Good morning, and welcome to our final results of our inaugural final results briefing for 2024. Let's see, who are we? I'm Eben. As I guess many of you know, I'm a silicon engineer by background. I've been running Raspberry Pi, founded the Raspberry Pi Foundation back in 2008. I've been running what was then Raspberry Pi Limited since we spun it out of the foundation at the tail end of 2012. I'm joined by my colleague, CFO Richard Boult. A very brief recap of who Raspberry Pi, who we are, who's Raspberry Pi. We build compute platforms, by which we mean hardware, the software that runs on them, the low-level software that runs on them, and the supporting collateral that helps our users design with them.
We sell into two markets: enthusiasts in education, which I guess was the original Raspberry Pi market, and now into industrial and embedded, which over the last few years has come to be the vast majority of our business. We design and build our products in the U.K. The vast majority of our R&D activity takes place in Cambridge. We manufacture the vast majority of our products with Sony in Bridgend in South Wales. We've sold a little over 65 million Raspberry Pi computers since selling the first on the 29th of February 2012. Over 90%—we make almost everything in the U.K. Over 90% of what we make is exported outside the U.K. We benefit from a very supportive shareholder base, including a couple of pre-IPO strategics, Arm and Sony. Of course, our longstanding and still largest shareholder, the Raspberry Pi Foundation.
Some highlights of 2024. Obviously, the IPO was on the 11th of June, was a standout event, raising $185 million to fund the work at the Raspberry Pi Foundation and raising $40 million of new money, which we're using to fund new technology development, development of future generations of our technology platforms, and to add some resilience to our supply chain. We were very pleased to welcome back two pre-IPO investors, corner stone in the transaction, Lansdowne, and Arm. Of course, met many new investors in the process. It was an interesting year from a market perspective. It was the year in which we completed our recovery from the shortages associated with the aftermath of the pandemic. We went into the year still with many of our products in a shortage situation. By the end of the first quarter, we had largely cleared those shortages.
As we discussed in the interims, we did see some elevated inventory levels in the aftermath of that clearance. Mid-year probably took, I think, one to two months longer for that situation to normalize than perhaps we had hoped. That process was then largely complete by the end of Q4. It was a record year for product launches. We launched 22 products in the second half of 2024. We launched more products than we had launched in any previous full year. Highlights of that, RP2350, our second-generation microcontroller, building on the success of RP2040, which we launched back in 2021. The board-level products, Pico 2 and Pico 2 W, are built on that chip platform. We launched our first AI products in partnership with Sony and with Hailo. We completed the ramp where we launched Raspberry Pi 5 at the tail end of 2023.
2024 was the year that we completed the ramp of Raspberry Pi 5 production. We launched the 2 gigabyte low-cost variant of that platform in August. The industrial-focused derivative, Compute Module 5, and the consumer-focused derivative, Raspberry Pi 500, were launched towards the end of the year. We also saw our first software product. Software has always been something we give away for free with our products. 2024 was the launch of Raspberry Pi Connect, our remote access solution, which is the first time that we've actually gone out and charged people money for software. In terms of strategic progress, I saw very strong uptake of Raspberry Pi 5, doing 1.9 million units of that product in the year. A substantial increase in accessory gross profit was driven by those product launches in the second half. There was a return to growth of our approved reseller base.
We access our market primarily through a network of over 100 Approved Resellers in 75 countries around the world. We paused the growth of that reseller base during the shortages, but with the return to availability, we were able to return to growth there. We have made some substantial progress in our direct-to-OEM strategy, really driven by the increased visibility of the organization in the aftermath of the IPO. With that, I'll turn over to Richard for financial highlights.
Thank you. Great. Let's see if we get this looking. Great. OK, I just want to take a moment, as I alluded to, a bit of a history lesson on unit sales. The graph there looks a little bit like a mountain range. That's actually our unit sales since 2019, since launch of Pi 4. Volumes moving nicely. We hit the pandemic. We hit the shortages. There's just that first sort of dip there. Then towards the back of 2023, you see that recovery as we start to fill orders. We have the supply that we wanted from Broadcom. We have a great closeout to the second half of 2023. Into really quarter one of 2024, it was pretty good. I think that general overstocking theme that we've talked about comes into effect.
