Rene Haas, Arm CEO: AI demand won’t slow

0:00This BBC podcast is supported by ads outside the UK. Your network was successfully breached by Comtit team. One of the most notorious gangs of hackers that ever existed. All of your files were encrypted with military-grade algorithms. Making outrageous demands. Let them pay millions. I'm Jeff White and I'll take you inside a business like no other. Where cybercriminals piled up the cash. Until divisions grew and somebody took revenge. Call me to Ukraine! Cyberhack season 4, The Comtit Files. All episodes available now. Wherever you get your BBC podcasts. With an Amazon Music subscription.

1:27I feel like they don't care what governments do. This is a war. The first thing that we want is the word to end. For this interview, I met René Haas, the American Chief Executive of the British Technology firm, Arm Holdings, in Cambridge. Founded in the early 1990s, the firm began its life operating out of an old turkey barn in the English countryside near Cambridge. As technology shrank and became increasingly mobile, Arm, whose main focus is the design of efficient chips, position themselves at the heart of the boom. It's now mostly owned by a large Japanese tech investment group and their chips can be found in almost every part of modern life from smartphones and cameras through to cars and supercomputers. With the AI revolution continuing to pace, analysts have said that Arm will have a key role to play in the coming years. Arm is the heart of the semiconductor industry and the electronics industry to that end. We are the CPU or the brain that sits inside every electronics device.

2:29And the brain is so important for every electronics device because that's what essentially runs the chip or the program or the end device. We're in data centers, we're in automobiles, we're in PCs, we're in smartphones, we're in earbuds. But we are the brain that makes those things go. We are the brain in today's robotics. And I do think in 10 years we will see physical AI robotics in a very large way, whether it's around manufacturing, cleaning services, security, infrastructure, remaining building bridges, doing repairs. Welcome to the interview from the BBC World Service with René Haas. I want to start pretty big, right? I mean you sit somewhere almost, nobody else gets to sit in this period of extraordinary change in technology and in our economy. You work inside and video, you work in semiconductor industry for three decades or more, and you now run Arm and your technology is nearly every phone, every device on Earth.

3:31So what does the world look like in 10 years time because of AI? And what's the thing the public has in understood that you can tell us about what's yet to come? Gosh, it's even hard to say what's going to look like in five years, but I think in 10 years artificial intelligence will be ubiquitous into every market that we know, every vertical we can think of, whether it's health, whether it's education, it will be pervasive everywhere. It will have some pretty dramatic effects I think on how we live, play and work. Some jobs will go away, some jobs will be created that we don't know about, but at the heart of it will be AI. And you have experience, not just if you're like a core supply chain of the chips that drive the AI, but also in the applications, you know, talks us about the key growth area, you're in every corner of smartphones or automotive or whatever. Robotics for example is something that you've been pushing hard. Yeah, maybe I'll link back your first question with robotics.

4:34So arm is the heart of the semiconductor industry and the electronics industry to that end. We are the CPU or the brain that sits inside every electronics device. And the brain is so important, as you can imagine, for every electronics device, because that's what essentially runs the chip or the program or the end device. We're in data centers, we're in automobiles, we're in PCs, we're in smartphones, we're in earbuds, but we are the brain that makes those things go. Robotics, to your first question, I think in 10 years, will be a huge change function. We are the brain in today's robotics. And I do think in 10 years, we will see physical AI, robotics in a very large way, whether it's around manufacturing, cleaning services, security. Infrastructure, remaining building bridges, doing repairs. We will see. People talk about robots and they're not, we always use the sci-fi future of actual robots.

