#38 Raspberry Pi Founder: People Overestimate What AI Can Do

19:55And then I got back on the plane, flipped around. Why? And it wasn't because we didn't meet really interesting, capable people there. But I think the big message that we brought back with us was, first of all, this idea that you will get a better valuation. That's the core of this.

20:13Felicity Hannah:That's what we hear over and over again. That's the core of it, that you get a better multiple. You take whatever number you like, EBITDA, and then you apply something to get your market cap. and that that multiplier will be larger in New York than it is in London. Guys at the Stock Exchange here will tell you is that that's a cohort effect. It's a data artifact. That when you say that, what you're saying is the sorts of companies which are listed in New York are the sorts of companies that command higher multiples. If you stratified, if you compare apples to apples, find two identical, otherwise identical companies that are listed in the two markets, you don't really see any difference on a company-by-company basis.

20:49and what we substantiated when we went out to New York was first of all that that really is true and almost we found people, there are a small number of true believers. I have met the occasional, I've met the occasional person, I met somebody this year actually on a roadshow earlier this year in Chicago who said I would invest in your company if you had a serious plan to relist in New York because I would expect to obtain a quick profit out of your doing that. But by and large, what most people would say is, well, that's impossible, right? How could – if I could buy a return, it's just – it doesn't make financial – and that's what I'd always thought, right?

21:25It doesn't make financial sense. If capital is globally mobile and if that capital can go and find a better return over there, it will go over there and it will bid up the assets over there until it equalizes out any – you have hedge funds that exist to – hedge funds run on these tiny little – I work by identifying and running on these tiny little gradients. and if there was this huge gradient that you could persistently buy return more cheaply in London well it's hard to imagine how that can exist and it was almost the case with one or two people who I described this to when you asked them the question what about the multiples gap they look at you as almost like you're destroying your credibility with them by believing in a fair by believing in a fairy tale so I don't think there's actually anywhere near as much belief it's again this thing about looking out and looking in I don't think there's anywhere near as much belief in a multiples gap looking from that direction as there is looking from this direction.

22:16So the key rationale went away. And then you're talking to people and they're saying, well, look, we will find you. Don't worry, we'll find you. If you are, A, we'll find you. B, you have to realize that you would be a small company. If you're going to list and, you know, the market cap this morning is like 1.3 billion pounds this morning, you would be a very small company in New York. And you risk, where you're a decent size, respectable FTSE 250 company here, that you would risk falling into the noise floor, failing to get investor attention, failing to get analyst attention. And in practice, that's what put me back on the plane, thinking, OK, let's go, let's do this in London.

22:53Felicity Hannah:I enjoy that you're not looking at the valuation every morning. I think I would be, first thing I'd do when I picked up my phone. But it's interesting, isn't it? Because you were initially valued at£542 million. Today, as you say, you're over£1.3 billion. What's behind such enormous growth in such a short period? I think, well, I hope, I mean, you can never tell, right? But I hope that there's an element here of us starting to get a reputation for doing what we say we'll do. You know, the thing I'm proudest of is there was pre-IPO analyst consensus for what our numbers would look like in 2024 and 2025.

23:27And our results were within a million dollars, hit those numbers, right? So we are very focused on making sure that we don't disappoint people, that we deliver the growth we say we'll deliver, that we don't disappoint people, that we don't let our messaging get ahead of the performance of the business. You could feel a substantial change in sentiment when we did our results at the end of March.

23:51Felicity Hannah:That's interesting, isn't it? What happened? What changed in March? Because 150 % growth? So I think some of that is about, as I say, having done what we said we'd do. Some of it was about currently facing an interesting situation around the supply of memory, the supply of DRAM, which is probably the largest single input factor for the business, which is suffering from both price increases and availability constraints. And I think we were able to say, again, you know, quite careful in what we say about this, because we don't want to commit that the worst thing in the world would be to say, don't worry, we've got DRAM covered, and in a six month's time, have some sort of DRAM problem.

24:27But I think we were able to say things at the results, which gave people a little comfort that we were, that we, in the first three months of the year, we'd moved in a good direction on that. And then on top of that, I think there is a growing appreciation. Obviously, there's enormous enthusiasm for anything AI adjacent in this world at the moment. And I think that we were able to articulate why we think... There had been a strange share price excursion actually about a month earlier, which was associated with these kind of open-claw, agentic AI platforms running on Raspberry Pi. And I think we were able to explain to people why, although the short-term excitement about that was people thinking, I'm sure people will run out and buy a million Raspberry Pis in order to...