I think a lot of people, we had kept telling them not to stuff themselves, to not gorge, but they did. I think in the background of the industry as a whole, destocking, clearly that affected our volumes. Towards the back of the year, as you can see there, it starts to pick up again. Partly, I think, a bit of firming up, as Eben's alluded to, also products like Pico and Zero, which are clearly not as high margin, also being launched and starting to pick up significant volumes. Towards the end of the year, we released further variants of Raspberry Pi 5, and Compute Module 5 came out, quite small volumes towards the end of the year. What did that mean for us overall?
We ended up with 7 million boards, down a bit on the year before, certainly down on where we sort of hoped to be at the beginning of the year, very much with that sort of destocking impact. Profit per board, $7.4, that's pretty much in line with where we expected to be at this time. It is only the profit per board, so single-board computers and Compute Modules, down from the $8.6 in 2023, which was very much a number that had the benefits of, again, that whole sort of restocking. Also, in 2024, we sold more of the lower margin units in Pico. Also, Pi 5, we had this cost for the first 2 million units of $5 on that first tranche. Gross profit per unit, $7.4. We'll talk a little bit more about it later.
That led overall to a gross profit, which was approximately flat. Yeah, the profit from boards is down, we see in the announcement. We had a really good result on accessory sales. They stepped up quite significantly through the period. Particularly the launches that Eben talked to, we launched 23 products towards the back. A real step up there, and accessories improved. We also had an improvement in microcontrollers and the volume that we sold there, which allowed us to release a provision that we had as well. There were some benefits as well. Overall, gross margin percentage, pretty steady. Driver of that, much more in the way of royalty sales. You may remember from the various previous presentations, we were nearly at 100% margin on royalty. For every dollar of royalty sale, that drops straight through.
Clearly, we sold more products through our licensee partner, Farnell, in 2024 compared to 2023. That margin percentage held pretty steady. As we keep warning people, our revenue is, it's not just apples and pears. I think it's a whole vegetable stand. Therefore, the number we really do focus on is that gross profit rather than revenue. I appreciate that's a bridge. Overall, that got us to an adjusted EBITDA of $37.2 million, approximately where we wanted to be, sort of as expected. Our adjusted operating profit was $26.5 million. You can see the EBITDA and then also a higher level of amortization. We'll talk about it a bit more. Clearly, with the launch of Raspberry Pi 5, which had some significant CapEx in it, we started to amortize that over four years for the board and six for the chip within it.
Therefore, the amortization charge went up. We finished the year with $45.8 million of cash. Just looking in a bit more detail at annual units and comparisons to the previous year. Red is the units that were sold through Farnell, our licensee partner. Blue is the ones we sold directly ourselves and have made. You can see in 2023, we had an exceptionally strong second half of 2023. We had 6.1 million units within that year, dropping to about 4.9. At the same time, Farnell then stepped up very much, as you'd expect, which is why we like having a partner like that. This has enabled us to launch new product with all of the inventory and balance sheet strain. You can see that clearly, that flow through of Pi 5 came through. We had product launches towards the end of the year.
Looking at the halves, just to sort of reiterate that sort of picture, you can see during 2023, really the driver was our own direct sales. Coming into 2024, Farnell did exactly what we wanted of them, what they promised to do, what we like them to work with us on, is they helped us manage that new product bulge as Pi 5 flowed through in the first half. You can see the step up there. As I said, during the year, I mean, very much quarter one of 2024, good volumes still from people on back order. Quarter two, quarter three, very much parts of the year where that overstock through the channel in the industry, we saw the effects of that volume slowed. Quarter four, it really started to step up again.
Partly, I think that returned to lower levels, more normal levels of inventory in the channels, and also some new product launches. Gross profit per unit, as I say, $8.6 came down to about $7.4. That's an overall decline of about 15%. Processor chips from Broadcom being more expensive than in the Pi 4 was part of that, including this $5 on the first. Also, yeah, quite a bit more of Pi Zeros and Picos where our margin is $2, $3 rather than a number well north of 10 for the bigger boards. As you can see, the other piece that I think was great to see and very much linked to that product release schedule is the margin per board.
