Redefining Chip Architecture with Arm CEO Rene Haas
- Arm has crossed from IP licensing into physical silicon, with Meta as the trigger. Meta wanted “a general-purpose agentic CPU” and Arm says no one else could provide it, leading to the Arm AGI CPU, introduced last March and presented at Hot Chips. Ecosystem pushback was milder than expected because more Arm-based software benefits customers broadly; launch congratulations included Jensen, Rani Borkar, Amin and James Hamilton. The move also adds supply-chain operations, memory allocation, back-end, layout, implementation and bring-up capabilities to a business Sarah noted had a 98.5% gross margin.
- AI already runs Arm’s engineering floor: 80% to 90% of engineers use it daily, especially on the true long pole of a 24- to 36-month chip cycle—verification, validation, debugging and documentation. Shutting it off would be like rationing 1990s internet access, prompting Sarah’s “there’d be anarchy” and Haas’s “the genie’s out of the bottle.” RTL generation and best-in-class physical design remain less mature because models rely on public data while key information is proprietary. Haas says Arm’s rich IP, documentation and test benches give it an advantage; Elad’s point is that unusable and untestable IP is untrainable and therefore unusable for AI.
- Haas sees idea-to-GDSII for straightforward designs as quite possible in 5-plus years, not necessarily 2 to 3. But a request for a design that is 10% faster than Vera Rubin, 20% cheaper and 30% more efficient will not be solved by pressing a button.
- Supply is likely to remain constrained for 3 to 5 years at least, so long as the transformer remains the unit of energy for AI training and inference. Data-center construction may become the next bottleneck: many projects are not ahead of schedule or using less labor than expected, and some parts of the US are discussing slowing or restricting development. Haas says that may be preferable to wafer and memory capacity becoming the binding constraint. Setting valuations aside, he says oversupply relative to demand is “not even close.”
- SoftBank could provide capital, ecosystem access and a potential home for chip startups. Haas advises young companies in CapEx-intensive industries to form strategic partnerships early with supply-chain participants, private equity and banks because access to capital is a gate. SoftBank Neo is the group’s intent to become a neocloud, potentially giving companies with chip technology an alternative to first winning a design slot at Microsoft or Google. Haas leads the direction of Ampere, Graphcore and Stack AV and helps Masa formulate and execute strategies around robotics, OpenAI, infrastructure and Arm.
- Robotics could become “almost like something out of The Jetsons,” across both humanoid and task-specific forms, but costs are high and business models remain unproven. Distribution centers may automate heavily, while Elad points to factory automation, delivery and autonomous trucks as early areas. Haas says Arm will be pervasive in robotics, from sensing and perception at the fingers to the compute in humanoids.
- Haas supports more US semiconductor manufacturing and says the export-control race is an infinite game with no winner; he warns that critical technologies could end up outside the US. Elad says that outcome would be bad and argues for staying at the technological forefront. Haas attributes data-center backlash mainly to fear of job loss, which he calls poorly grounded; Elad also points to organized media influence, while Sarah cites an electricians’ union asking that data centers not be banned. On CPUs, Haas says the accelerator focus after ChatGPT obscured the CPU’s continuing role: as workloads move from training toward reinforcement learning and inference, CPUs orchestrate where tokens go, alongside accelerators and memory. That applies from data centers to edge devices, where a 50-watt GPU is impractical.
1. Arm crossed from IP into products — because Meta asked for a chip nobody else could provide
- Haas describes two positions in the chip supply chain: Arm’s primary business licenses CPU IP used in “smartphones, data centers, automobiles—you name it,” giving it visibility across automotive, data centers and smartphones; since last March, Arm has also had its own product, the Arm AGI CPU. That puts the fabless company into buying substrates, wafers, memory and other supply-chain inputs itself.
- The evolution was individual IP components → compute subsystems, for which demand was “insane” despite initial skepticism that chip designers would want Arm to provide that assembly blueprint → a physical product. Meta wanted a general-purpose agentic CPU, and Haas says there was no one else who could give it to them.
- Arm expected more customer pushback but heard surprisingly little because more proprietary and open-source software in the wild benefits the broader ecosystem. NVIDIA, Amazon, Microsoft and Google—all builders of Arm-based server chips—were supportive. At launch, Jensen, Rani Borkar, Amin and James Hamilton congratulated the company.
- Sarah recalls Arm’s 98.5% gross margin. Haas contrasts that IP model with his first impression after joining Arm in 2013: “No inventory, no RMA, no scrap—what’s not to like?” Physical products now require supply-chain operations, work with TSMC and Samsung, memory allocation from Samsung, Micron and SK hynix, plus back-end, layout, implementation, physical infrastructure and bring-up labs. Leadership hires from Broadcom, Qualcomm and NVIDIA helped Arm build that capability quickly.
2. AI runs across Arm engineering, while documentation may make IP trainable
- Sarah cites that day’s OpenAI news about Jony Ive and a new chip, including a claim that AI tooling helped accelerate time to market. Haas says a chip design can take 24 to 36 months, but architecture, RTL generation and architecture mapping are not the largest time sinks; verification, validation, debugging and documentation are. AI is particularly good at those tasks.
- He estimates that 80% to 90% of Arm engineers use AI daily. Turning it off, he says, would be like having the internet but allowing access only from 2 to 4; Sarah adds that “there’d be anarchy,” and Haas concludes that “the genie’s out of the bottle.”
- The tools remain less mature for RTL generation and for physical design and implementation in best-in-class systems because models are trained on public material while much of the relevant information is proprietary. Haas says Arm has a built-in advantage: a rich IP portfolio accompanied by documentation, test benches and explanations of how to build the IP. Elad sharpens the point: if IP is unusable and untestable, it is “actually untrainable,” and therefore not usable for AI. Haas says Arm is working with model makers to address the gap.