25:05Felicity Hannah:That's when they realized they could run it on a Raspberry Pi. That's it. And you had people running these things on Apple Macs, on the other fruit company, the other fruit company's hardware. And, of course, you don't actually need that sort of hardware. The sort of mid-range Raspberry Pis are as good at$100, or as good a platform for running OpenClaw as a$600,$700 Mac. So I think some people felt that there was going to be a sudden surge in enthusiasm. What I think people forgot there was that everybody who's interested, Raspberry Pi is completely ubiquitous in the engineering enthusiast community.

25:36Although we're now really an industrial company, it is still completely ubiquitous in that original market. And so anybody who's interested in running OpenClaw already owns a Raspberry Pi, and they can take the next one off the shelf and go and use it. So I think people imagine there was some short-term story, but I think one of the things we were able to do successfully at the time of the results was to explain to people why that was a useful leading indicator for the ways in which we think our industrial and embedded customers will be using Raspberry Pi in the next five or six years.

26:08Felicity Hannah:What are your five-year aims for growth then when you've seen such an extraordinary short period? When we talk about growth, we mean the growth of the business rather than the growth of the value of the business. But I think we've got to a point now where we've demonstrated 70 or 80 percent of raspberry piers today sell into what we call industrial and embedded applications and i think we've demonstrated now that it is if you are building a you know an elevator car or an electric car charger or a interactive dartboard big business for us interactive dartboards post luke littler if you're building any of these devices if you build a device and you want to make a version of that device that has intelligence and connectivity in it then rather than building you shouldn't have to be if you want a company that makes you know dartboards.

26:50You shouldn't have to become a computer company. You shouldn't have to grow inside your organization the capacity to build computers. You should be able to outsource that to somebody else. And I think we've demonstrated that. And I think we've demonstrated that if you are going to outsource it, we're the place to outsource it to us. So I think we have a great opportunity to double down in that bit of the business. And at the same time, we've been building our own silicon. An interesting journey Rastry Pi has been on over the last five years is increasing the amount of the silicon in the platform that's our silicon rather than silicon from other people.

27:23As we build those devices, we don't just use them to build our products. We also sell them to other people. And so that idea that probably, you know, in year five, not necessarily this year or next year, but by year five, you're hopefully going to start to see significant economic impact on the business from the sale of semiconductors.

27:41Felicity Hannah:It's interesting because you're talking about how your Raspberry Pi is now being used so much by industrial, embedded business needs. And it's a long way away from the hobbyists. But another big shift for Raspberry Pi has been your semiconductors. You're manufacturing your own chips to your design. And that's now more than half of your business, isn't it? And I want to talk to you about that because they're made in Taiwan, aren't they? And so that is a much more difficult thing to oversee. You were talking about manufacturing in Wales and production in the Midlands. And suddenly you've got this big chunk of your business happening, not here in the UK, but in Taiwan.

28:19Yeah, that's right. So it is, as you say, it is more than half of our business. Of course, it's more than half of our business by volume rather than by revenue. So these are 50 cent chips versus$50 boards. We've come through this interesting moment over the last 12 months where strategically we feel like a semiconductor company now, right? But economically we are very definitely an electronics company, a board company. We have obviously a wonderful part. We have a couple of wonderful partners in Taiwan. We have TSMC, the world's biggest and most successful foundry, Silicon Foundry. And then we work with a company called ASE and also in Taiwan to package those devices for sale, for use in our products and for sale to other people.

28:57Yeah, it is a change for the organization. We've had to grow a more operational capability. This idea of ops is probably what we're going through this year, actually. I've just hired an old friend and colleague as our first chief operating officer. And this is an organization where kind of responsibility for operations, manufacturing operations in particular, manufacturing operations and engineering operations has been diffused through the organization.

29:21Felicity Hannah:You only just hired a COO. We only just hired a COO because we all – like COO is one of these roles which is quite, as I say, quite diffused through the organization. But we found ourselves having to acquire a silicon operations function. So we have a small silicon operations team. One of the early road bumps in going to scale with semiconductors was that you need operational capability there, dedicated operational capability. And actually, as we build out the ops capability for the rest of the organization, it's the silicon operations team that is the sort of template that we find ourselves recreating in other aspects of ops.