This is the profit that we make from accessories divided by the number of SBCs and Compute Modules, which is probably the best metric that we have, grew significantly. That's probably my wife wondering whether I think I'm doing. Growing from $0.60 per board to $1.20. We talk about a number of about a dollar, which is a sort of target level. We're a bit stronger than that. It's been very good to see that pick up since Eben got back into the office and started pushing these products down the slipway is probably the best way to describe it. Getting back to normal. Overall, what did that mean in terms of P&L? Revenue down very slightly. Gross profit alluded to just down a bit on year- on- year, margin percentage. R&D costs, so $26 million this year just closed, $22 million previously.
If you kind of look at it in halves, first half of 2023 was probably about 9 million-ish. That rose to, this is the aggregate of the two, to between 12 million-13 million for those remaining three halves. Half two of 2023, half one of 2024, half two of 2024. We have kept things tight. I mean, as soon as we could see that the general market was tough, we have sort of held back on any, we have been cautious on expenditure, it is fair to say, which has therefore kept us to about 26 million for the year. You can just adjust the EBITDA. The depreciation, which I alluded to, that rose to 10.7 million. Predominantly the Raspberry Pi 5, but also the microcontroller we launched in the summer of 2024. The amortization of that.
Our total spend on Pi 5 was about $20 million or so, including the RP1 chip that's included in there. We started spending on that, developing that in about 2016, 2017. It has been a burn for quite some time. The nice thing is we're pretty close. We're certainly closing in on gross profit from that product, already achieving that $20 million level. The profitability, the return that we have, I guess, payback of 18 months to two years, I think is quite something for a product of that complexity. Adjusted earnings per share and calculation. Our tax rate, the effective rate, 28.2%. That is partly because our profit for the year, that sort of profit before tax level has quite a lot of other stuff in it, like IPO costs of $2.9 million, which were not deductible. Are you listening, Rachel Reeves?
Other costs, we started a share-based remuneration for employees. There is that cost within that base number as opposed to the adjusted number. There are some other sort of effects within there. I think at an underlying level, once you start to strip out those non sort of the effects like the IPO, and also we're at the moment focusing on excluding the share-based payments, it's probably very much around that 25%. We'd see a similar number into the future. Cash level is definitely lower. We have research and development expenditure credits. It was about $3.4 million in the year. The amount we actually take to income was only $0.8 million because essentially they are grants to help us with CapEx. We release them as the CapEx comes on stream, and we actually start amortizing that, creating a splendid spreadsheet.
Overall, what does that mean in terms of our balance sheet? Intangibles, they rose to $73.2 million from $58 million. We continue to invest significantly in further CapEx development, particularly new boards, new semiconductor products. Our inventory at the end of the year was $156.7 million. That compares to $108 million at the end of 2023 and $146 million at the half year. Really since the half year, it's increased slightly, but it's pretty steady. The mix within there, I think, has been particularly satisfying, though. Finished boards, we finished the year with $63.8 million of finished boards and other finished products. I think when we talked to people around about the half year, we said that we had seen we were probably overstocked on things like Raspberry Pi 3 and Raspberry Pi 4 because the speed at which demand dropped away had caught us with some production coming through.
I think that resilience of our business model, that close contacts we have with our partners was very powerful. We're now down to a situation where we scarcely have enough. Pi 4s, we're back into quite significant production. We've almost got down to nothing in the warehouse at various moments. The overall, I think the slow-moving aspect of inventory has gone. What we're seeing here is really Pi 5, which, bear in mind, was only launched in sort of October 2023. Therefore, had very little there at the end of 2023. Compute Module, particularly the 5, but also the 4, which was in back order. Both of those now, we have probably what we feel is about the right level of stock. That's really the cause of the growth there.
In the other area and components, memory is probably a similar level to that of the year earlier, so December 2023, somewhere of the order of 30-35 million. That allows us to be pretty confident about the cost of our products through into quarter three, probably towards the end of the year in terms of the impact of memory. The other part that's grown is we now have more of the critical application processor chips that we get from Broadcom, which has given us that resilience. We do not want to go back through that roller coaster graph. You know, you like roller coasters. I have a terrible fear of them. Of the previous years when we just could not get enough chips from Broadcom. Quite deliberately, we got away with it once is what we feel.
