Raspberry PI Aktie

Good afternoon, ladies and gentlemen, and welcome to the Raspberry Pi Holdings plc Final Results Investor Presentation. [Operator Instructions] Before we begin, we'd like to submit the following poll, and I'm sure the company will be most grateful for your participation.

I'd now like to hand over to the management team. Good afternoon.

Good afternoon. So welcome to our results presentation for 2025. Richard and I are going to walk you through some abbreviated highlights of the year, a summary of our financial performance. We'll give an update on our progress against our strategy and conclude with some words on the outlook for the remainder of 2026.

So, in summary, 2025 was an absolutely standout year for us. Shipments of boards and modules increased by 9% to 7.8 million units, and that delivered a 25% increase in EBITDA. We saw demand for our products build steadily throughout 2025. The second half was noticeably stronger than the first half. And within that, the fourth quarter stronger than the third. We saw particular strength in our two largest markets, the United States and China. As we speculated in our half year results, this was indeed the crossover year for our semiconductors business. We saw a 47% year-on-year increase in chip sales to 8.4 million units, meaning that we sold more chips than boards for the first time.

Raspberry Pi is at heart a product company, and we love to launch products more than pretty much anything else. 2025 was a somewhat slower year for us in 2024 in terms of product launches, but still one of our strongest historically. We launched many new hardware products, including new microcontroller variants that, for the first time, embed nonvolatile memory alongside the RP2350 die. But I think the star of the show for us was our first software product, Raspberry Pi Connect. This was launched in beta in the middle of 2024. And we added many new features, including over-the-air firmware updates in 2025, those features really being very finely targeted on to the needs of our OEM, our paid for OEM users of Raspberry Pi Connect, Raspberry Pi Connect for organizations. We ended the year with very nearly 400,000 devices registered, and we're over 500,000 units, 500,000 devices registered today.

And finally, we've been strengthening our marketing outreach. We've been increasing our presence at physical events at trade shows. We've seen strong momentum in our board-to-board program, which aims to introduce and promote Raspberry Pi to major industrial OEMs at the C-suite level. We've seen particular interest from OEMs in smart home and in aerospace and defense. And we've been refining the structure of our reseller channel. For the -- we've retired a number of underperforming partners such that for the first year in a very long time, we actually left the year with slightly fewer resellers than we started the year with. But we've made a number of targeted additions of reseller partners targeting key geographies and key sectors.

So I'll hand over to Richard for the financials.

So, I think, as Eben said, 2025 was a very good year. Perhaps to understand the story, it's good to look at the shape of the unit sales over the last 24 months, essentially on a quarterly basis. Quarter 1, 2024, we came into the year of '24 really after that period of shortage through '22, '23 of electrical products. And we really had the final back order, particularly on compute modules, which caused our direct sales, as you see in blue, to be particularly strong. And also at the same time, it was really the first quarter where we were fully selling Pi 5 through Premier Farnell, our licensee partner, which is the orange. So that, that first quarter was very strong for a number of reasons. But that came into the middle of the year and really like a lot of the electronics sector, I think everybody had come out of that period of semiconductor shortage, had bought an awful lot of product, and there was a general industry-wide indigestion, which really took through quarter 2, quarter 3 and a bit into quarter 4 to really settle through, and you saw that in our sales volumes through that period.

One of the bright spots in that third quarter was also we sold -- we launched our second microcontroller product, which we sell on Pico boards. So the actual number of units of boards was helped, but it's a much more low-cost product, much lower unit margin, which we'll touch on. And then gradually, as we came into the start of '25, that demand had settled. There was much more opportunity in terms of industrial sector. So we saw, which we probably regard as one of our sort of bellwether products, the Pi 3, a product that's been with us since 2016, sorry, pick up in volume again as industrial customers really came back in, I think, after the overstocks had flared. And that sort of gentle progress really took place through the first three quarters of 2025. Until then in the final quarter, we saw really that strengthening in demand quite significantly. Some of it was new products. Some of it was probably a little bit of stimulation from memory prices going up. But most strongly, it was people building in new products into their development of compute modules and purchases of Pi 5s and Pi 4s. So overall, a very good finish to the year, and that has carried on into quarter 1, which is seeing very similar volumes. I guess the day is yet young, but hopefully, by the end of the quarter tonight, will be a very similar numbers to the end of 2025, that last quarter.

So overall, it was a year of 7.6 million units. That was up some 9% on 2024. Gross profit per unit. So it's one of the key metrics that we follow. So how much profit have we made on each of those boards that we've sold was $8.70, which is up on $7.40 from a year ago. A number of features there, definitely better mix. I've mentioned already, our Raspberry Pi 5 boards, Raspberry Pi 4 boards. Retail from $45 right up to over $100, which is much more substantial. Some of the other boards such as Pico, which is sort of $3, $4 in price, were essentially flat year-on-year. So that mix was a major reason for the lift in the higher margin that we make.

And the other factor was the first 2 million of the Pi 5 boards we made, the chip that went into there was an early version of earlier version of the Broadcom 2712, and it was $5 more expensive. Once we cleared through that, the costs of our boards became -- of the Pi 5 boards became $5 cheaper, which is quite a significant uptick as well in terms of that margin -- gross profit per board. So overall, our gross profit was up 23% as a result of those factors as a result of more units being sold and at a better margin per unit. And our consequence of that, our costs rose similar level to the gross profit such that our EBITDA was also up about 25% year-on-year to $46.4 million.

That gave us, as that flowed through an adjusted EPS, earnings per share rising 35% to $0.1450 from $0.1070 a year ago. And overall, the result of that with adjustments to working capital and things like that, we'll talk about later in terms of inflow and paying off some of the payables that we've had previously left us with cash of $28 million from the year compared to $45 million a year ago. And in addition to that, we have a revolving credit facility of $80 million. That has remained undrawn. So essentially, we've ended up with cash at the end of the year of $28 million, a clean number.

So overall, I talked about the units being up. You can see the direct share rose much more significantly. Royalties were flat, such that the direct share increasing means that of the total rose to about 76%, which is similar to the numbers that we expect to see in '26. We probably believe that our share of the boards that we sell as opposed to out of total boards, the other being royalty is in the range of about 70% to 80%. I think 70% was probably in 2024, about the lowest level, and we expect to see that rise closer to 80% in the year ahead.

The units at 7.6 million in total, we expect '26 to be better on that with that momentum that we saw in Q4 last year and quarter 1 of this year continuing for the year. And so numbers strengthening there. And gross profit per unit, as mentioned, $8.70. We expect that to come down in '26 really as memory prices, the cost of the memory that goes into our boards has been rising significantly. We will talk more about that later. Whilst we put up prices, then obviously, there is an effect there that means the margin on the boards in terms of dollars per board will come down, and we anticipate that to be the case.

Accessories, which is in the -- you call that bright pink on the top, it was another strong performance. That rose to about $1.40 per board. Those are things like that we sell alongside there. Some of them are products that we very much design ourselves. Others are products that we have bought from other people. We have intervened quite significantly in the design of quite a few of these items, so power supplies where we have been quite fastidious in terms of the quality of those. But those overall have had a very good year in '25. So that's areas like cameras, AI hats, displays, power supplies, all stepping forward very nicely relative to the number of boards and also includes areas like SD cards and SSD memory as well, which we started to bring through into '25. We've seen significant take-up of those. It's very important for the quality of the product that the SD card is reliable. And again, it's part of this culture of engineering that the team at Raspberry Pi have really made sure that those products are -- do not diminish the overall offer of the Raspberry Pi board itself. So those have performed strongly in the period, and we'd see hopefully similar numbers into 2026 in terms of that sort of profit per unit, maybe even stepping up a little bit from there.

So, in totality, we see revenue was up significantly. Increases in higher prices over the -- sorry, was up significantly, really driven by units and also slightly increased mix. We see that coming forward into 2026, where we've had to increase prices in certain boards really quite significantly to cope with memory price increases. But you would expect to see the revenue for 2026 increase substantially at the same time as units picking up, but that board cost increase going up.

Gross profit rose 23% while the margin at this stage stayed pretty constant at 24%, that's that margin that I think you would expect to come down in '26. We're very focused on that profit per unit. And as the base cost of the boards is going to -- price is going to increase as we accommodate more expensive cost of sales from memory, the revenue will increase, whereas the profit per unit will stay at similar sort of levels. So we would expect that margin to come down. It's not a number that we chase. We're much more focused on that profit per unit as a measure rather than the percentage.

Offset against that increase in gross profit, we saw an increase in research and development costs and in administrative costs. Research and development costs were up year-on-year. We saw the underlying salaries of people that we employ this area is essentially that line is dominated by the cost of our colleagues. That was -- their costs were up about 10%, a slight increase in the number of heads, but also increases in payroll. However, that total amount, we then deduct from that and capitalize up salaries of people and the projects they've been working on. In 2024, that was about 50% -- 56% of those salaries. That came down to about 42% in 2023, such that the net cost in that line, therefore, increased more substantially than just inflation. This is really a feature of depending on the projects that we've been working on in '26 -- '25, we had more in the way of software products, which we don't capitalize. '24 was a year when we were finishing the launch of the RP2350, our microcontroller and therefore, much more significant level of capitalization. We would think of this as being more typically about 50% into the future.

Admin costs and administrative costs were up really as a function of the first full year of being listed, they were up some 19%. So we had additional heads, additional costs, be that stock exchange fees, but also just generally increases in -- we had a good year in terms of bonuses as well, driving that administrative cost up. I think going into the future, we'd expect to see that grow at a lower rate in future periods. I think there was a bit of a sort of catch-up effect in these numbers.

Brings us down to an adjusted EBITDA number, as referenced of about 25%. Depreciation and amortization was essentially flat year-on-year. Similar products were being amortized in both periods. And then we did also have a slight reduction in the rate of amortization on some semiconductor products where we saw longer lives on those. We did a reassessment based on the demand for the products and the demand was of a strength where the idea that these economic lives of these products we exhausted within six years was just not practical. And so we've stretched that by another couple of years, and that's produced a slightly lower amortization charge.

Our tax rate -- taxation charge at the bottom. Our effective tax rate, which is the sum of current tax and deferred tax compared to our adjusted profit before tax was down to about 18%. That was really two factors bringing that down from, I think, what we would regard as the U.K. norm of 25%, the government corporation tax rate. We have obtained a patent in respect of one of the elements of Pi 5 and also of the semiconductors. This allows us to put the profits of those businesses into what's called the patent box and attract only 10% tax. And that taken with a release of a provision that we've set up in the past, brought that tax rate down. I think in the future, we'd expect that rate to be closer to 22% to 23% in the future. So those factors taken together leads us to an adjusted earnings per share of $0.1448 per share, which is up some 35% year-on-year.

Turning to the balance sheet. The key elements within here. our intangibles, which is the major part of our fixed assets, they were up to about $83 million from $73 million a year ago. That's essentially the capitalization, particularly of intangibles that we've incurred in the year normally at about sort of $20 million. This year it was about $18 million, less depreciation in the period. Our inventory, $145 million. The total amount improved. It came down slightly from about $156 million a year ago as finished goods reduced.

During '24, we had a period where they increased as we were continuing to manufacture demand slowed. We started to get on top of that. We pushed back quite hard in terms of production. So that by the end of the year, it was starting to come down. That continued through into 2025. And it's probably fair to say now at the sort of levels of, I think, about sort of $30 million to $40 million, that is probably about as low as we can realistically take it at the moment. So we may expect to see some increase there, whereas components, we did increase in 2025. Memory holdings went up. And also, we had some increase in the processor chips that we use, all part of that making sure that our business is resilient as possible to these fluctuations in the key components. So overall, inventory came down slightly, but with those two factors, the finished goods coming down and components going up.

On the other side, receivables increased. It was up to $59 million from $36 million a year ago. That's really a function of how strong that last quarter was when we saw significant sales volumes going out and also in truth, the increase in the prices of finished boards and such like, where you were seeing, therefore, higher value per individual units as well. So that increased.

Payables did come down. At the end of '24, we had about $52 million of extended payables. It's really a function of just how soft in truth the market for components was during '24, really quite a turnaround since. But at that period, we had the ability -- the offer of paying certain suppliers over longer terms. That has unwound during '25, we were essentially paying those off as well. So that came down by about $52 million on that line, such that our payables finished at $60 million compared to $96 million previously. If you sort of look at those together, that increase in overall working capital to $144 million from just shy of $100 million previously, that outflow really driven by those extended payables being resolved. We're now free of those to all intents purposes. And so the flow over the period caused a reduction in cash through the period at $28 million. And at that point, we, I think, have started to see an improvement in the cash position. Our RCF remains undrawn at $80 million. We increased that during the year in terms of the facility, but we haven't had cause to use it, but it's there in reserve should an opportunity arise.

