Raspberry Pi CEO Fireside Chat

Hello. I'm good, I think. Yes, I've just recovered from our first annual results road show as a public company. So I'm just coming down a little bit.

Ah, yes, that's what shareholders like. I was in Edinburgh talking to investors about tariffs and probably 75% of five meetings, 75% of the time talking about tariffs, got off the plane at Stansted, turned my phone on, and I'm like, god damn it, it's all changed.

I guess as an organization we've been on several journeys in parallel. So this is an organization we set up in 2008 trying to get young people excited about computers in the same way that I was excited about computers as a child, in the same way I'm still excited about that. Just to be clear, I am still excited about... We're children of the 1980s. Came from an environment, I guess, as young people where we were surrounded by ambient programmability by accident, right? I mean, there was nothing intentional about it. There was a period of time when the cheapest thing you could build to play computer games or the cheapest thing you could build to be a word processor happened to be a general purpose computer. And you're surrounded with these things and you buy them for one of these other reasons and you are then lured into programming and that's probably a general description of the people I was surrounded with. And that was like BASIC and then 6502 assembler. And the wonderful experience having a BBC Micro that had that kind of extremely smooth progression from BASIC into assembly language. And so that's probably a pretty good description of the people I was surrounded by at university. It's a pretty good description of the people I worked with when I was a semiconductor engineer at Broadcom. And the people, anyone between the ages of probably 30, 35 and 55, 60 at Raspberry Pi today has had some variant of that route into computing.

Yes, absolutely. You know, we try and hire competent people. Do competent work. But I think that there is always that sense that the people who tend to work in technical roles at Raspberry Pi tend to be people who've had both a hobbyist interest and some sort of academic grounding in computing. Right, so really Raspberry Pi, the disappearance of those computers, that programmable hardware from children's lives led to a collapse in interest in computing and a collapse in applicant numbers to the University of Cambridge. I think we went from something like 600 applicants a year in 1999 to roughly 200 in 2008, the year we founded the foundation. So you're kind of at a point, the university had got to a point that it wasn't admitting anyone who wasn't good, but it got to a point where it was admitting everybody who was good. And that's a situation where you can't accommodate any further shrinkage of the applicant pool without either having to shrink the course or compromise your standards.

And so this is kind of where Raspberry Pi comes from. It's really trying to, it starts off trying to get young people excited about computing. Takes us a little while, took us three or four years to go from that kind of initial concept and the idea was always to produce a physical object, a piece of hardware, to go from that realization, that concept of what we wanted to build, to having something we could build and to having something we put in the market. So we launch in February 2012, the 29th of February 2012. So we don't have many birthday parties. And the birthday parties we do have are generally spectacular but we save up for 3 years. So we launched this product and of course the people you sell to are not kids, right? So you sold 100,000 Raspberry Pis on the first day to people like the people in this room, right? You're selling to people who already know about computers, already know about electronics. And I guess the first journey we went on was from being very education focused to being very enthusiast focused.

And then the second journey that we went on, pair of journeys we went on, having sold to enthusiasts, those enthusiasts turned out to be the key to our impact in education because so many of those enthusiasts are parents or teachers or volunteers. We have an amazing tradition in this country of volunteering. Many of the enthusiasts we sold Raspberry Pis to on the 29th of February 2012 went and took those Raspberry Pis with them into code clubs, computing clubs that they were volunteering in. So a lot of the early impact, and obviously the Raspberry Pi Foundation which started as a charity, the foundation today does a lot of regular engagement with the formal education community, but the early educational impact of Raspberry Pi is this kind of anarchic clubs-based world. So a lot of our early impact was there but then the other journey we've been on, so many of those people are professional design engineers and they take Raspberry Pi with them into, and we're not the first people to walk down this road. I think Arduino walked down this road kind of five years before us but it is a well understood path from enthusiasm to industrial and embedded adoption.

I think a couple of things, there was definitely a transition. And we'll come back, this has happened to us again recently. There was a transition in 2015 when we launched Raspberry Pi 2, the quad Cortex-A7 based successor to the original platform. Partly because it was an enormously more powerful device. It was about six times, for a multi-threaded workload had about six times the CPU performance of the original product. But I think as important was the idea of having a product line, of having a second product. That when you launch your second product you are implicitly saying to people we're serious about this, we are not a flash in the pan, we will be around for a while. And so you saw this big uptick in industrial, in the aftermath of Raspberry Pi 2, a big uptick in industrial adoption not just of Raspberry Pi 2 but also of Raspberry Pi 1 because people could have confidence and people would look at that product line and say I'm going to pick that point on the line, I'm going to put that, it's not a point, it's a line now and we still sell Raspberry Pi 1s. Of course, I think a distinctive thing about us is that we still support, if you go and download Raspberry Pi OS from the website today and you were one of those 29th of February people, you can run it, you can run Raspberry Pi OS on your 29th of February 2012 Raspberry Pi quite slowly. But that idea of continuous support, I sold 451 Raspberry Pi 1s last month.