29:58Felicity Hannah:It brings a whole new element of risk, though, doesn't it? Just in the last 24 hours, President Xi told President Trump on this state visit, the Taiwan question is the most important issue in China-US relations. That's according to the state broadcaster. And he said, if mishandled, the two nations could collide or even come into conflict, pushing the entire China-US relationship into a highly perilous situation. Do you have any worries about what might develop, what it might mean for the work that you're doing in Taiwan? I think we should all be concerned. And I think it's a – people ask us sometimes, like, what's your mitigation?

30:32And the challenge is it's not really a mitigatable problem, that there is such dependence, particularly for the – not actually so much in the sorts of chips we build because the chips that we build ourselves are on relatively mature process nodes, sort of older, larger geometries, 40 nanometer geometries.

30:51Felicity Hannah:You might need to just simplify that for our listeners. Not me, obviously, but our listeners. Yeah, so a tiny fraction of the width of a human hair rather than a tiny, tiny fraction of a width of a human hair. If you think the AI chips that NVIDIA make, the mobile phone chips that Apple make, they're on characteristically on kind of a 2 to 3 nanometer process, 2 to 3 billionths of a meter. That's the smallest structure that's on those chips. Now, the Raspberry Pi, the stuff we build ourselves, is on a 40 nanometer process. So the structures are roughly 10 times larger. I mean, they're still incredibly staggeringly tiny.

31:28It's a measure of how fast this industry has developed, has developed, continues to develop, that when the chip that's in Raspberry Pi, that's in Raspberry Pi 1, which is a 40 nanometer chip, was pretty much bleeding edge at the point where we launched that product. So over the course of 15 years, the bleeding edge has moved from 40 to 28 to 16, all the way down to 2 nanometers. Now, TSMC and critically TSMC in Taiwan, because TSMC do have factories outside Taiwan, monopolize the supply of those very, very fine geometry process nodes. And that's one of the things I think leads people to believe that it isn't strategically feasible for the United States to – President Xi is right.

32:12It isn't actually strategically feasible for the United States to abandon Taiwan because you can't afford to lose access to that technology because it's increasingly – it has been the underpinning of our civilization for pretty much my entire life. It is doubly, particularly with the rise of AI technology, it is doubly the underpinning. It's the underpinning of all of this prospective growth that people are hoping to get out of the AI revolution.

32:38Felicity Hannah:We've talked about OpenClaw, this agentic AI. Now, that's AI that can operate autonomously. Is that the next seismic shift? How is it going to shape and change our economy? I think we are in an interesting moment, actually, because we've probably... There's a school of thought that says that this AI thing is naturally exponential, right, and that it's going to run away. We had an AI safety summit a couple of years, two or three years ago now, that Rishi Sunak ran. It was at Bletchley Park. And I went to the after party where Sunak was interviewing Elon Musk. And you listen to Elon Musk on the stage and he's saying, well, look, one of two things is going to happen.

33:16This is going to escape exponentially. 80 % chance it's going to usher in an era of unparalleled abundance, a post-economic era of abundance. 20 % chance it's going to kill us all.

33:28Felicity Hannah:He calls it the Terminator outcome, doesn't he? Yeah, there's the Terminator outcome and then there's the culture outcome. It was interesting to watch the novels of E.N.M. Banks, Banks wrote sort of 20 or 30 years ago, quite a good view of what a society that's kind of enabled by kind of run by benevolent super intelligence might look like. So you have these people selling these two curves. I suspect what you actually end up with is in the middle. You get a fraction of a percent on global GDP growth for the next 20 years, which is amazing. The same sort of outcome that the Internet has provided us with.

33:58So I think there's a sort of a desire to believe in these kind of outcomes. In practice, it is normal technology. And I think what we've seen now is we've seen the wave of LLMs. We've seen the wave of sort of chatbots being helpful. I use these tools all the time. They're super helpful. I think the thing that's happened over the last six months is people have come to realize that, yeah, providing some level of autonomy and some level of iteration. Because, you know, a chatbot, you say a thing to it, and then it says something to you. And then you say something to it, and it says something to you.

34:26And you are intrinsically in the loop with that object, giving the devices, whether it's something like Claude Code, which provides you with a agentic coding platform, a sort of somewhat self-sustaining coding platform, or something like OpenClaw, which gives you your own private agentic object that you can connect to a remote LLM to do the LLM magic and then to some of your data and some of your local environment. But definitely we are in the agentic moment.