Cash flow overall, slightly unsurprisingly is the result of the factors that we were describing on the previous page in terms of the movement in the balance sheet. Profit of $45.8 million or EBITDA, payables reducing, and therefore, we've consumed that level of payables. CapEx, as you can see on here, we spent about $18 million of cash CapEx, pretty close to the $20 million of guidance that we give. We've flagged in the past, we'd expect that to rise to circa 25% of gross profit over time. It is lumpy. There may be some periods where we spend more. We may buy IP, which you buy as a single piece in a given year. And we also, on some of the chip design programs, that final tape-out as we refer to it, where you send the designs across to TSMC typically in our case for them to turn it into full-scale production. Again, there are sometimes some significant one-off costs. But I think that overall sort of trend of 25% taken across the year should make a lot of sense.

We had a tax credit. We had two lots of research and development expenditure credit received in '25 as opposed to more normally an annual amount of, say, $3 million to $4 million. That overall, I think, brought us down to the cash position referred to before. Cash conversion, which is something really for the longer-term outlook. We think of this at the operating cash flow level. So essentially EBITDA minus a movement in working capital. I think over a period, we'd expect that to run at somewhere between 60% to 80%. So a smaller consumption of cash and working capital in the future. We don't expect any more extended payables to complicate things. But it will, as the phrase that we've used a few times, be lumpy. If there is an opportunity to buy memory in significant quantity, we would take up that opportunity. So -- but generally, as a trend, I think it will be of that order.

I'm going to say a little bit about memory and then hand back to Eben. Memory costs have increased significantly really since the summer of last year. We were in the situation where, as I've referred to, the level of inventory we brought into the year was substantial, and therefore, the impact on this current year has probably been slight, but it is a feature that is very much in our thinking for this year ahead. The overall trend for memory is one that you can see in the graph on the left-hand side there. And bear in mind, this is a logarithmic scale here. So the declines are radically more substantial. Since sort of the people started measuring this in 1975, I can remember buying a computer with very exciting 32,000 of RAM, which is probably.

I think mine had been 16,000. I think I may have had the same machine, but mine that started at 16,000 to 32,000. Extreme left. We are on the extreme left of this curve.

And then slowly sloping down. It's the most -- one of the best expressions, I think, of Moore's Law, you will see. It has trended down. But in the last six months, we have seen on the back of really the trend for investment in data centers who want a lot of particular category of memory called high-bandwidth memory, have been purchasing significant volumes.

The key memory manufacturers have diverted a lot of their production away from computing, the general purpose computing side of memory use and from mobile phones to make this high-bandwidth memory for the hyperscalers and friends. And that has produced a significant increase in the prices of that memory, as you can see, about 7x. But it's also producing some conditions now where we're having to work hard to make sure that we can properly source the memory we need. And that's even in a situation where we're one of the largest purchasers in Europe of DRAM. So we've seen significant increase in memory. We have started to take steps to address that -- sorry, we have started -- we have been taking steps since the summer of last year. We're continuing to.

We've increased the range of suppliers we're working with. Historically, we buy from really two of the major manufacturers, Micron and Samsung. We've now been talking much more widely to other people. The Raspberry Pi name is in order to be helpful with that. We find lot of places we go to. We haven't met people before, but they have used our products university or in the company.

Certain pieces of technical innovation that allow us to use memory more effectively to create 4 gigabytes from 2 gigabyte pieces of memory. And we're also just having to pay more, and then we are taking more substantial purchases where possible to manage all of that. So we've increased prices of our products already. As you can see here, there is a product, the 4-gigabyte compute module 4, which actually at one stage in early '26, I think, we reduced the price to tidy up. And since then, we've seen some quite substantial increases I'm afraid there's every indication of those price increases. There will be more coming over this year, but it is a step that we've done to basically maintain that gross profit per unit at a sensible level to be positive. About 25% of our cost of boards is made up of this kind of DRAM on the boards that use it. About 1/3 of our boards, the older ones, particularly do not use this. They use what's called LPDDR2, which you may guess from the name, 2 is an older generation, of which we have ample supplies.

So, in terms of outlook, I think we expect high elevated DRAM prices to continue into 2026 and probably into '27 as well. And really, the solution to that is coming from more fab capacity, more places to actually make this, which will take time. These are phenomenally sophisticated machines. Make this and require many billions to do so. And so I think the resolution by additional manufacturing is more likely to be '28 and '27. I think price increases will, however, help in that unfortunate Adam Smith's hand of the market. It will probably push some demand away. It will -- it has -- I think we're seeing a little bit of it already, some more to come where customers are revisiting the effectiveness of their software and looking for -- to use smaller amounts of memory for the purpose that they have. So again, there will be some reduction in demand there. But we're also widening our conversations to a much wider group of suppliers.

So overall, we're taking and have taken significant actions. I think this will continue to be a major theme for the year. We have -- at the moment, we came into the year with significant levels of inventory into '26. We're in a good place, I think compared to many a very good place in respect of 1 gigabyte and 2 gigabyte memory at the lower end. That should see us through right up to the end of 2026, almost to the end, I think, is what we said elsewhere.

On some of the higher 4 and 8, I think the doubling up the clamshell, we've referred to it. It's one of the solutions at the falls, but it is something we're having to work very hard to achieve. We think this -- we're comfortable with the numbers we have for '26, but this is an area of significant activity for us for the whole of this year. And certainly, there continues to be uncertainty that will take us through into '27.

I'll pass back to Eben.

Wonderful. Thank you. So a quick update on progress against our strategy. This was another great year for product launches at Raspberry Pi. I'm just going to -- out of the 13 or 14 products we launched, I'm just going to pick out three that I think help illustrate where we feel we're going as a business. Firstly, we have Radio Module 2 launched midyear. This product delivers turnkey wireless interfacing to our microcontroller customers, and it bridges the gap between these first-class deep embedded compute platforms that we build, RP2040 and RP2350 and the network. It's a fantastic complement to those microcontroller products, and it's become a successful product in its own right. Very much a demand-led product this, a product that we could only build because over the last five years, we've become enormously better at actually listening to our customers and listening to what our customers are telling us they need in order to make their Raspberry Pi-based designs a success.

Next, we have my personal favorite. This is Raspberry Pi 500+, our latest all-in-one PC. As Richard said, I think we both grew up with the same machine, both grew up with BBC Micros in our homes. So this is a love letter. This product is a love letter to our enthusiast base who still remain at the heart of the Raspberry Pi mission. And it's...

10 years. I was struggling to do things that you were doing at half my age.

Mine was a very old second beaten up secondhand of BBC Micro at the end of the 1980s. But this product is with its mechanical keyboard, with its backlit mechanical keyboard and 16 gigabytes of RAM, this is really a love letter of those machines that we grew up with. It's a wonderful demonstration for us. How does it fit in with our strategy? It's a wonderful demonstration that with the Raspberry Pi 5 platform, roughly 150x the performance of the Raspberry Pi 1 that we launched 14 years ago back in 2012, that we've now reached a performance level where this platform really can pull its weight in a modern high-performance client PC.

And finally, we have AI HAT+2, an honorable mention, didn't quite make it into the year-end, launched in the middle of January. This is the second output of our successful -- our very successful long-term collaboration with accelerator vendor Hailo. It builds on the success of our first-generation AI accelerator products that we launched in 2024 and for the first time, allows accelerated execution of modern AI workloads like vision transformer models, large language models and vision language models. I'll talk in a moment about why we believe that this product and the concept of running AI workloads on Raspberry Pi is very important to the future of our platform.

Of course, we can only build fantastic products because we first built a fantastic team. In 2025, we continued to hire into the engineering team. We added a number of new senior engineers into the silicon organization, and we continue to take fantastic graduates from our very successful and long-running summer internship program. And we continue to focus on retention. We had one retiree. So 2025 was a flagship year for us in another sense. It was the first year that somebody retired from the engineering workforce at the age of 71. And we have that one retiree, but otherwise, 100, we continued with our record of 100% retention rate in the engineering team.

We've made some targeted investments alongside growing the engineering team. We've made some targeted investments in business development, in application engineering, in communications and in the finance and legal team.

And last but very much not least, I'm incredibly excited that Tim Mamtora joined us this month in the newly created position of Chief Operating Officer. Tim was formerly the Chief Technology Officer at Imagination Technologies and is a long-time friend and former colleague of mine at Broadcom. I've known Tim for very nearly 20 years. We'll be centralizing and formalizing a lot of our operations activity into Tim's new organization, including engineering operations, manufacturing operations, supply chain, IT and cybersecurity. And really, when we make these changes to the organization, we see that as building the architecture that will support the next phase of our growth story.

Now as we build the capacity to interact with our customers, we also build our capacity to learn from them. I'm not going to belabor this slide, but I thought it was worth giving you a few examples of the sectors that our OEM customers operate in from digital signage to IoT gateways, retail and point of sale, smart home and industrial automation and many more.

Now it's striking when we talk to these customers that across all these sectors, our customers share both common strategic concerns. They're deeply concerned with time to market and with optimizing the transition from prototype to production. They care about gaining access to advanced technology without having to should a high fixed costs inside their organizations. And they're navigating, of course, an increasingly complex regulatory environment for IoT devices.

And they also share common product aspirations. Many of these customers are working to bring existing products to smaller form factors and lower price points. They're adding intelligence and connectivity to legacy products. And they're doing both those things in a world where security for IoT devices has gone from being a nice to have to being a legal necessity if you want to be able to ship your product. And all of these customers, as they confront these challenges and try to pursue these aspirations, they're all confronting what we call the R&D air pocket. This is the challenge of recruiting and retaining the talent required to design an intelligent product. bring that product rapidly to production and keep it in production in a rapidly changing environment.

We're positioning Raspberry Pi to these customers as a solution to these challenges. To our largest customers, we're doing that through our board-to-board initiative at the C-suite level. And to smaller customers, we're doing this by more traditional marketing activity and physical presence at trade shows and other events. I was at embedded world in Nuremberg a couple of weeks ago, largely to speak to DRAM vendors. But while I was there, I was struck by the hunger for solutions to these challenges among OEMs and by the growing awareness and really gratified by the growing awareness of and enthusiasm for Raspberry Pi as a supplier in this space.

I'll give you a couple of examples of some of our favorite OEM customers. Firstly, Sixfab, they were a guest on our standard embedded world. They use our platform to build ruggedized industrial computers, and they build those around our compute module products. Our value proposition to companies like Sixfab is really very simple. We provide reliable, available, embeddable general purpose computing at an attractive price point, and we support those computers with world-class software and other collateral.

Then we have ProGlove. They're a long-standing German industrial OEM customer who we serve through our direct-to-OEM channel model. They build wireless gateways for warehouse operations around Zero 2 W, our $15, our most cost-effective SBC product, most cost-effective modern SBC product. And again, a very, very simple proposition to customers like ProGlove. We're providing them with the perfect mix of compute interfacing and connectivity that in ProGlove's case, allow them to rapidly add new functionality to an existing platform, and we're backing that with long-term software support and long-term availability guarantees.

Now a word on the edge AI opportunity for Raspberry Pi. First of all, I think we have to remember that we come to this opportunity from a position of enormous strength. We're already the most popular platform for edge computing and many of our customers' workloads already have an AI flavor to them generally because there is a lag in terms of how new technology makes its way into the OEM space. When we say an AI flavor, what we tend to mean is visual AI, vision AI, generally the recognition and classification of objects in still and video images coming into the Raspberry Pi. Some of those workloads will run on the CPU, some of them may be GPU accelerated. Some of them may be accelerated by a dedicated accelerator platform for one of our partners, notably Hailo and Sony. And we have fantastic relationships with these accelerator vendors and with the model vendors. And we leverage these relationships over time to expand the platform performance envelope to expand the range of AI workloads, which will run performantly on Raspberry Pi products.

There's a cluster of megatrends, I think, that define and underpin the general edge AI opportunity. Frontier models in the cloud are hugely capable. But when we build IoT applications on those cloud models, we build in latency, we build in ongoing cost, and we build in a dependence on the reliability and availability of the network, which translates into inherent brittleness in the resulting product. In contrast, if we run -- if we are able to run AI workloads at the edge of the network, those -- that has the potential to deliver IoT applications, which are more robust, more cost effective and which deliver lower latency and address pervasive concerns around privacy and data sovereignty.

Now we know that over time, AI capabilities trickle down the performance ladder. Models become smaller, they become sparser, they are more heavily quantized. -- all without sacrificing performance. We've seen this already in Vision AI. The workloads that people run our largest OEM customers today are AI customers. They are running Vision AI applications on Raspberry Pi. And the workloads that they're running today on Raspberry Pi would have required a large NVIDIA GPU a decade ago. So we've seen this happen already. We've seen the confluence of the increased performance of edge devices like Raspberry Pi and the reduction in compute intensity of models at constant quality output. We've seen those converge to enable Vision AI at scale on Raspberry Pi platforms. And we're seeing this same trend replay itself today in large language models and other generative workloads. Taken together, this confluence of expanding performance and shrinking performance requirements will gradually bring more workloads into play for Edge AI.