5:36Absolutely. As workers, physical workers, fixing stuff, cleaning stuff. Absolutely. Yeah, 100%. If you almost work backwards, you'd ask, well, why can that not happen? Yeah. Because artificial intelligence today, one of the things that enables robots is for the robots to learn. That's a very key point because robots prior to AI were built for a certain purpose. You see them in factories today, they install a tire and a factory or inside a distribution area, they're the forklifts that move up and down. With artificial intelligence, these robots can see, learn, and essentially be reprogrammed for new tasks. So, in the service industry, the robot that was programmed to make a bed can now make a bed, but can also learn how to arrange the towels in a room or clean the dust bins or whatever you want to go off and do. This is only enabled by AI. AI allows these robots to learn, these humanoids. So, what will happen is you'll have these general purpose systems that because of AI, we'll be able to learn and deliver new types of workflows.

6:43That is going to happen. 100%. We need to get the costs down. We need to be able to get them to be lighter, faster. But that's just an engineering cost issue that we're going to get our hands around. Well, massive consequences for that. I want to just take you back to the origin, your origins. For those that don't know, I mean obviously you're ubiquitous in the tech industry, but you're not a brand that consumer would buy at this point. Just describe what it is, the core of what you have done and where you've come from and just how many of these devices you're in. Yeah, so ARM is business to business, meaning that we're not a consumer brand, we don't sell to consumers. Our customers are people in the semiconductor value chain or people in the electronics value chain. But the way to think about is if you're building a chip and that chip is essentially inside of your vehicle, your data center, your automobile, your consumer appliance, that chip that makes all those things go, we are the company that supplies to those chip makers.

7:46So what does that mean exactly? We build a CPU, as I mentioned earlier, which is the brain, the essential element. But what we do is we design that brain and we license it to a company building chips. So it's the blueprint. If you're building homes, we're providing the blueprint to how you build that house. And then it's a very wide business model that essentially Apple, Nvidia, Qualcomm, Samsung, Tesla, Amazon, Google, Microsoft. It would be very hard to name a company that doesn't use us, as opposed to who does use us. So we are in every major device you can think of, over 350 billion chips have been shipped since the company was in seventh. So it's starting to number up. It's starting to number up. It's starting to be a huge number. Yeah, it's probably low. I think we've figured out that not only have we shipped enough chips for every person on Earth, we've shipped more chips than every person who's ever lived on Earth, add them all up, and we're still larger than that. So this all begins where we are right now. In Cambridge in the 1980s, a team behind the BBC Micro, the ACON, computer, some engineers, with not a great deal of money,

8:54I understand in a barn, designing chips that were so frugal that they even worked when the circuit board wasn't plugged in. Oh gosh, yes, we're in a beautiful facility right now, so I can assure you that this is not where it all started. It started in a bar not far from here. It was a joint venture between a number of companies, and what the company was looking to do was to design a new microprocessor, a new CPU, for use. In fact, the first use case was the first PDA, something called the Apple Newton, and there were two big requirements for that first PDA. This is 1990s, now long, long time ago. It had to be low power, because you're running off of battery, and it also had to be rather inexpensive, because it had to go inside a plastic package. Back then, they were in very heavy ceramic packages, so we needed it, so it didn't generate a lot of heat. So that was design requirement. Make it inexpensive, and don't draw a lot of power. And to your comment, yes, the engineers overachieved, because what they found out was the very first microprocessor,

9:56the way these test boards are built is you have a test board, has a bunch of other components, and you plug it into wall in the processor runs. What they found out overnight on the very first run was when they unplugged it, so no power is on the board anymore. So technically, the CPU should not work, the CPU was still working. And it was still working because there's tiny little bits of leakage currents that exist on the board. That sipping amount of power was still enough to make the CPU go. So they thought, oh my gosh, we're really onto something. We've got something that's incredibly power efficient, which is kind of the heart and DNA of the company. And the critical point that you still retain to this day is the combination of compute power and power efficiency in terms of energy use. That's right. The company then grew quite a bit, and the first designs were the GSM phones, and then smartphones. And now, as I said, we're really everywhere, and power efficiency is the key to it. But I think one of the most important things that happens with any company when it's conceived are the habits you learn, relative to designing products.