Our partners, our customers trusted us then. We can't do it to them again. Receivables down a little bit, as you saw from the charts, very much a bumper end to 2023. Payables up slightly. We said during the year, there was an opportunity where we could take on components and things that were really quite favorable terms. We have continued to benefit from that. There is some step up in payables because of that. I think we see that as probably starting to unwind over 2025 with maybe inventory levels, similar-ish levels, maybe come down a little bit, probably similar levels to what we have here. Cash overall, $45.8 million. Generally, that is the balance sheet. Tell people once, tell them again. Cash flow, really just illustrating many of the sort of facets we have talked about already. We started the year with $42 million.
EBITDA contributed 37. Spend on memory and components, that increased by $25 million over the year. Finished good, an increase of $23 million from the beginning of the year. Very much that we now have the stock that was zero at the end of 2023 in respect of Pi 5s and Compute Modules. Other working capital mentioned the benefit of the better payment terms, the terms we've been offered. CapEx, $23 million overall. About $20 million of that is intangibles, product development. The rest is sort of the usual sort of things that fit out and warehouse work. Proceeds from the IPO and then a number of other bits and pieces bringing us to about $45.8 million at the end of the year.
Finally, one of the joys of being back in the office and having some time is I've been actually able to do some things with all the data that we have. One of the charts that came out of this was a graph that picks up across the bottom average selling price for our products. The blobs are actually individual products. The gross margin % we make on the Y axis. And the size of the dots is actually the gross profit we've made from those. There's a couple of key things I just wanted to leave you with before handing back to Eben. You can see there's essentially a trend. As the average selling price goes across, that margin % increases. I think we've always said that as the selling price increases, obviously the gross profit almost automatically increases.
Actually, one of the nice features of this is the margin percentage is also big. The profit per unit is rising as the selling price increases. As we put more memory, as we put more performance, as we put extra features like extended temperature ranges onto products, we do see additional profit per board, which is very much that strategic aim of growing unit profit. What we also see, I guess, is in places where we have moved either to selling more direct rather than through royalty channels, or we sell direct to OEM. Again, those bubbles start to rise slightly because gross profit participation improves. Finally, in the sort of top corner there, Raspberry Pi Pico, our first board that actually contains our own silicon. You can see the margin percentage is there significantly higher than the rest, really quite different.
Yes, the selling price is quite low. The dollar value of those is lower, but the margin percentage is great. That is very much the benefit of our own silicon being there. We have much more control of the product. It gives us great competitive moat. We can really optimize that price structure. We like to think that is kind of the model as we go forward. As we are making investments in silicon, it gets us into that space on other boards over time. Surprisingly, it is quite a large bubble, actually, which is the gross profit we have generated. Please do not get out your rulers and set squares and calipers or whatever and try and interpolate too much from this at that level. As a general trend, it really does summarize what has been happening intrinsically within the business.
As the ASP increases, as we sell more, our margins and our profit per units improve. If we put our own silicon in, it puts us into a very good position.
Fantastic. Thank you very much. The astrophysicist in the room, I did point out he had drawn a Hertzsprung- Russell diagram there, but he did not accede to my request to turn the circles into little stars. Good. Right. A little bit more, yeah, no, you're right. A little bit more update on some of those 2024 highlights and how they align with our strategic objectives. As I say, 2024, second half of 2024 had more product launches than any prior full year. There was an obvious, as Richard says, there was an obvious positive impact on our in-year profitability.
Some appeal to our enthusiast base, really reassuring our enthusiast base that as a listed company, we were going to continue to build the interesting products that get them excited about Raspberry Pi. Some of our products over here on the right, the Raspberry Pi 500 and the Raspberry Pi official monitor that we launched at the same time, obviously are explicitly targeted onto that market. Really, the overarching story of 2024 is about platform refresh. Pico 2, which we launched in August, powered by RP2350, refreshes our low-end platform, refreshes our microcontroller-based platform. Raspberry Pi 5, which we launched at the end of 2023 and completed the ramp of in the first half of last year, and its derivative products, in particular, Compute Module 5, mean that over the last 18 months, we have refreshed the entire Raspberry Pi technology roadmap.