Yeah, that's it. It's an easy number to remember because it's the temperature at which paper catches fire.

Yeah. So I think you see probably by 20, the tipping point's probably 2019, 2018, 2019 around about the time of launch of Raspberry Pi 4, you see us going over 50% industrial and embedded. So industrial people using Raspberry Pi themselves in an industrial context. So that's people doing factory automation. I think Brompton's a great example from our website. The Brompton bicycle factory is entirely automated with Raspberry Pis right now. With people buying them and building their own bespoke industrial process automation, production automation, test automation solution or embedded people building products around Raspberry Pi. And that tipping point probably came in 2018, 2019 and we're now at a point where of the 7 or 8 million Raspberry Pis that we sell in a year something like 80% of them are going into either an industrial or an embedded context. And I think I'll come on to education in a second because it's actually quite a good follow-on from there.

Yeah, I think there's this idea, there's some kind of boiling point as well. Yeah, there is a kind of boiling point. And that boiling point has moved over time. So I think successive generations of SOC, successive generations of silicon are becoming harder to work with both from a software, hardware perspective, a hardware design perspective. The signal integrity and power integrity challenges for doing a good implementation of a modern high performance ARM SOC are quite substantial and they also become more challenging to work with from a software perspective. Just the size of the code base, the size of the board support increases over the years. And what that tends to do, I think, is push not just towards Raspberry Pi, but it tends to push people at mid-volume OEMs, OEMs out to 10, 20, 30,000 units a year, tends to push OEMs to consider modular solutions. There are quite a few module vendors here. It's the show, right? And really I think that's a demonstration of this trend towards people wanting a degree of abstraction between them and the more modern silicon platforms that's gone up over time.

I think there's another interesting point, of course it's kind of, that's an economic calculation, right? That it is expensive to maintain the team that can maintain a chip-on-board implementation, hardware implementation and board support in production for an extended period of time. And therefore, if you need to amortize the cost of that team across a thousand units, the allocated per unit cost of that team is very high. So there is a kind of an economically rational ball point and our kind of current estimate of that is about 50,000 units a year is the point where we find our modular customers starting to kind of look at the door a little bit. You know, if they've made a make-by calculation at a thousand units and they've chosen to buy at 50,000 units maybe they do the calculation again, they run the spreadsheet again and decide to make. I think what's interesting though and back to the point about education and back to there still being problems even though the foundation has had an amazing impact on computing education in the UK there are still talent pipeline challenges to solve. We actually find an increasing number of our modular customers staying with us well beyond what you might imagine is the economically rational point to transition back to make simply because of the challenges involved in sourcing the engineering talent. Just because you think well I've got to staff up a 5 to 10 engineer team to do this well, you can build a spreadsheet that tells you to do it but where are you going to find the 5 to 10 engineers.

Yeah, that's right. And I think that's a lesson that we've learned over the years. Whenever we have, we're all technologists, right? We're all enthusiasts. We're all perfectionists. We all like to build perfect things. We all like to build beautiful things. And the temptation when you do a new platform is always to change everything. And I think we were lucky with Raspberry Pi 2 and Raspberry Pi 3, they are very incremental evolutions of the Raspberry Pi 1 platform because our volumes then were small enough we couldn't justify a complete new SOC development and so we were kind of forced for a couple of generations to have very incremental changes to the platform and that got us used to incremental change.

And so Raspberry Pi 4 is the first time in 2019, it's the first time that the entire Raspberry platform had been reimplemented so it's a port from 40 nanometers down to 28. So you have this big change from Raspberry Pi 3, 3 Plus to Raspberry Pi 4. But what we'd learned by that time was not to take advantage of all of the flexibility that that change brought with us. And the same thing with Raspberry Pi 5. Now, Raspberry Pi 5 launched in 2023. Actually, an even more radical change to the philosophical underpinnings of the platform because of course we go from a monolithic SOC, the 40 nanometer monolithic SOC compute and IO, 28 nanometer monolithic SOC to a chiplet architecture for Raspberry Pi 5 where we have the BCM2712 16 nanometer logic chip which is almost purely a fast digits chip. It's almost entirely digital and then you have the RP1 chip. For the first time a big Raspberry Pi with Raspberry Pi design silicon on it. That's a huge philosophical change but when we were designing the change we did our very very very best to minimize the impact, the amount of impact of that change above the abstraction layer that we present to our customers. And that's been good for customers I think because it's allowed them to move backwards and forwards up and down the product line. But more importantly, it's been good for us because it's allowed us to support three quite different generations of technology, Classic, 4 and 5, with one software team. And that's how I am still able to sell Raspberry Pi 1s, supported Raspberry Pi 1s, 451 of them. Because I don't have to have a Raspberry Pi 1 software team. I just have a Raspberry Pi software team.