11:01And we were fortunate, and I'm super fortunate now, as to see the company, that DNA was forged around low power and being very power efficient. And that really found itself into every single product that we build, which is why now even in data centers, when people think about, oh my gosh, ARM, aren't you known for smartphones chips? How are you possibly being used in these data centers? Of these data centers, you use hundreds of megawatts and gigawatts. Anything you can do to be more power efficient is incredibly critical. You're listening to the interview from the BBC World Service. Your network was successfully breached by CONTE team. One of the most notorious gangs of hackers that ever existed. All of your files were encrypted with military-grade algorithms, making our rages demands. Let them pay millions. I'm Jeff White, and I'll take you inside a business like no other, where cybercriminals piled up the cash, until divisions grew, and somebody took revenge.

12:09CyberHack, season 4, the CONTE files, all episodes available now, wherever you get your BBC podcasts. I'm at the Cambridge headquarters of ARM Holdings to interview its CEO, René Haas. His company is the hidden hand in the chips that control hundreds of billions of devices and AI data centers. The company is rooted in British know-how and a culture of the experiments with 1980s computers, such as the ACORN and BBC Micro. It now plans to be at the heart of a world of self-learning humanoid robots and AI data centers. Okay, let's return to my conversation with René Haas.

13:11You're now involved in the sticky business of supply chain manufacturing, trying to get hold of materials, testing equipment, everything. And the moment when the world's most advanced industry just can't make enough of this stuff. Yeah, yeah. As you can imagine, given where we sit inside the ecosystem, relationships with people who make memory, the microns, the Samsung, the Heinigs of the World, the TSMCs of the world, these are all people that we know and have strong relationships with. It's not like we showed up and said, hey, we're going to now start to become a customer of yours. They were like, who are you and what do you do? They know us all very, very well. But yes, to your point, we are in an absolutely supply constrained environment, which I think we will be in for a bit, because the demand for artificial intelligence touches everywhere and everything. And as a result, chips are at the center of it and supply is going to be tight. And memory chip prices have obviously nearly doubled because of the AI data centers who've run the puller supply,

14:13making smartphones more expensive as affected demand for smartphones as well. Is this just a bump in the road? What we're seeing in terms of the chip market, or could there be something more fundamental going on here? If it's a bump, it's a really, really big bump. Because the demand for chips and artificial intelligence and memory is at a scale we've not worked on before. And you say, well, why is it different this time? If you look at artificial intelligence and the way it works today, these models with lots and lots of parameters that need to train and then essentially provide all the information from an inference standpoint, there's a lot of things that need to remember. When people say trillions of parameters, that's really the amount of information being used to weigh out, how answers are derived. So when people say, oh, the model has taken all the information that we have on the internet and then some, all the information in the internet and then some, that's a lot of memory. When you think about what's required there. So it's different this time just because of the sheer scale.

15:15And when you think about the scale, the capital expenditures for building out new factories is immense. It's tens of billions of dollars. It takes a long time to build these factories in two, three years. And then when you look at the amount of spending that's taking place with the hyperscalers, hundreds of billions of dollars to build these new data centers that essentially will get the compute and the power and the cooler, etc. These are big, big numbers that are somewhat circular in terms of driving a world of hand. And sustainable, is trillions of demand going in a particular direction? Does people are saying that they'll run out of steam? Are you seeing any sense of that at all? You know, the way I look at it is when I went public in 1999 and I was at a different company at the time. And so I lived through that dot com bubble. And at the dot com bubble, we went through obviously a huge build out of infrastructure. At that time there was, you know, end quote, a lot of dark fiber that was not being used.