Both of those, both the new microcontroller products and the new big products, the new Linux-based products, they both expand the range of addressable opportunities in our OEM customer base. Lastly, down the bottom, Raspberry Pi Connect, as I said earlier, for us, software has always been a cost center, not a profit center. It has always been something that runs on the device. We build the hardware, and we build the low-level software that runs on the device. I think what Raspberry Pi Connect has done for us is to broaden our notion of what we mean when we say compute platform to encompass some software that also runs in the cloud. Right now, Raspberry Pi Connect is a remote access solution.
It allows our enthusiasts and our OEM customers to get access to a Raspberry Pi after that Raspberry Pi has been deployed into the field on its own or as part of a customer product. As I say, it's the first time we've charged for something that isn't hardware. There is a free tier, and we launched in May last year with a free tier targeted at the enthusiast customer base. In December, we launched Raspberry Pi Connect for organizations, which was a paid tier of the same product, adding some of the features that we know our OEM customers need. We are exiting beta. This has been in a public beta, very much that kind of Google style of launching products and holding them in beta for a long time.
This product is now exiting beta as we get to the point where we're confident that we have the set of features that our OEM customers need. Really, Connect will become, while it is now purely a remote access solution, Connect will become the nucleus of all of our off-device work as we expand the off-device offering for OEM customers to include features like over-the-air updates. That is products last year. What does the product roadmap look like in 2025? We were always relatively closed about where we're going with the products. I think there are some things that we can say about the sort of general direction. Obviously, the ramp to volume of Compute Module 5, Compute Module 5 launched in December of last year. What's the benchmark for that ramp?
The ramp of Compute Module 4 after its launch at the tail end of 2020. CM4 sold about 230,000 units in its first year on sale, about almost exactly 1 million units cumulative over its first two years. Now, that's not a totally clean data set because it's a pandemic-era data set that overlaps both sort of pandemic-era demand and also the beginnings of pandemic-era shortage. I think it's a good benchmark for Compute Module 5. The ramp of Compute Module 5, I think, is underpinned though by two things that we didn't have on our side when we launched Compute Module 4. First of all, this is a product which has enormously more compatibility with its predecessor than Compute Module 4 did with Compute Module 3.
In fact, about roughly a year before we launched Compute Module 5, we published a forward guidance document to our OEM customers to help them understand how to design Compute Module 4-based products, into which they could drop Compute Module 5 without disruption. We have that compatibility win. We also ran a much more extensive early access program. This is really, I think what you're measuring there is the larger size of the commercial team, the larger commercial capacity within the organization that we were able to run an early access program to give some tens of our best OEMs early access to Compute Module 5. That's why we were able to announce, unlike Compute Module 4, with Compute Module 5, we were able to announce OEM wins, OEM product launches which were concurrent with the launch of the Compute Module 5 platform.
Secondly, we have range expansion for Pico 2. With the Pico 1, in the Pico 1 generation, we really just shipped two Pico 1. We shipped a non-connected and a local area wireless and Bluetooth connected Pico 1 product, Pico W. We've obviously done that for Pico 2 in August and November. We're going to continue, I think. I think we're perceiving some opportunities throughout new connectivity options for the Pico 2 platforms, both wired and wireless. Some of those, for our OEM customers, the Pico products, I think, serve as development platforms, reference designs and development platforms for our silicon. I think some of the connectivity options that we're going to add to the Pico 2 platform give us an opportunity to sell complete Pico 2 products into some of our OEM customers. Next, AI products. Obviously, last year we launched our first AI products.
They've been a success for us. They have some limitations. In particular, I think they are based on silicon, which was conceived in the era when AI meant recognizing features in images. That's not really what AI means anymore. They have very limited generative AI capabilities in particular. I think it's likely you'll see us align with our vendor roadmaps. We have some fantastic vendor relationships in this space, and you'll see us align with our vendor roadmaps to launch new generations of AI products. Finally, RP2350. As I say, been a success for us already. Put it into distribution last month. I think we put it into general availability into distribution a little earlier than we were expecting. Really, the reason for that is it was becoming we were supporting early access customers in a kind of white glove hand-holding way.