So what's the opportunity here for Raspberry Pi? Well, quite simply, the opportunity is to leverage our existing position in edge computing to become the platform of choice for AI at the edge. We have the opportunity to serve the existing demand for edge AI compute across markets from industrial automation to defense. We have the opportunity to become the default embedded host for Agentic AI. And we've already -- we're already seeing this in the popularity of Agentic platforms like OpenClaw running on Raspberry Pi. And together, these give us the opportunity to define the future direction of AI as inference migrates to the edge over the next decade.

Now 2025 was a milestone year for us in semiconductors. As I said earlier, we sold 8.4 million devices. That's up 47% year-on-year, meaning that for the first time, we sold more chips than we sold boards. And we really are just getting started. We believe that semiconductors let us make better boards and modules. And we believe that over time, they will become a valuable second franchise in their own right. Our road map for the next five years is informed by both these beliefs. I think you're going to see regular microcontroller releases aimed at expanding our TAM, whether that's by hitting new price performance points, whether it's by integrating new interfacing capabilities and new connectivity features into the platform, whether it's by qualifying our existing products at extended operating points, all of those grow the total addressable market for our microcontroller products. And then building on the success of RP1, we expect to be able to deliver differentiated features and improved margins on our boards and on our modules.

So, Raspberry Pi Connect, we're less than two years into the Raspberry Pi Connect story, and it continues -- the Raspberry Pi Connect platform continues to go from strength to strength. We have 370,000 registered devices on the Connect platform at the end of 2025 and over 0.5 million registered devices today. In 2025, we launched many new features aimed at OEM customer pain points, including over-the-air firmware updates and deeper integration with our imagery utility to simplify the onboarding of devices and of users. Connect is the latest iteration, I think, of our mission to simplify the OEM product journey. And over time, it's becoming both a valuable complement to an enabler for our hardware business and an increasingly credible line of business in its own right.

Finally, a word on outlook. Well, we said that in 2025, the second half was stronger than the first half and that Q4 was stronger than Q3. We brought that momentum and substantial customer backlogs with us into 2026. We're very, very confident in the short-term demand environment and in the work that we're doing to grow demand in the medium to long term and particularly in the potential of the board-to-board initiative. On supply chain, clearly, the DRAM environment remains a concern, but we're comfortable that we have the inventory and the relationships to navigate it. And we really do see this more as an opportunity to take market share and build customer loyalty than as a threat to our business. And on finances, we're expecting profitability for the year to be in line with market estimates, but we are flagging that obviously, with price increases on the product with an increase in product ASP, revenue is likely to be substantially higher than market estimates.

So I'm going to leave you -- not going to belabor this slide either. I'm going to leave you with some quotes from some of our favorite OEM customers attesting to their enthusiasm for the Raspberry Pi platform, and I'll be very happy to take questions.

That's great. Eben, thank you very much indeed and Richard for updating investors. [Operator Instructions] I'd just like to remind you a recording of this presentation along with a copy of the slides and the published Q&A will be available via the Investor Meet Company dashboard.

Eben, Richard, you received a number of questions from investors today. Thank you so much to everybody for your engagement this afternoon. Eben, if I may just hand back to you, if I could ask you just to read out the questions, of course, where it's appropriate to do so, and I'll pick up from you at the end.

Okay. Well, I'll start with one which is very much at the top of my mind at the moment. This is, do we have a -- we don't have a name attributed to this. But Raspberry's mission is rooted in education and accessibility. I'm concerned about reports of boards and add-on systems ending up in conflict and weapon systems via the gray marketplace. How is the company ensuring that products remain aligned with its educational mission and what safeguards are being strengthened to protect the brand and its ESG standing.

So we are aware of the use of our products in particularly drone platforms, particularly Russian drone platforms in the context of the invasion of Ukraine. We -- this news has prompted us to take, I think, another look, a very detailed look at, as you say, the gray market, I would classify these under the sales under the heading diversion. We put in place -- we've put a lot of work in over the last six months, in particular, to strengthen the -- particularly for unit selling into China to strengthen the information that we require from that our resellers provide us in terms of the end use of our products. We've been very gratified by the assistance of our licensee partner, Premier Farnell and our Chinese approved reseller partners have given us in allowing us to get that extra level of detail.

We do believe it's possible. I don't think this is not a council of despair. We do believe that it is possible to meaningfully limit hopefully, to very, very small numbers, but potentially to zero to limit the flow of our products into these sort of applications. Nobody is happy when they wake up in the morning and see to these stories, and we are doing our very best to get in the way of this diversion of our products into these...

Clear to our resellers in China and a few other intermediate locations unacceptability and reinforced how clearly that's set out in our terms of trade with them.

Fairly obvious question, but what is the effect of high memory prices on your sales? I think it's fair to say we've yet to see significant evidence of elasticity of overall elasticity of demand of people not buying Raspberry Pis when they would otherwise support Raspberry Pis. I think we believe that over time, we will see some density elasticity. We will see people trading down. I think we've come through an era of DRAM being incredibly expensive, incredibly cheap coming into an era of being incredibly expensive. We come through an era of it being incredibly cheap. And in that environment, both individual users and smaller volume OEMs are not heavily incentivized to trim their usage of memory.

I think as we come into this more expensive era, it will be a time-limited era, but we do believe it will be with us for a while. As we come into that era, we do expect to see -- I think we've seen maybe a little evidence of enthusiast customers trading down already. I think we will see OEM customers. Obviously, there's a time lag for OEM customers because there's engineering involved in tuning a software stack to lower memory debt.

But I think taken in conjunction with, I think, an increase in boards, which we look forward to for '26, we do have higher prices. So I think the revenue line in our accounts will increase substantially on that basis.

But the thing I think that we just want to stress to you, we do focus on profit per unit. Therefore, the gross profit will not rise in the same way. And there probably as a margin percentage, there will be some impact, some reduction in that margin percentage as a consequence of the maths in that.

Are there plans to bring Raspberry Pi Connect to the microcontroller family? Yes. There are plans to bring Raspberry Pi Connect to the microcontroller family. I think we are refining at the moment if you think that Raspberry Pi Connect has its roots as a remote access solution. I think we are refining what the proposition there might look like for remote access. I think the proposition for over-the-air updates, which we added to Raspberry Pi Connect at the end of last year, microcontrollers is much more straightforward. You may even imagine a world in which we ship OTA before we ship remote access in that space. Let's see.

Do direct unit sales include sales to approved resellers or these B2B direct? Now when we say direct unit sales, that's the sum of the -- if you think we have three channels, we have licensee, we have direct to reseller and we have direct to OEM. So direct to reseller is mediated sales to OEMs, to industrial users and to enthusiasts. Direct for us is the sum of the latter two direct for us is everything that is licensee. That's the number that we see in this kind of 70% to 80% corridor in any given year.

Can you comment on the success of the China-only compute Module 0? It seems like a nice sweet spot of price performance that the world would buy in large quantities. So 2025 was also the year that we launched our first geographic-specific product. This is the CM0 low-cost compute module that we launched to the Chinese market. I think that we are -- we have seen -- that has been a success for us. I think we have seen inquiries. We have seen interest outside of China. I think the challenge for us with that platform is we are not currently -- it is not currently clear when we launch a new product.

Increasingly, we are offering 10- to 15-year windows of basically guaranteed availability. It isn't clear to us because of the chipset that we use in the compute Module 0 platform that we can meet that criterion for compute Module 0. That's important for a global market product. It's a little bit less important for a China market product. And so I think we just have to understand how we either get comfortable with the long-term availability of compute module 0 or we get comfortable that we can message the reduced availability window for that product. So it's likely to be a 5- to 10-year product rather than a 10- to 15-year product. We just need to make sure that we message this. It's an enormous gest of faith when an OEM customer designs our product in, and we want to be very clear that people aren't designing the product in on the basis of an unrealistic expectation about product availability.

Let's see. Admin expenses include sales personnel, correct?

Yes, they do. Yes.

Let's have a look. How high is the risk of sourcing DRAM, et cetera, compared to bigger players like Lenovo? What is your strategy on this? There's always risk associated with sourcing components of any sort during a shortage for that component. I think that we benefit in general, I think in any -- my feeling and some of the feedback, I think we've had from vendors is that being a large player in a particular market is advantageous to us in this situation. And so just as I think Lenovo will probably benefit from being one of the largest, possibly still the largest laptop vendor. We benefit from being one of the largest vendors of modular and single-board computers.

So there's always risk. But as I said, I think the formulation we've used is we're comfortable that we have the inventory levels because, of course, we did come into this year with substantial inventories of DRAM, particularly at lower densities. And then we have the relationships. And part of that is the brand equity associated with the Raspberry brand that we can leverage to secure ongoing supply. So I think we are optimistic. There is always danger there, but we are always -- we are optimistic about that.

Let's have a look. How do you manage different currencies? Do you use options futures?

Essentially, all of our numbers down to gross profit are all in dollars. We take some -- we sell some products to customers in euros, but that price is actually pegged to the dollar and sort of reevaluated certainly quarterly and more often if the price moves. We have overheads in sterling. We do hedge those out into the reported dollar number using essentially forward contracts. We keep it as simple as possible just to lock in that, take one area of risk out. But certainly, we haven't looked at any more complex derivatives. And I think we've got other things to concentrate on.

There are no DRAM futures.

Sadly. That would be very good actually.

Did we see a significant increase in demand related to Agentic AI and OpenClaw? No, I don't think we saw a prompt increase in demand. I think the reason we are excited about Agentic AI is because it's a sign of things to come. Often, what we find is that our enthusiast customer base are ahead of the game. The things our enthusiasts are doing now, our OEM customers will do our industrial customers, our OEM customers we'll be doing in three, four, five years' time. The popularity there is undoubtedly substantial use of Raspberry Pi in this space is an indicator that Agentic AI, particularly in this kind of future agentic AI in which more of the computers migrated into the agent has migrated into the local client device is a powerful -- I guess it's a powerful abstraction. It's a powerful abstraction to put around the underlying AI technology. So it's more a forward-looking thing than a source of very substantial demand in here.

Are we concerned about Chinese companies reverse engineering of products and competing? Obviously, we're always concerned about competition. We're very alive to the threat of competition. I think what's striking about Raspberry Pi is that we've been able to build a sustainable business around not just building the hardware, but putting high-quality software on it and building high-quality collateral around the platform. I think that's probably what distinguishes us from nominally similar Chinese competitors. We continue to be very alive to the risks of competition, continue to try and bolster the platform.

You can see Raspberry Pi Connect as being an important part of this, right? We are very focused, particularly for our OEM customers on identifying their pain points and doing the engineering that's required to address those. And the more of that you do, the more compelling, there's a nonlinear benefit to being able to solve all of OEM or all of an OEM's problems rather than just a subset of them.

But then on things like Raspberry Pi 5 and compute module, it has our own silicon on it. So you would have to reverse engineer the silicon to actually produce competing product.

Let's have a look -- we have a Max B on the line. Many questions, as always, from Max. Let's have a look. Many thanks to investor for the opportunity of this meeting. A lot of us on here, I guess, would like to own the shares. However, the small free float is contributing to a highly volatile share price. What can you say publicly about the likelihood of the free float increasing?

I don't think we can necessarily say anything about that. I mean, obviously, it's -- we understand why the free float for the company is limited. We have a nearly 50% shareholder in the Raspberry Pi Foundation. We have a number of other large shareholders in the business. I don't think that we have a perspective on that. That's a shareholder issue rather than a company issue, and I'm not sure we really have a valuable perspective on that.

From George, what types of partnerships could accelerate your next phase of growth?

I think that the partnerships, they look like customer relationships sometimes. But I think some of the companies that we're pursuing through the Board-to-Board initiative, the sort of super massive companies that we are pursuing through the Board-to-Board initiative, they do feel more like partnerships than customer relationships, both because they're large enough and deep enough to...

They run for quite a while to establish them.

Yes, that's right. So when you have a large multidivisional prospective customer, often you are interacting with that customer at the C-suite level. So you're not interacting with anybody who does, who makes anything, you're interacting with that kind of central organization. And you often find yourself, I think, in a partnership relationship with the central element of a multidivisional organization. I think the Board-to-Board engagements are working particularly well are working well because we found in that central organization, somebody who is prepared to be our partner, who is prepared to take Raspberry, take the Raspberry Pi story, take the Raspberry Pi value proposition out into the organization. I think those have been very valuable.