16:19Today we don't have that issue, all the GPUs are being used. But one could argue, well, it's gush, will we go to come level of capital correction or valuation correction? I think that's not impossible. I think it's quite potentially could happen. That doesn't mean that the demand for artificial intelligence goes away. And if you, again, let's go back to that time frame with the internet build out. We did not have an issue that the internet 20 years later is not a valuable utility. In fact, it's almost too valuable, right? If you don't have access to a smartphone, try to get into a sporting event or a concert. It's not going to happen. That's the internet at its backbone. So artificial intelligence will be the same. It's going to be table stakes for everything that we do. We may see a short term bump in the road or some pause or whatever you want to call it. And then people will freak out and say, you know, end quote, a i's dead, which is complete nonsense. Because it's only going to get better from here. And it's going to be a greater utility. So I don't worry about it too much. You know, we may see something like that. But the long term demand prognosis.

17:21I don't know how people can argue against it. That's logical to me. So trillions of market value have been created in recent years. I'm intrigued by sort of arms position in that ecosystem. Clearly, the value has gone to the GPU graphics processing units creators. Because they help create the models, train the models. And where you sit has traditionally helped more with the use of the models, the inference of the models. Do you think that that's up for grabs? Like, you know, so much of the valuations have gone towards the creation of the models. But and towards the AI companies. Do you think that could be up for grabs as AI evolves in the direction that you set? Well, right now I think the value has kind of gone to anybody who's in the electronics industry. Right? If you look at the top 10 market kept companies in video, Apple, Google, slash alpha bet, Amazon, Meta, go back 20 years.

18:21It's general motors. It's exon. It's completely different industry. So now all the value top 10 value companies are really in our world. Then when you sub segment inside the world, you know, semiconductors, memory companies are huge, valuable. Embidies very valuable, AMD's very valuable, broad comms very valuable. We are quite valuable. I think it's a recognition that semiconductors are essential. Arm, we are very unique because we have two ways that we deliver value to our customers. One is our IP business model, providing this blueprint to everybody in the industry. And then our new business, the AGI CPU business is huge value because frankly, CPU growth has been outpacing GPU growth over the last year or so. But it may be that the frontier AI companies that are investing so much and getting the absolute best form of their AI models. It might be that the impetus is towards maybe more value for money models, cheaper models, generic models.

19:26You know, we've seen this move towards between open and closed. How are you positioned if there's a move away from the more expensive frontier models towards something that's a little bit more commoditized? Yeah, it's a great point because I think the story is still to be told relative to the delta between these frontier models that run the best you can possibly buy versus the open source models which are 90% as good but much less costly. I think it can be 90% as good. Could be. Yeah, could be. Thankfully for me, I don't care. Because the frontier models are going to run on arm and the open source models are going to run arm. So I think it's going to be a little bit more plumbing that all that's going to run through independent of it. So thankfully, I don't have to pick a winner in that case. I think it'll be interesting to see how it plays out but in either way it's fine for us. Surely in arm world, I do not benefit if the AI happens more closer to what we wear and what we use as consumers rather than in some data center that's operated in the desert.

20:27You still on both cases? You'll hatch. I do well in both cases. We are over 50% market share today in the data center, Amazon builds on us, Microsoft builds on us, Google builds on us, Nvidia builds on us. And they build on us whether it's open source or closed. So it's fine for me. And then if it ends up on the edge, which it will, and you'll have some hybrid, whether it's to your point on your lapel or in your car or some new device that's not your phone, that's all going to be on arm. When I'm told by the AI, the Frontier AI companies, that they are just at the point of exponential growth of capacity of these Frontier models that they'll be able to replace all knowledge work by 2030. You're at the core of the supply chain here, aren't you? It's like, do your chips have the capacity to supply, given all the constraints we've talked about, that level of compute in two or three or four years time? There's a bunch of assumptions being made that they'll just carry on going up. But like, aren't we hitting the buffers now in terms of how much compute is available to fulfill these sorts of promises?