34:51Felicity Hannah:And I should say, well, when we say LLM, I'm meaning large language model, which is how they've sort of learned to talk back to us. You said earlier on that there's a sort of tendency to overestimate what these tools can do. Are you trying to mitigate that risk? I think we just, I think we all need to make sure that we use them in responsible ways, ways that actually generate value. There's a wonderful study of some of the AI-assisted coding tools that suggested that they went and they took a group of people, a control group and a tools-assisted group. And the interesting thing about the tools-assisted group is they felt they were asked to assess – they had their performance externally assessed and they were asked to assess their own performance.

35:34And they assessed themselves as being 20 % more productive and they were actually 20 % less productive. So I think it is possible to use these tools in ways which undermine your capacity as an engineer. But on the other hand, we use them every day at Raspberry Pi. The interesting thing, though, is I looked at our token spend, and you see these stories about individual engineers at Facebook blowing through hundreds of thousands of dollars of tokens a month. You have Jensen saying you should be spending about as much money on tokens as you do on engineer salaries. Last month, we spent£1 ,700 on tokens at Raspberry Pi, and I certainly spent more than£1 ,700 on engineer salaries.

36:10They are useful tools, but you have to have discretion and you have to have taste in how you use them. And this is back to the question about, can we stop teaching kids to code now? Absolutely, we can't.

36:20Felicity Hannah:We should say token in AI, a sort of unit of data, isn't it? Like a word? How much AI you're doing. How much AI you're devouring. Some number which is broadly proportional to the amount of carbon that you're pumping into the atmosphere. Which is another very, very important aspect of this. Another important aspect is what it means for the students that you began your career with. I mean, I was thinking in particular about A-level students right now sitting their exams, wondering what they should study next or where they should go next. What will future-proof them against this AI world that Raspberry Pi is helping to usher in?

36:57Oh, I think people should just do what they were doing before. You know, it's possible to get caught up in this. This is this risk of damage, right? in this moment of incredible enthusiasm for what are genuinely incredible tools, the idea that we should suddenly, you read in the paper, you know, what guidance should you give your child about what GCSEs to choose in the context of an AI future? And it's like, we have no data to inform a rational decision on that. The answer is wait five years, wait 10 years, and then maybe we might know something. But it's, the advice I always give everyone, because I come from quite a quanti background, is do as much maths as you can.

37:38The only thing that people have a misconception about quantitative subjects, about science subjects, that they are narrowing somehow, that you're inevitably on a path to wearing a lab coat, and they are broadening. If you don't have aptitude for it, don't do it. But if you have aptitude for it, however much aptitude you have, and the big lesson, my lesson from arriving at Cambridge, is everybody's got a ceiling, right? Somewhere I found mine. But however much ceiling you have on the quantitative side, you'll probably be well served as a person if you try to get close to that ceiling.

38:16Felicity Hannah:Sorry, I'm going to play this to my would-be engineer 13-year-old later. Also, engineering is, I should say this, engineering has such an image problem. A sort of Dilbert, I'm sorry, ma 'am, he's going to be an engineer. Yes, I was really terribly sorry. And I always say this to parents, like, why wouldn't you want your kid to be an engineer? Because they pay you money to mess about. And it's indoors in the warm. It doesn't have to be indoors in the warm. If you like to be out in the cold, you know, we have a whole North Sea you can go do engineering in, right? But, you know, this is a it is just the most incredible job and it just serves you up in my job.

38:58I don't do it. I do much less engineering than I probably would naturally do these days. But it's just it's a never ending puzzle box of exciting and well remunerated fun your entire life. So, yeah, everyone should be an engineer.

39:12Felicity Hannah:We've talked about the pressures of the energy costs on a business like yours, and that's not been helped by what we're seeing in Iran. What we're also seeing in Iran is it's affecting helium supply and helium we are now discovering is for more than balloons and it feeds into tech in all sorts of important ways when it comes to manufacturing semiconductors. Are you seeing any kind of problems there? We're not seeing any problems. We have not had our direct silicon suppliers, so TSMC, or our indirect suppliers, so our DRAM vendors, our other logic vendors. We haven't had them communicate anything to us.

39:46I think the challenge is you don't know. We really don't know what the consumption intensity of helium inside or other chemicals inside the semiconductor process is. We don't know how big the semiconductor vendors' stockpiles are. We don't know, although we can guess, how capable those companies are of bidding up for supply. I mean, I suspect that the party balloon industry probably falls over before the semiconductor industry. You don't. Yeah, probably its reservation price is probably lower than your average two nanometer fab. But we just don't know. I don't think we are. I think it's possibly characteristic of the age that when something like the Iran crisis happens, that the natural tendency is to cast around for some way in which this is the apocalypse.