I think we've also -- I mean, with the help of particularly our brokers, we've actually been able to get quite a senior level access to these organizations. So perhaps something that maybe a partner in a purely sales view would provide is actually something we've been very lucky in terms of getting access at the right level in organizations in the last years.

Yes. I guess, obviously, the other -- the simplest example is what sorts of partnerships? Well, the sorts of partnerships we've been building for a long time. We run a partner event every October now. We have roughly 400 people. You look at the sorts of partners who are there, originally, it was a reseller event. I think all of these organizations are go to the Arm partner meeting every year in Cambridge in August. And that's an event that has, over time, evolved from being simply a meeting for Arm's IP licensees to being a meeting of the ecosystem that is convened around the ARM platform. I think we see that change. So we see a world where five5 years ago, the event was about resellers. Now it's about resellers. It's about ISVs, it's about IHVs. It's about the entire ecosystem that's grown up around the platform, some of our vendors as well, some of our component vendors.

So we continue to try and nurture those partnerships, grow them, deliver value to our partners, whether they're upstream from us or downstream from us in the supply chain or whether they're off to one site, whether they're producing complementary products that benefit from where they benefit from the existence of Raspberry Pi and we benefit from the existence of their solutions regardless of how -- of what the business relationship is with the partners. I think that we have got better over time at understanding what their needs are and articulating an advantage to them in being involved with Raspberry Pi.

Let's see who else -- let's see. What was happening throughout the year that led to more MCU units being sold in the first half versus the second half of 2025? That's a good question. Nobody has asked this yet. What's the outlook for the MCU business in 2026 and also the medium term? Will demand likely follow the product launches? Or should we expect to see continuous growth for existing products?

A couple of thoughts there. I'm surprised that nobody has asked us this question before. Yes, we definitely did sell more MCUs, fractionally more MCUs in the first half of the year than the second half of the year. We had a couple of big hits in the first half. So generally, like a lot of our business, our MCU business is made up of a very large number of relatively small orders. In the first half of the year, we had a roughly 0.5 million unit order and an 800,000 unit order, the first from a Taiwanese customer, the second from a Scandinavian customer. Those were sufficient to tip the balance, even though I think we were seeing underlying month-on-month growth, those two were large enough. Our business is still small enough that our aggregate 1.3 million units is enough to tip the balance back into the first half of the year. I think we're seeing good momentum into this year. We've had some good -- yes. So I think we are -- I think we -- I would expect to see continuous growth. The interesting aspect of those sales, that 8.4 million units last year, dominated by RP2040, a real reflection. That's a product we launched in 2021, nominally superseded by RP2350 in some ways. in 2024, but the design cycles are sufficiently long and the cost structure advantage that 2040 has over 2350 is very real that we continue to see.

I suspect I've said to somebody the other day in the context of amortization periods, I said these are conversations I have now. I used to be an engineer that I would not be surprised if in 2036, we sell more RP2040s than we sold in 2035. I think silicon products are -- silicon products, particularly deep embedded silicon products like these microcontrollers are weird, right?

You've been seeing that for 20 years.

Yes, that's it. That's it. I mean lots of -- these products have very -- well, you talk about them having very long tails in practice, they may continue to ramp pretty much indefinitely. So that's quite exciting.

Somebody, Mark saying, I can see the advantage of getting your desktop 500-plus kits into schools in Africa, energy-efficient, low cost, obviously, solid state devices and therefore, much more durable and rugged environments. We have a couple of. One of the things we've done over the last five years is to build out in Sub-Saharan Africa, a reseller community. Some of the -- some of my favorite interactions at our partner event are with our African resellers. We've built out a reseller community in sub-Saharan Africa, very much like the reseller community we have in Europe. And certainly, the 400 series product, which remains extremely cost effective.

So the 500 has got a memory.

Yes, 500 has memory challenges, 400, we're lucky enough to have come into the situation with reasonable inventory of that product with old price DRAM in it. So we see 400, 500, 500 plus as being particularly interesting in that market.

Let's see what I only have Max questions left. Let me have a look. Max has asked a bunch of technical questions about the RCF. What other types of accessories do we envisage adding to the lineup? I mean that's very interesting. We always sort of feel that we must have got to the end of accessories. And you look at the accessory sales, they're quite dominated by cameras. Displays, pass storage. Those are the kind of -- that's the heart of the -- and that's kind of the heart of the business because every device needs a -- because every device needs at least the power supply and flash storage and many of our -- certainly our OEM applications, to the extent the vision AI applications, they need a way of capturing images and sometimes a way of displaying images in the field. So -- but then we have a large number of other accessory devices.

Enclosures, I suppose, is another one. We always think we're done. We have various hats. We always think we're done, but I'm sure we'll think of something. You think about the things we've not done, we've not done anything with cellular radio. We've not done anything with batteries. There's always been a little tiny squeamishness in the business about batteries. We hold ourselves to very high engineering standards. Battery engineering is extremely challenging engineering. We've not done that yet. I suspect those two, we might take a look at. But then at the same time, I suspect you're going to see more cameras, you're going to see more displays. You're going to see us doubling down on the things that really work for us in the business.

But they're all things that we actually participate in the engineering of. They're not just sticking a name on.

Let's see, what is the time -- another Max question. What is the time line for Pi 6? Has it been impacted by the memory situation? I mean we've all seen a couple of reports about PlayStation time lines?

No. I mean, I think that the time line for Pi 6, if you look at our history, our products generally have four or five years in the sun. Every product has four or five years as the flagship product. That puts us about halfway through the nominal lifespan of Raspberry Pi 5. I suspect Raspberry Pi 5 has legs as a product. It's such an enormous step up. Pi 4 was a very large step up over Pi 3. To some extent, I feel that Pi 4 is [indiscernible] compute are only coming into their own [indiscernible] So to the extent that we had two successive very large steps up in performance in 2019 and 2023, I suspect that Pi 5 may have some more -- may have more life to it than a typical flagship product. So whatever happens, I will be -- even at the most optimistic schedule for that product, I will be surprised, saddened, disappointed if we're not out of the woods on DRAM by the time that product reaches its natural launch moment.

So I think we're covering the -- perhaps a few closing remarks.

Yes. Let me just jump in there, and thank you to everybody for your engagement as more questions keep coming in there. Eben, every question you answer, another one comes through. So thank you to everybody for engagement. Eben and Richard, I know investor feedback is particularly important to you both. I'll shortly redirect those on the call to provide you with their feedback. But before doing so, I wondered if I may, Eben, just ask you for a couple of closing comments.

Sure. Well, thank you all for your time today. 2025, as I said, was a remarkable year for Raspberry Pi. 2026, I think, is shaping up to be a remarkable year for us as well. So again, thank you for your time. Looking forward to seeing what the next year has in store.

That's great. Eben and Richard, thank you once again. Could I please ask investors not to conclude the session as we'll now redirect investors to provide you with their feedback.

On behalf of the management team of Raspberry Pi, I'd like to thank you for attending today's presentation, and wish you all a good afternoon.

Welcome to our results presentation for 2025. Fourth time we've done this, first time we've done it in this particular room. Rich and I are going to walk you through some abbreviated highlights of the year, give a summary of our financial performance, an update on our progress against strategy in the year, and then we'll conclude with some words on the outlook for the rest of 2026. In summary, 2025 was absolutely standout year for us. Shipments of boards and modules increased 9% to 7.6 million units, and that delivered a 25% increase in EBITDA. We saw demand build steadily through the year. The second half of the year was better than the first half. The fourth quarter was better than the third quarter. And we've really taken that demand with us, that demand momentum with us into the first quarter of 2026.

We saw particular strength in our 2 largest markets, the United States and China. And as we speculated when we did our half year results, this was, in the end, the crossover year for semiconductors, the first year in which we sold more semiconductor products than we sold boards and modules. We saw a 47% year-on-year increase in chip sales to 8.4 million units. Look, we're at heart of Raspberry Pi. We're at a heart of product company, and we love launching products more than pretty much anything else we do. This was a somewhat slower year for us than 2024, but still one of our strongest years historically.

A bunch of new hardware products, including some new microcontroller variants, we launched in August, we launched variance of the RP2350, the new -- the second-generation microcontroller. We launched variant of that product, which, for the first time, integrate nonvolatile memory into the package. But really, the star of the show for us was the performance of our first software product, Raspberry Pi Connect. We launched Connect in beta in the middle of 2024, added a lot of new features, including over-the-air firmware updates, which just squeaked in at the end of 2025. All of those features really intended to increase the utility of the Raspberry Pi Connect platform for our Raspberry Pi Connect for organizations, OEM customers.

And we ended the year with very nearly 400,000 devices registered with that platform. On the commercial side, we've been strengthening. We continue to strengthen our marketing outreach, increasing our presence at physical events. We've seen strong momentum in the board-to-board program. This is the initiative that -- this is our initiative to promote Raspberry Pi directly to major industrial OEMs at the C-suite level. And we've seen particular interest from OEMs in the smart home and aerospace and defense sectors. And we've been refining the structure of our reseller channel, too. We've retired a number of underperforming partners to the extent that this is the first year for the first time we left the year with slightly fewer, very slightly fewer reseller partners that we entered the year with. But we've been making a number of focused additions targeting key geographies and key sectors. So with that, I'm going to hand over to Richard for the financial review.

So overall, as Eben said, 2025 was a very good year. I think to understand that story, you need to just look at the shape of unit sales over the last 24 months, which, as you can see here, we've broken down by quarter. Why particularly quarter 1 2024 was really the period when we saw the last of the orders that came through from the shortages that had happened in '22 and '23 and really the orders that people have placed finally started to unwind in '24. And also it was the period really when the unit volume from the launch of Pi 5 a bit earlier came through. So that was a particularly strong quarter for Farnell with the Pi 5 volume that they really took on, but also compute modules were incredibly strong in that period.

Then coming into the sort of middle period of 2024, really the quarters that straddled the half year was when I think as the whole of the electronics sector saw this significant indigestion of people who had really sort of filled their warehouses, filled their boots with products. And that really caused that sort of dip through the middle of the year. It's sometimes described as a [indiscernible]. I think Rich just might have been a better way to describe it through that period. That started to -- the indigestion, should we say, started to clear by quarter 4 of '24. And then coming into '25, we started to see that buildup of products like demand for Pi 3s, compute module, those classic products that we would see as very much the ones where there's industrial demand. That saw a steady improvement through 2025 to really having then a very strong close to the 2025 as the industrial customers really came into strength.

We saw compute module demand coming through. And that was really that period in quarter 4, about 2.2 million units. And we've seen similar momentum coming into the next quarter into quarter 1 of this year. Inevitably probably some interest from people looking to anticipate memory increases, but really demand that we regulated quite rigorously through the sales team in terms of stopping people from overbuying and really holding people to either prices these days that they if there's going to be a price increase, which something anticipated, we've said to them, you will get that future price, you don't get the price at the date you're placing the order. So we have taken quite significant steps to try and prevent people from building up inventory.

So overall, our unit volumes, 7.6 million units, up 9%. Our gross profit per unit was $8.70, which you can see is up significantly on the $7.40 of last year. Some themes within there. The first tranche of 2 million Pi 5s, they cost us about an additional $5 because of the initial cost for Broadcom chips that we agreed with them. So that obviously unwound in 2025, improving that margin. But also, we've seen a better mix. Pico's and those sort of low-margin products have been flat through the period, whereas Pi 5 and Pi 4 and the compute modules have picked up. And we've also seen a growth in direct, which for the same board sees a better margin -- sorry, profit per unit. So gross profit per unit overall was up 23% and the gross profit was also up sort of similar amounts.

EBITDA was up about 25% on the back of that. And EPS, talk a little bit more detail later, adjusted EPS was up about 35% on the back of that stronger profit. Cash finished the year at about $28 million, down on the previous year. But during the year, we had some significant extended payables that coming into '25, which we cleared through the year. So overall, finished at about $28 million and with an undrawn RCF. So as I said, units up overall to 7.6 million. We expect that sort of strength that we saw in the last quarters to continue into '26. So there's good momentum there coming in. So we would see some improvement on that number as the case that we're comfortable with for the following year.

Direct, particularly has increased while royalty sales have stayed flat. So we've gone from about 70% of our sales being through the direct channel to 76%. I think that number trending between 70% to 80% towards the 80% is the sort of expectation we have for '26. Gross profit per unit rose to $8.70, as we said, from $7. We expect that to come down in '26. I think the pressure on memory increases is such that we will see some pricing biting on that. So that profit per unit will come down in the sort of '26 in this year. And then accessories, which was the other sort of standout thing of the year. In terms of profit per board, it rose to about $1.40 from $1.20 in the previous year. And that's a mixture of cameras were very strong.