21:36The talk now is about multiple gigawatts of data centers, right? There was a data center announcement that Softbank made, for example, a 10 gigawatt build out of a data center in Ohio that we just did as Per Softbank Group. There's been multiple gigawatts of data centers announced in France up to five gigawatts. These are giant scales, right? And then people are talking about Elon and Jeff Bezos about terawatt data centers in space. The constraint for making all that happen is going to be back to the earlier chat about wafers and memory. Can you get enough chips? We come, we come. Right now it's quite constrained. We need more fabs before we can put a data center in space. I'll tell you that much. And we'll probably only put data centers in space when the biggest impediment to data centers is the cost of the data center. I'm struck that some of the main deployments, most advanced deployments of AI, is happening in the AI industry, in the tech industry. Tell me, how have you used AI within your company almost to make the AI? It becomes a little bit circular, but like, is it replacing jobs? Is it augmenting? What's the balance?

22:41I think it's definitely augmenting. I wouldn't be able to tell you today that proudly my head count is reduced by 25% because we have AI bots doing all the work. But we have a lot of engineers using the work just to make them allow their work to be easier and to go faster. One of the things that's a big part of when you design a product or a chip, it's not the design, it's verifying that design. So we throw all the bugs and testing the bugs, identifying the bugs. AI is wonderful at solving bugs, addressing bugs, making it easier to find out what was the cause of the bugs. And we have engineers literally, I had many of them tell me, they'll run a report on a weekend and come in on a Monday in the past. They would have to do all the parados of exactly what they saw when they come back. Two-thirds of this might have been done and fixed. That just allows them to go faster. So I think probably we have 90% of engineering using it on a daily basis. I use it all the time myself personally. So yeah, we use it very heavily inside arm.

23:42And what about in your home life, how would you advise people use it with their children and some concerns as well as some opportunities there? Yeah, for people to say gosh, AI is going to replace humans and replace jobs and knowledge work. AI is going to be there, period. We know that. And then as for all things we are talking about. So I would tell anyone who's talking to their kids or my case grandkids, learn it, embrace it and become comfortable with it. Because ultimately AI is a utility that's going to allow us to be more productive and effective. So don't be afraid of it. And then well, you should be afraid, but it's quite difficult not to hear from the AI leaders that keep on saying how doom, mungery the future might be and how we all need to protect against it and government need to be quite hard. You know, you're actually running the core technology that provides the capacity for this exponential power to replace human labor to do some mildly scary things we're told by the AI companies. So do you think about that responsibility? Is that your job or are you just going to provide as much compute power as possible?

24:46You know, a little bit, but I would say at the same time the estimates of jobs going away and being a blue replaced by machines is a bit overstated. I was looking at it, it was really an article this morning, I came to tell you why I was reading it, but it was about the economy in 1776 when the United States became an independent nation. I was 16% of the revenue and labor was farming. Now in the US about 4%. But where people worried about, oh, the farming jobs are going away and as a result because the farmers have no jobs, there are no jobs almost kind of nonsensical. But I view AI as a similar type of thing, will it change how we work and live and play? Of course it will. But at the same time it's going to create opportunities that we can't imagine today in terms of what that means. So I tend to be an optimist by heart, I would say embrace it, learn it. And I feel an arms role, I feel very lucky to be leading a company that's at the heart of it because I think we can make the planet a better place. Thank you for listening to the interview. If you enjoyed this conversation you can find many more episodes of the interview wherever you get your BBC podcasts including ones with AI experts, Parmi Olsen, Hinge CEO Jackie Jantos and the Kinsey China's Joe Nye. Until the next time, bye for now.

26:13Your network was successfully breached by Comtty team. One of the most notorious gangs of hackers that ever existed. All of your files were encrypted with military-grade algorithms. Making outrageous demands. Let them pay millions. I'm Jeff White and I'll take you inside a business like no other where cybercriminals piled up the cash until divisions grew and somebody took revenge. CYBERHACK Season 4 The Comtty Files All episodes available now, wherever you get your BBC podcasts. The Comtty Files All episodes available now, wherever you get your BBC podcasts.