We just burnt through the entire capacity of the commercial team. The commercial team could no longer cope with the amount of interest, the amount of demand for that product. Put it into general availability into distribution a little earlier than we were expecting. Again, the ramp to 2350, the rather accelerated ramp to 2350, I think is supported again by compatibility with the predecessor. There is a lot of compatibility between 2350 and 2040. Again, by that early access program that we have been able to run. Excuse me a second. There we go. Growth in the reseller channel again. Super exciting. I mean, these are some incredible companies ranging from multi-billion dollar global distributors to individual one and two employee mom-and-pop shops. They are the primary way in which we meet our customers. They are the primary way in which we sell.
They are the primary way in which we learn about our customers' needs so we can be responsive to those. Adding some really kind of two things going on here. Adding some new territories: Colombia, Argentina, Qatar, Ukraine. Super exciting. There are still worlds there. Even today, there are still worlds to conquer with Raspberry Pi readily available in over 75 countries. There are places where you still can't easily buy Raspberry Pi, and we're looking to address that. You see some existing territories here. Good examples, France, Poland, Italy. These are either territories in which we are generally under-penetrated. We do have benchmarks for how many reseller partners and, for example, what total credit capacity of partner we want to see in a territory relative to its population and the size of its economy.
There are territories in which we're under-penetrated, and France is a really good example of that. Or where we lack specific sorts of local representation. Poland and Italy are examples of places where we're adding those mid-tier industrial-focused distributors, which we've had for a long time, say, in the U.K. or in some of our other core markets like the U.K. and Germany. Busy slide. We made some substantial investments in headcount in the commercial team in the latter part of 2024 and then going in 2023 and going through into 2024. That's really given us the capacity to engage more closely with OEMs in support of our aspiration to do more direct to OEM business. We have a much better developed market segmentation.
These are the 15 market segments that we use as the kind of reference, the kind of framing for the efforts of the commercial team. By getting closer to our OEMs, we are coming to understand their priorities a little bit better. What do our—I think it's an extremely non-exhaustive list of OEM priorities. What do OEMs care about? They care about time to market, and in particular, having that seamless transition from prototyping, which is often already on Raspberry Pi, into production. They're interested in access to advanced technology. It's interesting that the scale at which many are—still fairly substantial scale—at which many of our OEM customers operate, they are simply not able to get access to the most advanced silicon technology if they choose to follow a make rather than buy strategy.
They are looking for reductions both in fixed costs, not having to maintain the engineering team that develops and maintains an in-house compute platform. That includes compliance and global market access expense and also variable costs. They are looking to reduce the amount of the impact of the compute subsystem on the bill of materials of their products. What is our value proposition to these OEMs? We are making products which are reliable, which are available and supported. When we say available and supported, we do not just mean available and supported today. We mean available and supported in 10 years' time. I had a little look at the numbers we just got on our March flash numbers. I do not think it is material to say that we sold 451 Raspberry Pi 1 in March. Right?
This is a product that we launched in July of 2014 and nominally obsoleted in February of the following year. We now have that track record that if you decide to outsource the compute element of your product to Raspberry Pi, we can give you confidence that in 10 years' time, you will be able to buy that product from us. You will be able to run the software that you designed today on that product, and you will even be able to run the software that we are shipping in 10 years' time on that product. We are offering availability and support. We are offering what we believe is a very differentiated price performance proposition underpinned by the structural cost advantages of our technology.
A streamlined development flow that you can start with a—you can start with a Raspberry Pi single board computer, migrate to a Raspberry Pi module, and then at the highest volume levels, even migrate to a customized Raspberry Pi product. Documentation, design support programs, both inside the organization, our own application engineering capabilities, and the Approved Design Partner program, which gives people access to hourly rate engineering support to design Raspberry Pi into your products. In U.K. manufacturing, we believe that the U.K. is the best place in the world to build products like Raspberry Pi. This is a drum we've been banging for a while. It is obviously a story which is becoming increasingly salient, becoming increasingly salient over time in the context of this slightly complicated geopolitical environment that we live in. Ian talked kind of summarizing all of those OEM interactions.