I think that vision, whether it's Vision AI, as Eben alluded to elsewhere, but that combination of cameras and Raspberry Pi boards there has been quite significant demand. We have sold quite a lot of SSD memory and SD cards, and that has certainly helped in '25. And then we've also had AI hats and also some really good volume on displays as well in the '26 -- sorry, '25 year. So in totality, revenue was up some 25%. It's not a number that we use as a key metric because it includes both royalty income and -- which is near 100% margin and sales of direct boards, which obviously is operating at lower sort of margins, say, 20%, 30%. But it was up 25%, in line with the other increases in units. We do expect that to increase really quite significantly in '26 as a result of the memory price increases that we're reflecting in the price of our boards that we're selling to people.

So we do expect to see significant increase from that and also from the increase in number of units. Gross profit was up 23% and as a result of the gross profit per unit that we've talked about previously. And the margins stayed pretty consistent at 24%. I think mathematically, that will come down in '26 as the revenue increases, gross profit per unit stays reasonably steady, comes down a little. So the effect of the math is that percentage comes down through that period. We are very much focused on that gross profit per unit rather than the margin percentage. Overall, overheads increased about 22% in total. Our research and development costs went up about 28%. Salaries, the people costs, which are the major part of that underlying increased about 10%, but we capitalize a proportion of those costs -- it was about 42% were capitalized in 2025, which is really quite a low percentage compared to, say, 56% in the previous year.

It really does fluctuate depending on the nature of the products that we're developing at the time and also where they are in the cycle. So in 2025, it was much more probably of an emphasis on software. '24 has seen the release of the RP2350, our latest microcontroller products. So there was quite a lot of capitalization in that period. Since then, people have been doing work to keep the product updated, but we don't capitalize that. So there is a function. I think in the future, we'll probably see that more at the 50% sort of level. Other areas of work, Pi Connect, something that given the wide range of possible outcomes, we've stood back from capitalizing that one at the moment, for example, and there's a number of other products of similar nature.

Administrative costs were up. They were up some 19%. This was our first year as a listed company. There were a number of costs that came in, whether it was grossly more expensive CFOs and their bonuses or fees to brokers and others who are looking at this luxurious building here and wondering whether they're in the right business. But that was up as a result of that. We wouldn't expect to see that continue into the future. We'd expect certainly that overall total bundle of overheads to grow at a lower level in the future. Depreciation was overall approximately flat. It's a very similar portfolio of products to the previous year. And some of the products that we have, we now have slightly longer lives based on the success of them RP2040, we originally were depreciating over 6 years. Its fifth year of trading was its best ever.

So we looked at those lives and we stretched that out a couple of years. So there's a little bit of benefit there on the admin costs. So overall adjusted EBITDA, which is probably one of our key metrics was up 25%. Operating profit up 35% because of that essentially flat depreciation and amortization. And then tax on that, which is an effective tax rate, and that's combined current tax and deferred tax compared to adjusted profit tax was down to about 18% or so. We had a couple of things that came into our favor. RP, one of the pieces that is in the latest microcontrollers and is also in Pi 5 allows us to use the patent box regime [Audio Gap] bit of a catch-up in these numbers because you can look back once you've got that patent. And there was also a provision that we put in previously that unwound.

So there's some benefits in there, got the rate down to about 18%. I think in the future, we'd expect to be about 22% to 23% tax admittedly including deferred tax. So a lot of the good stuff that my colleagues do in terms of whether it's research and development expenditure credits, whether it's the government's various other schemes for R&D, they help on the current tax, the deferred tax kind of puts it back into the numbers. So I think as a tax rate, 22% to 23%. Turning to the balance sheet. Fixed assets increased by about $10 million. That's really capitalization of the intangibles we've talked about plus the depreciation, some of them.

Inventory improved. That's how the finance person describes it. I think my colleague here may say it's a shame that the inventory is not up, but it's down about $10 million. What we saw was a significant reduction in finished goods going through '24. We certainly had some buildup, which we have taken quite significant steps by the end of '24 to bring down. That flows through into '25. It's probably fair to say that we're now down probably as low as we can realistically go on finished goods and speed at which demand has taken up has certainly impacted that. On components, memory, we've improved the holdings as at the end of '25 and similarly on processor chips. So that's really the other side, I think we flagged that it's increased by $21 million. And then receivables are up year-on-year. Quarter 4 of '25, as you saw from the unit volume charts was a very strong quarter, and that's flowed through into those receivables.

It's probably also an element of we now increased prices, and therefore, we're selling at the same board but at a higher price. Payables has come down. Normally, payables and receivables actually pretty similar. They stay in lockstep. The reason for that is at the end of '24, partly is a sign of how weak suppliers were at that stage, they were offering us quite significant extended payable terms, which we took up. We've unwound those through the period of 2025 such that we're now down at more normal level of payables. But the effect of that clearly was an outflow on the cash flow and increase in the net working capital number, but it's something that we cleared. I think we flagged that at the half year, and that's really come through to us. And we really have a couple of million or so of extended payables with longer than the sort of typical 30- to 45-day term that we used to.

That resulted in cash at the end of the year of $28 million, as I mentioned previously, down predominantly I think because of those extended payables, but now starting to sort of settle. And we have $80 million RCF. We uprated that in the first quarter of '25 from $40 million. That remains undrawn. Cash flow, I guess, very much a sort of reprise of just described on the balance sheet as you'd hope. You can see the outflows in respect to the payable reductions, the receivables. CapEx, we spent about $18.2 million of cash on primarily intangibles product development in the year, very much in the range of guidance that we've given at $20 million a year.

I think over time, we'd expect that to rise to sort of 25% of gross profit. But it will be lumpy. We've got some products where come '27 -- back end of '27 into '28 as you actually go to TSMC and ask them to make the chip, there is an upfront payment. If we're buying an item from Arm Cadence, Synopsys and some piece of IP. Again, that may come in particularly even it's for the benefit of the next 3 or 4 years of development. Tax credit, 2025, we had a 2 years' worth of research and development expenditure credit. We didn't quite get the filing in time at the end of '24. So we saw 2 lots of money in '25. So there's a little bit of a lumpiness there. And tax paid, we paid about $4.1 million of tax overall, similar level to the previous year.

So overall, I think $28 million of cash and cash conversion, which is really our EBITDA minus the growth in working capital. I think we would expect over time that to sort of turn back to the cash-generative business that we recognize. I think that sort of conversion of probably about 60% to 80% of EBITDA in the medium term is realistic, essentially no more extended payables. And hopefully, the level of inventory will continue to manage, but it will fluctuate if there's an opportunity to buy lots of memory at a very good price, we will do that. It's really quite necessary. I'm going to finish with a little bit on memory. Any difficult questions though, I will very much point towards this gentleman here who is living it on a daily basis. The long-term trend of memory, as you can see in the black chart from one of the more vocal analysts in the sector on semi analysis over -- since 1973, over the last 50 years, there has been a downward sloping logarithmic curve.

So this is logarithmic. It's -- if you could actually plot it with a normal y-axis, you would see that very neutral downward curve, I think Eben frequently demonstrates on the radio. And that has continued. It's Moore's Law in action. However, what we've seen in the last year on the back of demand for high-bandwidth memory used in data centers is a sudden switch in production. Initially SK Hynix, I think strongly regarded in that space, but rapidly subsequently followed by Samsung and Micron to move their production in that area. That has created significant increases in cost of memory. It's increased the demand -- sorry, because of that reduction, there is more strain on supply as well.

So, we have seen start to see increases. I think it says here, we've seen an increase of about 7x on board. Memory is about 25% of the cost of our boards. We've already put through price increases. We started putting them through in the last quarter of last year. There were some more in the first quarter of this year. I think sadly, there will be more this year as we continue. We'll continue to focus on that profit per unit as being the key measure. So we will take steps. We have good inventory levels, particularly at the lower memory capacity, so the 1 gigabyte, 2 gigabyte. And we have historically had good volumes, but we are eating into those on the 4 and 8 at low historic prices. And we have been using that to partly manage in a sort of controlled way the increases that we're now having to put through to our customers.

So we have probably got enough supply for '26 at those lower memory variants, so 1 gigabyte, 2 gigabytes. We have got good conversations going with a number of suppliers at the other areas. The reputation that Raspberry Pi has is actually been incredibly helpful in some of the conversations we've had with people that we've never talked to before that know of us. And for them, I think that association has certainly smooth conversations. But it is something that we are working hard through, particularly as we go into the last half of '26 and then into '27, making sure we've got an adequate level of memory is a cause of great focus. So we've got more suppliers. We've taken some technical steps. We have the ability to use smaller memory chips to sort of double them up, if you will, to create larger memory.

So that's given us some scope there. We talked to other suppliers. We are making strategic purchases. We will not relying just for that immediate demand, but it is dependent on what you can get. And we have put through price increases, and we'll continue to do that. You can see on the Compute Module 4, 4-gigabit to memory. I think that was something we reduced the price of a year ago, and we've now had to put those prices back up, and that is continuing. The outlook for memory prices, well, clearly, there is people are starting to invest in new fabs. ASML in the Netherlands, basically the go-to supplier of equipment for chip production has signaled significant increases in demand from, I think they said slightly [indiscernible] Korea and Taiwan -- well, Taiwan, obviously, for a lot of uses, but for Korea and clearly for memory production.

But there are a number of other names on the chart here that we are also engaged with in terms of conversation. So we're looking for that. Price increases inevitably will have some of that effect of probably suppressing some demand from other people, which hopefully will therefore mean availability. And so we are seeing -- we are talking to customers. We're helping customers with downsizing of their memory. A few years ago or sorry a year ago, I think the reflex reaction from a lot of people would be, well, I'm going to have 8 gigabytes. That's the biggest number on the box.

They're now thinking hard given those price increases as to whether the software that they intend to run on that board operates at 2 gigabytes, 4 gigabytes of a stretch. And we're certainly for our large customers, helping them with that. So I think we will see some of those effects start to percolate through. So that's the view from me, over to Eben.

So a quick update on progress -- on our progress against strategy. As I said before, this is another fantastic year for product launches. I'm going to pick out just 3 of the 13 or 14 that I think help to illustrate where we're going as a business. Firstly, we have Radio Module 2 launched midyear. This delivers turnkey wireless interfacing for our microcontroller customers, and it bridges the gap between these first-class deep embedded compute platforms, RP2040 and RP2350 and the network. It's a fantastic complement to our microcontroller products, and it's also become a very successful product in its own right. It's very much a demand-led product. It's a product that we built because our customers told us that they wanted it. And that's something we -- that we could not have done historically.

We've become much, much better as an organization over the last 5 years at listening to our customers. Next, we have Raspberry Pi 500 Plus. That's my personal favorite product in the year. It's our latest all-in-one PC, a successor to Raspberry Pi 400 and Raspberry Pi 500. It's really a love letter to our enthusiast base. Our enthusiasts, even as our enthusiast customer base has been outgrown by our industrial and embedded customer base remains incredibly important to us as an organization, both from a mission perspective and also as the primary way, one of -- still one of the primary ways in which industrial and embedded customers learn about our products.

We are our best salespeople. They are the enthusiasts who take our products with them into work. So it's a love letter to our enthusiast base. It's a love letter to the computers in the 1980s that inspired the Raspberry Pi story. And it really does demonstrate that the Raspberry Pi 5 platform really now can pull its weight in a modern client PC. Finally, we have AI Hat+2 arrived just didn't quite make it into '25, arrived just after the year-end. It's the second output of our very successful long-term collaboration with our accelerator partners at Hailo, builds on the success of the first-generation product that we launched in 2024. And for the first time, it accelerates modern AI workloads, including vision transformers for Vision AI applications and on-device large language models and visual language models.

And we'll talk in a moment about why we believe that these capabilities are very important for the future of the Raspberry Pi platform. Of course, we can only build all these amazing products because we, first of all, built an amazing team. We continue to hire into the engineering team in 2025. We added a number of new senior engineers to the silicon team, and we continue to add fantastic graduates from our long-running and very successful summer internship program. We continue to focus on retention. We did have one retiree. 2025 was another landmark year for us. It's the first year that we had somebody retire from our engineering workforce. But other than that, we had 100% retention in the engineering team. Terry, who retired, he was 71 years old. So we did manage to retain him for as long as we possibly could, but he had to go and get married.

We made targeted investments in our business development, application engineering, communications and finance and legal teams. And last but very much not least, I'm excited that Tim Mamtora joined us this month as Chief Operating -- in the new role of Chief Operating Officer. He's the former Chief Technology Officer of Imagination Technologies, and he's a long-time friend and colleague from Broadcom. I've known him for 20 years. He's an amazing individual. We'll be centralizing and formalizing a lot of our operations activity in his newly formed unit within the business, including engineering operations, manufacturing operations, supply chain and IT and cybersecurity. And really, when we do this, as we restructure the organization, we believe that we're building the architecture that will support the next phase of our growth story.