40:33But we'll see, again, slow thinking. I think we've just got to give it a few months and see what happens.

40:38Felicity Hannah:I was thinking, I was wondering what kind of change it's been for you as a person to go from academic to entrepreneur to the CEO of FTSE 250 firm, sitting on earnings calls, going to Downing Street, taking photos of Larry the Cat. It's perhaps not where you expected to be. it's where I always wanted to be I am I am absolutely where I wanted to be some really sad story there's a there's a television program called Bergerac from the 1980s and John Nettles I think and there is a character in it called Charlie Hungerford who is a who's a slightly disreputable entrepreneur of dubious legality and he is and as a child I was always fascinated with him I just remember watching Bergerac and being fascinated with this character who was always It's sort of trying to find new ways to do business.

41:27I hope to be more legitimate than Charlie Hunker. But it's something I've always been interested in. I've always been – I've started my first company when I was 20 years old. I started a games company in London, which is still running actually, Marmalade, called Marmalade Game Studio. And so I've always been as interested. I love technology. I love engineering. But I've always been as interested in the business of engineering, the business of technology as I have in technology. So I am where I want to be. It's enormously good fun. you read these articles about how stressed public company CEOs are and how hard the job is and everything is.

41:59Well, you know, you don't have to do it. You can stop. I think it's amazing. I love the investor interactions. You are always, you know, you may be meeting a specialist who knows your field very well, or you may be meeting a generalist, but you're not meeting, you're never meeting stupid people, right?

42:15Felicity Hannah:What's your advice for someone who's at the startup stage now who's working away long hours somewhere because they believe in an idea oh persist find ways to persist some people have it inside themselves to endure and persist i don't actually and and you know raspberry pops through in the the the time between founding the foundation in 2008 and launching the first product in 2012 i i got bored and wandered off several several times and went and tried to do other things and i was lucky that i had my family and particularly my wife who went on to to be my co-founder, steered me back onto it every time.

42:51I'd talk to my parents and say, I've got this great new idea. And they'd say, what about that raspberry pie thing, though? You should go back and do that. And so finding ways, whatever your own capacity. And now, you know, I'm surrounded. I have about 100. We're a very small company, actually. I'm about 150 people. But they are absolutely incredible people. You are surrounded just to do the engineering function, finance, commercial, communications. They're all really first-rate people. And their presence and their enthusiasm. And their hard work helps me to endure. So find a way to do that inside yourself or in the people you have around you.

43:25Felicity Hannah:That was a beautiful point to end. But it does just pose another important question. Is there a skills shortage in the UK? Is that a problem? Is it hard to surround yourself with 150 brilliant people? I think it's incredibly difficult to find 150 people as good as my 150 people. But we do. I think we do have a skills shortage. We have a quantitative skills shortage. Back to the point that people don't realize what a good job engineering is and people don't understand how important it is that they try to push whatever capacity they have for maths as far as they can. We're in a moment where the conversation about AI has the potential to distort people's choices in ways that make that skill shortage worse and not better.

44:11but I am optimistic that a lot of the infrastructure we've built, not least the Raspberry Pi Foundation, right? That a lot of the... This is your charity. This is a charity. That's a huge amount. Yeah, and this is the original thing that we founded and that still owns the best part of half of my business. That infrastructure we've built will provide the kind of pushback in the short to medium term against some of those narratives and help people understand that whatever the world looks like in 2030, 2035, 2040, it is going to need more people with quantitative skills and more people who are able to make good engineering judgments in order to use whatever the great, great, great grandchild of today's AI tools turns out to be.

44:49Felicity Hannah:Do more maths. Do more maths. Dr Eben Upton, thank you very, very much for joining us on Big Boss Interview. Thank you.

44:59Felicity Hannah:Our thanks to Dr Eben Upton, the founder and chief executive of Raspberry Pi. They're speaking to Felicity Hanna. And thanks so much to you for listening to the Big Boss Interview podcast. If you don't already, please do subscribe and give us a five-star rating while you're there. It really does help other people find the podcast. And remember, you can get in touch with us. The email is bigboss at bbc.co.uk. It is Big Boss, isn't it? Yeah, not Big Boss interview. Brilliant. Do I need to say that again or do I bleed into it? I'm going to leave it in at the end.