Ian here has developed the kind of the idea of this idea of an R&D air pocket, that there are many OEMs who wish to add, often for the first time, wish to add significant compute capabilities into their products, but do not wish to build, do not wish to take on the cost, do not wish to take on the complexity of building an in-house team that can build the compute element of the platform. Really, those, I think that is quite a good summary of kind of the heart of our OEM business. Post-IPO, we have been able to run a round of C-suite level engagement at U.K. industrial OEMs, really aiming to do three things, I think. To detect and address those kind of residual misperceptions.
The success, the historical success of Raspberry Pi in the enthusiast and education community continues to generate, continues to sustain misperceptions about what Raspberry Pi is and what our technology can do. Going out there and finding those misperceptions and correcting them, providing people in those organizations with information about the path to production, the information that there is that route from SBC to module to custom that will sustain an OEM as they use Raspberry Pi technology, and finally building trust. Easier, obviously, with that track record of keeping products in production for an extended period, and obviously, interestingly, easier as a public company than it was as a private company. Really, often the great thing about those engagements is you're going at the top of organizations and you're not forcing your product down those organizations. You're going into those organizations.
As the people in the C-suite look down the organization, they discover an engineering team who are already very excited about Raspberry Pi. Other engineering teams, often they're already using Raspberry Pi, at least at the production stage. Really, very often what you're doing is you're giving those teams permission to scale with Raspberry Pi. Just a few brief examples of recent OEM activity. Giatec over on the left-hand side. We love these guys. We love these guys largely because they embed our products in concrete. They're a wonderful opportunity for repeat business. They don't tend to use a given product twice. Using Raspberry Pi in their distributed concrete delivery and curing monitoring system. KUNBUS, one of our oldest partners based in Germany, taking Raspberry Pi technology and packaging it as an industrial computer.
They've been building products based on Compute Modules since Compute Module 1 in 2016. Obviously, they have first-hand experience of what that long-term support guarantee means to their business. The last one, really, I guess a standout moment of 2024, the HMI, human-machine interface collaboration with SECO, based in Italy. Wonderful opportunity to collaborate with another absolutely first-rate player in the computer hardware and software field. Really, a situation where the whole bringing together their IoT platform and our compute hardware really does generate something where the whole is worth a lot more than the sum of the parts. Each of these partners, we've sort of had a little bit of an idea of what we think each of these partners values most in our value proposition. Really, what is the common thread here? The common thread across all of these companies is software.
For a hardware company, it's fascinating the extent to which our OEM customers see our differentiating value to a very great degree in being in the software that we run on our hardware. Quick thought about what the limits to growth might look like. I'm going to go out into any public space today, and you'll already see an incredible amount of distributed computing around you. Even though the cost of doing a lot of that compute is really, really very substantial, there are many, many, many pieces of compute in your environment that have positive return on investment. Lowering the cost of computing obviously creates new positive ROI opportunities to compute. I believe that over the next decade, we're going to drive a step change in the cost of computing.
I think we're then very, very well positioned to claim much of the resulting increase in compute volume. We've shipped very nearly 70 million units, but I really do think we've only really started down this road. Finally, a few quick words on 2025 outlook. Channel inventory has now normalized. By the end of last year, we got to a position where the overstock in our reseller and distributor channel had normalized. We do believe there are still some limited inventory pools at OEMs. What we expect to see this year is a steady build in unit demand as those pools deplete and as we start to see existing OEMs come back for another bite. New post-shortage design wins go to scale.
The typical time required for a design win to mature into volume is 18 months-24 months, which is now 18 months-24 months since Raspberry Pi products started to come back into good general availability after the pandemic. That is likely to be the source of increase in demand this year. We are projecting modest increase in the economics. As Richard said, we have some improved logic pricing, which I think probably in the interims we suggested was going to meet elevated memory pricing largely canceled out. I think where we are now is we are expecting that improved logic pricing to meet largely flat memory pricing and give us a very modest improvement in easiest economics. Confidence on a long-term outlook is underpinned by some of those direct OEM engagements and by the expanded addressable market due to that platform refresh across both our Linux machines and our microcontrollers.
Thank you very much.