As we build the capacity to interact with our customers, we, of course, build our capacity to learn from them. I'm not going to [indiscernible] this slide, but I just thought we'd share a few examples of the sectors that our OEM customers operate in from digital signage and network gateways to retail and point of sale, smart home, industrial automation and many, many more. And as we talk to our customers, it's striking to me that across all these sectors, our customers share both common strategic concerns and emphasis on time to market and on optimizing that transition from prototype to production. They have a desire to gain access to advanced technology without carrying, without shouldering a high fixed cost base inside their organizations.

And they need to navigate an increasingly complex regulatory environment for IoT devices. And they have -- they tend to have common product aspirations, too. Many of them are looking to bring existing intelligent products to smaller form factors and lower price points. Others are looking to add intelligence and connectivity to legacy products. And they need to accomplish these things in a world where security has gone from a nice-to-have feature to a legal necessity if you want to be able to ship your IoT products. As they do this, all of these customers are confronting what we've come to call the R&D air pocket. That's the challenge of recruiting and retaining the talent required to design intelligent products to bring those products rapidly to market and to keep them in production in a rapidly changing supply environment.

We're positioning Raspberry Pi to our customers as a solution to these challenges. For our larger customers, we're doing that through the board-to-board initiative, direct engagement at the C-suite level with large industrial OEMs. And for smaller customers, we're doing this more traditional marketing activity and through enhancing our physical presence at events like Embedded World. I was at Embedded World in Nuremberg a couple of weeks ago just for a flying visit mostly to talk to DRAM vendors. But while I was there, I was struck both by the hunger for solutions to the recognition of these challenges, the hunger for solutions to them and the growing awareness of and enthusiasm for Raspberry Pi as a supplier in this area. Highlight a couple of our OEM customers. Firstly, Sixfab, they were on our stand.

We actually hosted 6 of our OEM customers on our standard embedded world. Sixfab was one of these running a Vision AI demo, an automated crowd sentiment analysis Vision AI demo. They build ruggedized industrial computers around our compute module platform. And our value proposition to companies like Sixfab is very, very simple. We provide reliable, available, embeddable general purpose compute at an attractive price point, supported by world-class software and collateral documentation collateral. Then we have ProGlove, a long-standing German industrial OEM customer who we serve through our direct-to-OEM channel. In fact we moved them from our direct to resell channel. The one of our poster children for our ability to move high-volume OEMs as OEMs go to ultra-high volume, being able to move those OEMs from the direct reseller channel to the direct to OEM channel with a corresponding improvement in our margin participation.

They build wireless gateway products for warehouse operations around our 02W, our very cost-effective 02W single board computer. Again, a very simple proposition to customers like this that we're providing them with that perfect mix of compute, interfacing and connectivity that in ProGlove's case, they were able to use to rapidly add new interactive features to an existing platform and that we back those products with long-term software support and long-term availability guarantees. We had that moment where you realize that you have fewer pieces of paper than you have slides. Let's see how we go on. A word on the Edge AI opportunity for Raspberry Pi. I think the first thing to say is, as we talk about this opportunity, we come to this opportunity from a position of enormous strength.

We are already the most popular platform for edge computing and many of our customers' workloads already have an AI flavor to them. These applications are either running on the CPU or sometimes they're accelerated by our accelerator partners, including Hailo and Sony. And we have fantastic relationships with our accelerator and model vendor partners, which we leverage over time to expand the platform performance envelope to expand the range of things that people can do with our products. There's a cluster of megatrends, I think, that help to define and underpin what this edge AI opportunity looks like. Frontier models in the cloud, they're hugely efficient and they're hugely capable. But when if you choose to build your -- if you're trying to build an intelligent IoT application and you try to build that on top of those cloud AI platforms, what you're building into your IoT application is you're building in latency, you're building in ongoing cost and you're building in a dependency on the availability of the network, which translates inherently into [indiscernible].

Now running your -- building your IoT application on AI compute that happens at the edge of the network in contrast potentially brings with it improvements in cost structure, it brings improvements in latency and it brings resilience to network disruption alongside addressing pervasive concerns about privacy and certain concerns around regulatory compliance. What's the opportunity for Raspberry Pi here really? It is to translate our dominance in Edge Compute into a dominance in Edge AI. What's going to support us in doing this? The growing salience of low latency, particularly in applications, including robotics and the gradual reduction that we see every generation of AI technology.

We see a migration when a new innovation appears in AI, we see that innovation begin in the context of very large models running on very high-performance computers at the center of that. We certainly saw this in Vision AI. Our largest OEM customers today are Vision AI customers, the sorts of workloads they are running on Raspberry Pi at the Edge of the network today, a decade ago would have required the largest NVIDIA GPUs running on large servers at the edge of the network. Two things have happened in that decade. One, the performance of Raspberry Pi computers and performance of the devices at the edge of the network, even unaccelerated even without the assistance of our partners at Hailo and Sony, the performance of those devices increased by roughly a factor of 30.

At the same time, the compute demand models as people have refined the model architectures found ways to make those models smaller, faster and more quantized, the compute requirements of those models have come down by roughly 2 orders of magnitude. We've seen this in Vision AI. It's been an enormous enabler for the current wave of AI applications on Raspberry Pi. We expect this to happen with generative AI applications. We expect this to happen with LLMs and vision language models. We're already seeing it happen today. There's been a lot of noise recently about people running OpenClaw, about people running Agentic AI on Raspberry Pi platforms. Now the interesting thing about this first wave of agentic use on Raspberry Pi and lot of other small, slightly less low-cost computers from other companies is that all of the compute, all of the generative compute in these applications still runs at the center of network effectively.

These devices are shins, which connect the local environment to remote compute. And I think what we're going to see over time, I think the opportunity for us is by becoming first the dominant platform to run that sort of thin local -- thin local agentic AI, it positions us well to intercept the local compute opportunity as the performance of the local devices increases and as the computing demand of the models declines. A word on semiconductors. As I say, this was a crossover year for us. We sold more semiconductor devices than we sold boards for the first time. Financially, this is relatively immaterial. We are talking about $0.50 microcontrollers and $50 boards. But the vision of the business, what do we want the business to look like in 8 years' time? What do we want the business to look like a decade after the IPO? The vision is to have a 2-franchise business in which we have an electronic products business, larger than the electronic products business today, but at the same time, to have the semiconductor business grow up alongside it to become a financially [indiscernible] business.

That implies that we need to be selling at least hundreds of millions of microcontroller devices if the devices we're selling in 8 years' time are at the current ASP, we do need to be selling hundreds of millions of them to realize that ambition. And I think it's hugely significant that barely 4 years after we launched our first microcontroller, we'll be able to get to a point where we're selling more microcontrollers, more semiconductor products than we are selling [indiscernible] products. What does the road ahead look like? Continued sales growth in microcontrollers. We took -- we had good momentum in the second half of the year, took that good momentum with us into 2026. New microcontrollers and Raspberry Pi Pico products and other silicon devices.

You can split our silicon group work really into 2 clusters -- we have the cluster that generates microcontroller products, the cluster that generates products which we use ourselves and sell to third parties, and we have the products which exist purely to enhance the performance of our in-house products. In terms of the microcontroller road map, you'll see, I think, from us over the next 5 years, a regular cadence of product releases. I think we are going to be a little bit more open not today, but probably starting in 6 to 12 months, we're going to be a little bit more open with people about what that road map looks like. You're going to see a regular cadence of releases, whether those are addressing new price performance points, the 2 products, RP2040 and 2350 that we produce today sit very squarely in the center of that kind of parameter space -- you'll see us exploring new price performance points.

You'll see us trying to find new ways to integrate interfacing and connectivity options into the devices, perhaps qualifying extended operating points. We haven't indicated the desire to get involved in producing microcontrollers to the automotive market, but I think we do know what we need to do if we wanted to do that. What's the focus here? All of these things are intended to grow TAM and they're intended to grow TAM without blowing up the number of products we make. We always want to be an organization. We have to be a low mix organization.

We have technical innovations, which we believe enable us to address large parts of that parameter space with relatively small numbers of [indiscernible] then in terms of the devices that we use to make our board and modular products better, we've had huge success with RP1. This is the IO controller that we designed for the Raspberry Pi 5 platform. And I think you'll see us build on that, try to build devices which bring higher performance, richer feature sets to our board level products, obviously improved unit economics as we just place vendor margin out of the margin stack in our products.

All of these things, they deepen our competitive moats and they give us that security of supply and long-term availability. Today, when we launch a product, we communicate a desire to be able to produce that product at least into the late 2030s. We have some fantastic supplier relationships, which underpin those commitments, but there is no underpinning more powerful than knowing that you build the silicon platform yourself. A word on Raspberry Pi Connect. Really, I think one of the standout stories of 2025 historically, what does Raspberry Pi do? What is Raspberry Pi's value proposition? What is its core if you boil down our value proposition to our OEM customers. What we are doing is we are detecting things that many, many, many OEM customers do. We are doing them once. We're doing them well, and we're selling the fruits of that work to our OEM customers and making margin.

Now historically, when we've looked for things that OEM customers do, we've had a, I guess, a mental block, right? We've had a tendency to look only at the boundary of the Edge product, the physical product that we sell, whether that's the hardware or whether it's the low-level software that runs on that device. Really Raspberry Pi Connect represents us acknowledging that particularly in the Cyber Resilience Act era, our OEM customers, there are important network level pieces of software that our OEM customers need to develop in order to make our products operative, operational inside their products.

Connect launch in 2024. Last year, we had an over-the-air update capability. That's a key underpin for people wanting to comply with the cyber resilience while building their products around our hardware and software products. We added that at the end of last year, left the year with very nearly 400,000 connected devices on Connect. That number is over 500,000 units today. So still seeing very, very robust growth, both in the total number of registered devices and into the proportion of those devices that we are converting from the free tier Raspberry Pi Connect to the paying tier Raspberry Pi Connect for organizations. So an increasingly important part of our business. And then a word on outlook. As I said, we took -- saw very, very strong momentum, building momentum last year, brought that momentum with us into the start of 2026.

We have the demand creation initiatives, notably the board-to-board program, which we believe will drive continued growth through this year and beyond. In terms of supply chain, we do expect, as Richard said, that supply chain challenge in DRAM will only be decisively terminated by the addition of foundry capacity in the much then Far East. We do expect that to take time. We don't expect relief from that source this year or next, although we do believe that the relatively extreme levels of DRAM pricing we're seeing now are going to drive -- discover and drive elasticity in the market over time. Still somewhat limited visibility in the second half, but we are well provided for our density, and we do believe that we have the vendor relationships, old, long-standing and new vendor relationships required to navigate these challenges effectively.

And of course, we do like to remind people that about 1/3 of our products by volume, the classic Raspberry Pi products, the Raspberry Pi 3 and earlier and the Pico-based products either use a different supplier from. The use LPDDR2, which we have very, very large wafer level buffers or in the case of Raspberry Pi Pico products don't require SDRAM at all. So roughly 1/3 of our sales are unaffected by this dynamic. In terms of pricing outlook, we are indicating that we expect profitability to be in line with market estimates, but we are flagging that the ASP increases that we put through in the past and the ASP increases that we may need to put through later in the year are likely to lead to revenue being substantially higher than market estimates.

I'll leave you with -- I'm not going to belabor this slide either. I'll leave you with some quotes from some of our favorite OEM customers. I'd just like to take this opportunity to thank the team for all the work that's gone into getting the results ready this year, particularly also to thank Richard. This is likely to be Richard's last final results with us. He joined us in 2019, when I think the finance team was about 3 people, was instrumental in building a finance team, which was fit for purpose and obviously seen us through the IPO process and has been an extremely valued colleague and a very good friend for these last 6 or 7 years. So I'd just like to extend my personal thanks to him for everything that he's done for us. We certainly would not be where we are without his efforts. [Audio Gap].

Good morning, everyone, and welcome to our 2025 first half interim results. We're going to start by sharing a few highlights of the half, and Richard is going to take us through the financials, and we'll round off with an update on some of our strategic programs and an outlook for the second half and beyond.

We're very pleased with our performance in the first half of the year. We saw a 9% sequential increase in unit volumes, underlying demand growth from our existing and new OEM customers. And this happened in the context of a normalized channel inventory position. Recall that heavy inventory both in channel and at our OEM customers was a significant drag on the performance of the business in the second half of last year.

So an 8% sequential increase in direct unit sales and a 27% increase in direct revenue, reflecting the success of our strategy to move more of our large OEM customers into the direct channel.

A good product mix, certainly a better product mix than in the second half of last year, which together gave us a 19% sequential increase in adjusted EBITDA. Just to probably address a couple of potential headwinds on the company. Obviously, we are now in a high and dynamic, probably more importantly, dynamic tariff environment in respect to sales into the U.S., we've looked and we've seen no evidence of a significant impact of those tariffs on our demand from U.S. OEM or consumer customers.

We do see this in the medium to long term as a significant source of differential competitive advantage over many of our competitors who tend to be manufactured in the Far East. And then there's been a lot of noise recently about DRAM pricing and DRAM availability. We entered the year with substantial stockpiles of DRAM, tens of millions of dollars of stockpile of DRAM across all densities.

We are -- those stockpiles will take us through lower densities. That hedge will take us through into 2026. We do have some exposure to market pricing to a contract market pricing, not spot market pricing at higher densities at the 8 and 16 gigabyte points but sufficient supply to take us through to the end of the year.

In the longer term, of course, we do have a number of technical and commercial levers that we can pull to mitigate our vulnerability to sustained high DRAM pricing. It's been a good half for product launches, where 2024, I think, was a year of platform refresh.

We refreshed both the large Raspberry Pi, the Linux-based Raspberry Pi platforms and the small Raspberry Pi, the microcontroller-based platforms. 2025 is really a year of sustaining innovation. We'll talk a little bit more -- we'll give a little bit more detail on that later.

And we'll also talk a little bit more about our board-to-board initiative. We're seeing encouraging progress in our board-to-board initiative that aims to open up CTO level conversations with major industrial OEMs and the progress -- the very substantial progress we've been making this year in our semiconductor franchise.

So I'll turn it over to Richard for a little bit for some financials.

Thank you. Good morning. The financial review of the first half. I'm going to start with the first slide, if you would. And so really, for a bit of context, we've got this first graph here showing units by quarter. And within the categorized, you can see below the line really are our major boards. So what is that? That's sort of reading up from the bottom, it's the compute modules. It's Pi 3, which we're still selling significant volumes of having even launched the product in 2016, but continuing to be a major component of our unit volumes, Pi 4, through the period and then Pi 5, which was most recently launched at the last quarter of 2023.

And really, why do I put this up? It's the changes we went through in '24 into '25, very helpful, I hope, in explaining some of the trends that underlie the particularly half 1 against half 1 comparison. So quarter 1 of 2024 came into the year, we were still seeing the backlog of orders that have been placed during the time of shortages in '22 and '23, particularly in the area of compute modules. And that was really the last significant part there. And then you can see the lighter blue Pi 5, actually very strong in that period. We've just launched the product and therefore, a lot of volume come through into quarter 2 and then quarter 3 and 4 of last year, where we saw this channel indigestion with so many people had bought boards.

And really the period has that worked its way through. Key thing though, in quarter 3, obviously, one of the things that helped it. We launched Pico, board at $3 or $4, though, so not a lot of gross profit per unit. But clearly, as a unit volume quite significant. And the piece that we've -- Eben has already mentioned, we mentioned actually at the time of the full year results.

Really that return to industrial OEM demand in quarter 1, quarter 2 of this year. Has the channel stabilized? And what did we see? We saw that strength. You can see in the bright orange Pi 3, very much an industrial board. Something of that age is very much a product that people have built into a design and they continue to use.

And so we saw that come through and that return to volume in the second half -- sorry, in the first half of this year where those main boards have come through. So that's the underlying sort of trends. I don't think we'll give quarterly numbers very regularly, but I think is something of the pattern of what we've been through over the last year and into that period sort of I think much more confident growth now is, hopefully helpful.

So overall, just looking at some of the key metrics, unit volume. We sold 3.6 million boards in the first half. That's down very slightly on $3.7 million we did a year ago, but up on that second half of last year, some 10% up on the second half of last year. Gross profit per unit down a little on 2024. Why is that? Well, that was the very strong gross -- sorry, strong profits that we had on compute modules in the first half of last year offset by some of the additional profit we're now making because we have the cheaper Broadcom chip in the Raspberry Pi 5.

That goes to a gross profit that was in truth more or less in line with 2024, aided by essentially the same sort of performance in accessories, but also additional profit that we're now making on semiconductors, which we'll talk a bit more about later.

Overall, that came to an EBITDA of about $19.4 million, more or less in line with 2024. With our administrative costs, talk a bit about on the second page, continue to be sort of tightly controlled. One of the figure here. Cash at the end of the year was $34.3 million. That's down on the position at the end of the year, really with a reduction in payables.

Inventory, we have started to -- inventory is in a good place particularly in finished goods, we've brought that down. And the real reduction is a one-off reduction in our extended payables, which we'll talk about a bit more later. Thank you.

Just quickly on units per half, very much as previously described. It's worth noting the change in the sort of license volumes there, down after a strong Pi 5 launch was direct sales, the sales that we make to OEMs and resellers have risen consistently through the period, partly because of new products. And so things like Pi 5 with 2 gigabytes and memory, but also Pi 3s and compute modules returning to strength.

Gross profit per unit in the half 1 of 2024, I think we've alluded to, compute module volumes, which are a much better margin product was strong and that was offset by the additional cost of the 2712 C1 chip that we had from Broadcom, which was the first chip, $5 more expensive than where we are now, that we use in the Pi 5 initially.

During the second half of last year, we had then sequentially this Pico launch, which was definitely at a lower volume and a general slowdown in the mix of products. And then in the half -- first half of this year, more 2 gigabytes and 1 gigabyte products being sold, very much this return to OEM demand, together with Pi 3 offset in part by that cheaper Broadcom chip, which helped sort of push the profit per unit back up.

And I think we'll see some of that trend into the second half. I think we'll continue to see a slight improvement through that period. We had a few other sort of one-off costs within there that should benefit that profit per unit.

So moving to the next slide, I think it was Frederick the Great when presented by the Brandenburg Concertos complained too many notes -- my apologies that there are probably too many numbers on this one, a little bit of exuberance. As you can see, gross profit per unit -- sorry, gross profit down a little bit, but then admin expenses overall, a slight increase, a few more heads and some more costs from being a listed company, whether that was additional sort of areas like sort of Investor Relations, but offset by -- in this period by some foreign exchange gains on balances of about $700,000 or $800,000.

So we have various sterling receivables, which we had some small gains on over the period. That brought down to an EBITDA of about $19.4 million. Amortization was up, I think as we flagged up by about $1 million. Principally, we relaunched the Raspberry Pi RP2350 chip, codenamed Amethyst at the time, which was launched in August. And from that period, we started amortizing. So no amortization on that chip in 2024 and about $1 million of amortization in 2025, which is why there's a step-up there.

That brings you through to adjusted operating profit for the period of $13.2 million. After that, we've got adjustments. These numbers do not include share-based payment costs, the cost of the new scheme compared to the cost of the residue of the pre-IPO scheme that we had. And sadly, depending if you're someone who would like Rachel Reeves, fortunately, we now suffer and we have to accrue national insurance on that.

So there are additional charges within there. Financing costs, those are up. We have some leases. They are based in sterling, and therefore, we have a foreign exchange loss on those, that's included in that financing charge. And we also have some discounts on some of the longer-term payables, which we'll talk about a little bit more.

And those will unwind through this period. So that increased our financing cost by circa $1 million. And then tax is lower, tax is low for two reasons, partly yes, a bit less profit -- but the other aspect is we actually had within there some gains on sterling receivables. We had a tax receivable at the end of last year as sterling appreciated against the dollar. We have about a $700,000 gain which is in that line.

Underlying tax rate is very much 25%, but that gain on foreign exchange means that the effective rate that you'll see for those cognizant of tax accounting, is about 13%, which is low but not something that could be repeated into the future. That brings us down to an earnings per share.

Adjusted profit before tax. So this is excluding the cost of share schemes and the one-off costs that we had in 2024, which were relating to the IPO. That's down to $9.2 million from $11.9 million previously. Finance costs, I think I said, slightly higher, brought that down.

Share-based payments, higher by about $1.6 million, but tax offsetting by $1.5 million. So that's overall there. Shares in issue coming to that adjusted basic EPS, $193 million. That's up significantly on the number for last year, but please remember the number for last year was actually pre-IPO, and therefore, is really before the effects of the primary raise that we did in June and also the crystallization of the pre-IPO LTIP. So really a comparison on that line. It's not just apples and pears, it's apples and potatoes. It's a whole variety of things within there.

But going forward, that $193 million, more or less stable in the future. If you want to bring in dilution effects from the share option schemes and things, I'm sure we can help you, but it is quite a complex story at the moment.

Looking to the balance sheet side. On inventory, particularly broadly flat year-on-year, so up only slightly compared to the year-end. We've managed overall to bring control on inventory very much something that we said at the half year last year that -- we have seen the slowdown.

Our production inevitably ran about a month ahead, and then we've gradually brought that in. We've reduced the overstock that we had at that stage of Pi 3 is working very closely with our manufacturing partner, to bring that down. So finished goods, I think we're in a very good place. We've intimated that maybe in some areas, we're really about as low as you could possibly go.

I think Eben may talk a bit more about we do have some order backlog at the moment, which we'll see unwind. So maybe a little bit of an increase there, but pretty much now down to that sort of level.

And then what we've seen on the other part of our inventory, which is components to what is this, this is memory stock. This is stock of processor chips and other sort of key components and that has continued to rise slightly. We've continued to see some increase in particularly processor chips. We have never wanted to have a repeat of what happened to us in '22, '23. So we're continuing to see some step up, but on overall level, but I think we'll stay fairly steady into the future in terms of inventory at around about that sort of $150 million mark and may even come down slightly over the period as some of the delivery patterns change into 2026.

Trade payables. Trade payables, we finished June '25 with about $80 million of trade payables. That is a reduction on where we were at the year-end. I think at the year-end, we were at about $96 million, and that's predominantly in the area of extended payables that we had taken from suppliers at the end of 2024.

The market at the end of '24 was definitely one in which it was really quite soft for them. So they were offering us in addition to some very strong payment prices on memory, also terms in terms of paying over longer periods, be that 6 months, even up to 12 months in some cases. So over that period, we took the benefit of that. That has been running down.

That's really the key theme within the cash flow for the first half of this year is that reduction in those payables. That will continue into the second half. We'll be pretty much clear of those payables, which is this number here, about $30 million by the end of -- $22 million by the end of this year. So we would expect to see a reduction in cash through to the end of the year. But there is not that pretty much -- there is no more after that.

What do we see after that? We see payables and inventory really staying pretty steady through the remainder of the '26. And therefore, our profit offset by CapEx, but therefore, generating some cash, which should start to actually make our cash number in '25, '26 start to improve.

Looking at our cash flow for the first half, very much the pattern we've just mentioned, finished the year '24 at $45 million. EBITDA of about $19 million, small increase in inventory, which we had to pay for. And then that reduction in payables that you can see of about $18 million, which is the significant outflow. CapEx, the $19.5 million, consistently at that around about $20 million a year level that we've talked about before.

And so you see by the -- at the end of the period, $34 million. The payables, we see there is more to be paid by the year-end, but then it stabilizes after that. So cash comes down at the end of the year to $34 million and then -- sorry, at the end of the period and maybe comes down by about another somewhere between $15 million and $20 million by the end of the year as we clear out the last of those extended payables.

Clear out perhaps we're not being offered those anymore before moving into a period of positive cash flow in '26 as really that inventory and payables balances match one another through the year. And therefore, we see positive cash flow and an increase in cash coming into through '26. Finishing off, just a little bit on our reminder in some ways on our business model.

On the revenue side, the relative profits of the channels. If we are selling through a licensee royalty channel, we expect a sort of profit of circa $5. If we're selling to a reseller, we would expect to make a profit of about $10. And if we're selling actually to a direct customer, say it's about $15. And it's really that sort of step, that ratio 10, 20, 30 as being the ratios between the channels. And therefore, that move between the channels has quite a significant effect on our profit per board.

In terms of the unit economics, what goes into those; we've got the cost. We've got the costs being by size, really the process of cost, so that's what we're buying from Broadcom, our DRAM costs, so higher memory variance, more expensive DRAM. And then other components. And then finally, some costs, not a large proportion, don't tell Sony, but not a large proportion of our costs being the manufacturing in Wales.

In terms of how do we manage the margin on that and margin upside. Memory density, the more memory that we sell, we do make a margin. If you're selling a chip with 16 gigabytes of DRAM, you are going to make more margin than if you're selling 1 with 2 gigabytes of DRAM, very much, I guess, the Apple iPhone model. Custom boards, we have been making more boards for particular customers who have the volume to justify the engineering investment and that ability to bring that together to bring some savings for them and saving for us has opportunity to improve our margin. We've certainly seen that in the case of the ones that we've done in the last year.

Additional variants. So if we're making them with higher temperature ranges, we can charge a slightly higher premium for those and that opportunity. And of course, there's accessories. We continue to sell accessories. We're making at the moment, about $1.10. It has been up to about $1.30. I think there's a bit of upside there still to come.

And the nice thing we're seeing with accessories, particularly at the moment is actually people are buying them in their own right rather than an attachment to a Raspberry pie. So there's an increasing trend for people buying our power supplies because they're actually just very well engineered power supplies with the USB Connector.

And finally, there is that portfolio lever opportunity to increase prices on products. We have done it in the past. We've reduced prices on products in the past. It has always been there, but it's something that is not entered into lightly because that is very much the integral part of how we have a consistent proposition to customers. We are -- still -- we still surprise new people by just how cheaper a Raspberry pie is, and we are perpetually aware of the risk that somebody in the Far East will design a board that will be cheaper.

And I guess very much with Christensen's Innovator's Dilemma in mind, someone else will come in beneath us if we become too greedy on margin. Nonetheless, our target is to grow gross profit over time. And that is through those margin upside opportunities that we have there and that mix of products.

Okay. So let's take just a little look at how things have been going on the strategic side of the business. Starting off with products. As I said, another strong year, a strong half for product launches with a different character in 2025 to 2024. In 2024, we completed the rollout of Raspberry Pi 5. We launched Compute Module 5 and Raspberry Pi 500. So essentially completing the refresh of the large Raspberry Pi platform. We launched Raspberry Pi Pico 2 and Pico 2W in the second half, powered by RP2350, our second-generation microcontroller.

2025 really more about sustaining innovation, identifying gaps in our product lineup, addressing specific pieces of customer feedback and finding ways that we can put incremental improvements into existing products. I have a couple of examples here.

Richard already mentioned extended temperature variants of Raspberry Pi Compute Module 4. Historically, our compute module products have been commercial temperature range products certified for operation between minus 20 degrees and plus 70 degrees Celsius. The extended temperature variances of these push that out to minus 40 to plus 85. So they broaden the range of applications that these products can be used in, obviously, in addition to growing the addressable market.

These are features that we can charge, that we can generate incremental margin from selling the higher-performance devices. And then we have Radio Module 2, we have a lot of OEM customers prototyping with our Pico W and Pico 2W products, which have integrated modular wireless. Radio Module 2 provides a path to scale for those customers. It allows people to take the radio functionality from the Pico products, go chip on board with our microcontroller products without facing a substantial radio conformance burden. We're not done yet for the year coming up in the second half.

We don't talk in detail about future product releases, but we have a major platform release, which is fairly imminent, which targets our education enthusiasts and enterprise customer base. Last year, we launched our first AI product, the AI camera, AI Kit and AI HAT products. We have a refresh of one of those products, which adds support for more modern workloads, particularly for generative workloads.

We have a cost-engineered compute module. In fact, this was launched this morning in China. We have a cost-engineered compute module product specifically for the Chinese OEM market. It's the first time we've done a geographic-specific product. And there, we're really recognizing the very high level of price sensitivity in the Chinese modular computing market.

And finally, on the software side, we're planning to roll out. We're seeing good momentum in our Raspberry Pi Connect IoT cloud platform, and we're planning to roll out at the end of the second half or possibly might slide slightly into next year, extra functionality, in particular around over-the-air updates. We are providing our OEM customers with a turnkey solution for over-the-air and regulatory compliant solution for over-the-air firmware updates for their deployed IoT objects.

Talked before about our board-to-board initiative. We initiated this in the aftermath of last year's IPO. And what we're trying to do here is reach out and have conversations with senior decision-makers at large industrial OEMs to understand the opportunity for Raspberry Pi at those companies.

The goal there to generate individual design wins, which move the needle. In our general OEM customer base, our median OEM customer is under 1,000 units a year of volume. We have a very broad OEM customer base. What we're trying to do here is generate more of what we call whales, the 50,000 plus 100,000 plus customers. Really three elements to this effort, discuss with these -- once we got in the door, discuss and identify the barriers to scale the use of Raspberry Pi technology.

We almost always find that these customers are using Raspberry Pi in some capacity generally in prototyping or in production and test automation. Some of them are already at scale with Raspberry Pi products, some of them aren't. So trying to understand what the barriers are, which are preventing them from being at scale with Raspberry Pi products, whether that's misconceptions about our business, whether that is misconceptions about our products, whether there are genuine gaps in our product offering that we can address through additional documentation, through additional performance testing or through changes to the product lineup. That's where the industrial temperature -- that was the genesis of the industrial temperature grade Raspberry Pi Compute Module 4 products.

And then internal barriers to adoption, internal lack of capacity, we'll talk a little bit in a moment, finding significant number of OEM customers who could be designing with Raspberry Pi, but where the barriers to adoption are largely inside those organizations rather than in our organization and trying to understand what we can do to address those.

Then once we've had those discussions, working at the senior level to develop a shared understanding of the opportunities inside that organization and then to assist in matching those opportunities to our products and/or to our capabilities.

And finally, then working with engineers to design products with Raspberry Pi and then bring those designs rapidly to scale production. These are -- we say rapidly, these are multiyear efforts. The very quickest that we can reasonably expect to do this is about 12 months, but we expect some of these will be a 2- or 3-year effort before they generate material impact, material financial impact on the business.

But I think you'll see us address these opportunities with a mix of standard products and customization. Traditionally, we offer customized products to customers who are already at scale with our standard products, but the scale of some of these opportunities probably indicates that we may address them out of the gate. They may be suitable for customization, candidates for customization right out of the gate.

And what sort of things are we finding when we talk to these OEM customers? The first thing to emphasize, great domain-specific engineering capabilities. These are amazing engineering organizations, which are already very, very good at whatever it is they do. They have great enthusiasm often for the idea of adding connectivity and for the idea of adding intelligence to their existing product line, but they face a series of common barriers, really not just barriers to adoption of Raspberry Pi products, but more broadly barriers to the addition of intelligence and the addition of connectivity to those products.

What are the barriers? Well, the hardware and software platforms that they need to work with in order to add connectivity and intelligence are becoming more complex to work with, both the hardware and software environments on their own and also the co-optimization of the hardware and software elements of the platform to meet the required price performance point. There's obviously an increasing focus on security, reliability, availability, serviceability. Some of this is reputational.

This is these companies not wanting to put product into the field, which is going -- which is going to perform poorly in the field. Some of this is regulatory, particularly in the security area, there is increased regulatory scrutiny of the risks associated with deploying insecure IoT devices at scale.

They have the usual commercial considerations, but also, I think, probably more modern post-pandemic concerns about supply chain integrity in what's an increasingly complex geopolitical environment. And this was an enormous surprise for us, a difficulty in recruiting and retaining design engineers within the organization. I think you're going to see a lot of focus from us on understanding the extent to which people are not designing with -- understanding and mitigating the extent to which people are not designing Raspberry Pi simply because they lack the internal electrical and software engineering capabilities to work with our products.

We do see Raspberry Pi technology as an answer, a complete or partial answer to all of these issues, very easy to work with, very highly integrated, a turnkey software platform, which, to the greatest possible degree, reduces the amount of internal capacity you need inside the organization and of course, the U.K. supply chain.

It's important not to neglect the broader market. As I say, our median OEM customer is under 1,000 units a year. So even as we focus on developing more of these board-to-board relationships, we do need to keep investing in the engine that's powered the growth of Raspberry Pi over the last decade.

How are we doing that? Investing more in physical events, primarily trade shows. I think we're attending 6. We have significant attendance at 6 trade shows this year, up from 3 last year. Virtual events as well. First and third party, we have -- we've begun a very successful webinar program, particularly focused on non-English language webinars on our own platform and on other people's platforms.

Better use of our existing communications assets. We have some fantastic assets in our website, in our newsletter readership and in our social channels, really trying to use those -- make more use of those assets and make use of those assets specifically in the context of supporting OEM customers. We've been working with partners on messaging. That's particularly important in China, where there can be significant cultural and language barriers to understanding the value proposition of our products.

And finally, early access programs. I think if you look at the demand -- the early demand curve for RP2350 and Compute Module 5 and compare that to the early demand curve for Compute Module 4 and RP2040, you really see the fruits of a big effort that we put in, in the tail end of 2023 and into 2024 to get our key OEM customers early access to these next-generation platforms, and you'll see more of that from us over the next couple of years.

Case study slide. We love case studies. We think they bring our customer value proposition and competitive advantage to life. The examples here are taken from a recent visit to Brazil by Roger Thornton, who runs our application engineering team. 3 -- I think 3 very typical -- if you look at our global OEM base, 3 very typical applications. The first one, Pro Data to make transport products. This is a payment gateway for public transport solutions in Brazil.

Second one is an Agritech is a precision agriculture, precision irrigation solution. All of these are compute module. All of these are compute module based. Precision agriculture and then fuel station automation. So I think really kind of good examples right in the heartland of our OEM -- of our traditional OEM business.

Why Brazil? Well, historically, we've seen South America is the last market, the last significant market that we've yet to make significant direct investments in. What's fascinating was to go there, was for Roger to go there and see even though this is somewhere that hasn't benefited from enormous investments from us in the past, we're still seeing a landscape and OEM demand landscape, which is very, very familiar to us from our work in Europe, North America and Asia Pacific.

A word on semiconductors. We've always described our electronic product offering and our semiconductor offering as mutually supporting franchises. The semiconductors that we make allow us to build better electronic products and those electronic products then become the shop window for the semiconductors. They are the first place in which design engineers who may in future work with our semiconductor products encounter those products.

We have an aspiration over the next decade to reach a point where the semiconductor side of our business, the semiconductor franchise is making an equal contribution to our business alongside the electronic products. A really significant milestone for us in the first half in that we sold more chips than boards for the first time. So you have to remember, these are $0.50 chips versus $50 board.

So this is not yet financially material for the business. But this is the first half year in which more of the compute experiences that people are having with Raspberry Pi, more of the boxes that people buy that have a Raspberry Pi logo on them are chips rather than board-level products. It's driven by RP2040. Gives you an indication of the length of the design cycles in this industry. We launched Raspberry Pi -- RP2040 in January of 2021.

So it's very nearly a 5-year-old product, and it's still ramping extremely aggressively in the OEM space. Shipments were up 69% sequentially. A very broad customer base for that broad general purpose microcontroller customer base, but we did have a smaller number of very substantial wins in the half.

I'll give an example here, WIZnet, Taiwanese networking product vendor. They make Ethernet chips. What have they done with RP2040? They've taken their Ethernet silicon and they've taken our compute silicon, and they packaged them together in what's called a system-in-a-package to produce an intelligent Ethernet controller for IoT applications.

Absolutely not a line of business that we envisaged being in when we launched RP2040, but I think a really good demonstration of the fact that if you make great product and you put it out in the wild and you make it available to people, the market will discover -- this is really the story for Raspberry Pi, right.

If you put great products in the wild, the market will discover what those products are for, and you have to be agile enough to respond to that demand signal.

RP2350, obviously, earlier in the adoption curve. We launched that in August of last year. We put it in general availability in March. We put a -- we had a one significant analog bug in the A2 launch silicon. And we had a number of security upgrades, which came out of a hacking contest that we sponsored in the second half of last year.

Last month, we launched the A4 stepping of the silicon, which addresses all of those issues, which rolls up all those fixes. We had intended to defer the general availability, the general distribution availability of RP2350 until the A4 silicon was available. We got into the first quarter of this year and the volume of we have a gentleman, Chris Boross, who works for us, who runs the early access program for the silicon.

And simply in order to stop his inbox melting, we put the A2 silicon in the market just as a way of managing the very, very high level of interest in that product, put in the -- put that into general availability in March, released A4 in August and cycled all of that residual A2 inventory out of the channel, replaced it with A4, seeing really, really good adoption and inquiries from Tier 1 consumer electronics companies where I think RP2040 probably a kind of sort of Tier 2, Tier 3 companies, definitely kind of Tier 1 consumer electronics companies looking at RP2350 now. Really an illustration of, as Richard said earlier, the second product that you put into the market is the one that really establishes your credibility. It builds on the momentum of the first product, and it establishes a product line. People are much more to invest in a product line than they are to invest in a product. We certainly saw that with Raspberry Pi 2 back in 2015, and we're seeing this now in the semiconductor franchise.

Finally, to the outlook for second half and beyond. We do continue to see improvement in underlying demand, and we expect alongside that in the second half to be able to address much of the very substantial customer backlog that accumulated in the first half of the year, which obviously will make a contribution to second half volumes.

We've got confidence in the unit economics, a very similar product mix to the first half and sufficient DRAM supply, much of which was acquired at what now looks like extremely favorable historical prices and certainly at the lower density -- the lower densities enough for that to last through the year and into the quarter of 2026.

In consequence, we're not expecting any change to our profit expectations. And then looking further out, continuing to see a good demand environment, continuing to see good progress and perhaps the first contribution in 2026 to volumes and revenue from the board-to-board work, good levers for mitigating any continued strength in the pricing of DRAM and the opportunity for continued growth in our semiconductor business. Thank you very much indeed.