- Asked whether the 24- to 36-month cycle could shrink to 6 to 12 months, Haas says he does not know if that is only 2 to 3 years away, but that in 5-plus years it may be quite possible to go from an idea to a GDSII file for straightforward designs, removing much of the design and verification work. A request for something 10% faster than Vera Rubin, 20% cheaper and 30% more efficient on a given model will not be a one-button task.
3. Constrained for 3 to 5 years; data-center construction may be next
- The proliferation of AI-chip companies does not eliminate the industrial bottleneck. Haas describes young companies with innovative designs and substantial funding selling into an industry with massive capital requirements, where relationships with memory and substrate vendors are critical.
- He expects a constrained environment for “3 to 5 years at least,” and says that will not end in 12 or 24 months. The reason is that transformers are compute- and memory-intensive as the unit of energy for AI training and inference.
- Asked about the next bottleneck after packaging and memory, Haas points to data-center buildout. He says not many projects are ahead of schedule or need less labor than expected, while some parts of the US are discussing slowing development or imposing restrictions. That may be acceptable because otherwise wafer or memory capacity could become the binding constraint. Multiple “governors”—not state governors, but different constraints—will throttle growth.
- On the AI-bubble question, Haas distinguishes valuation bubbles from oversupply relative to demand. Setting valuations aside, he says the answer to whether supply is close to exceeding demand is “not even close,” because demand remains insatiable given how the models work.
4. SoftBank Neo as a potential home for chip startups; Haas helps execute Masa’s strategy
- Haas says Arm’s publicly traded structure and very large single shareholder give him frequent informal investor discussions with SoftBank. His advice to young companies in CapEx-intensive industries is to form strategic partnerships early with supply-chain participants, private equity and banks. Semiconductor startups are attracting investment again, but access to capital remains the gate.
- SoftBank Neo is not described as an operating neocloud already; Haas calls it SoftBank’s intent to become a neocloud. In that world, it could become a home for young companies with chip technology that might otherwise have to pursue a design win at Microsoft or Google.
- Haas leads the direction of Ampere, Graphcore and Stack AV, which works on autonomy. More broadly, he says he is involved in many discussions with Masa, helping formulate and execute strategies around robotics, OpenAI, infrastructure and Arm.
- SoftBank’s involvement in robotics, energy and data-center infrastructure gives Arm a broad view of industry direction and could provide a home for Arm products. Haas says that would not necessarily mean entering the broad merchant-chip business; Arm could make products for SoftBank.
5. Robotics: Jetsons-scale potential, with distribution and automation early
- Haas broadly agrees that robotics is showing better generalization across tasks and environments but has not reached widespread deployment. Robotics 1.0 was purpose-built: a new automobile line or other equipment could require ripping up the existing line. Robots that can learn from training or what they see, paired with mechanically general-purpose designs and falling costs, could change the scope dramatically—“almost like something out of The Jetsons”—across construction, infrastructure, service and security.
- He expects both humanoid and specialized forms. Some work is optimized for a roughly six-foot human body and existing tools, but other applications will favor task-specific designs.
- Haas says Arm will be everywhere in robotics, including real-time sensing, perception and microprocessors at the fingers. He also says that most of the brains seen in humanoids today run on Arm, citing NVIDIA and work Qualcomm does.
- He cautions that the business models have not been figured out and that robots are currently expensive enough to make direct purchases difficult. Distribution centers are a clear early application and could eventually be automated through delivery. Elad adds factory automation, delivery and distribution, including autonomous trucks, which he considers robots “of sorts,” as likely early areas.
6. The leadership doctrine — and why CPUs move the tokens
- Speaking as an American citizen and semiconductor veteran, Haas recalls the US response to Japan Inc.’s aggressive memory pricing in the 1980s through SEMATECH, which aimed to refortify the US semiconductor industry. He argues that the US needs more fabs for national security and supply-chain diversification, and says the UK should pursue the same goal to a lesser extent given its smaller scale.
- On export controls, Haas frames the effort to limit chips so China does not “win the race” as an infinite game in which there will not be a winner. He warns that the US could instead reach a position where critical technologies are not US-based. Elad responds that this would not be good for national-security and economic reasons, arguing that technological leadership also creates surrounding ecosystems.
- On data-center resistance, Elad points to organized media influence. Haas says the backlash is largely fear that AI means job loss, a fear he calls “not well grounded at all,” and says some claims about data centers—such as fake, tainted water concerns—are being made up to create fear. Sarah cites an electricians’ labor union asking that data centers not be banned because they create skilled work.
- Elad supplies the postwar Detroit auto-industry analogy, where surrounding states and companies benefited from the ecosystem. Haas says data centers work similarly: energy, liquid cooling and other infrastructure create jobs even when the facility itself appears to have few workers. His first-principles conclusion is that there is no downside to being the technology leader; laggards have the entire script dictated to them.
- In the closing CPU discussion, Haas says that after ChatGPT’s explosion, attention shifted heavily toward accelerators and the CPU was overlooked. But every computing problem still uses or can use a microprocessor. As workloads move from training toward reinforcement learning and inference, something must orchestrate, arbitrate and decide where tokens go: “Where are the trucks that are going to take the tokens away and give them to the users? That’s what CPUs do.”
- He frames the system as CPU, accelerator and memory, applicable to data centers, automobiles, robots, phones and wearables. Arm is especially well positioned at smaller footprints, where edge AI needs local processing and a 50-watt GPU cannot simply be placed on someone’s head.
Full transcript
There's no computing problem that's ever been invented that doesn't utilize and can't utilize the microprocessor. It is the heart of everything. All roads lead through it, around it, and past it. Something has to do the orchestration, arbitration, and decision-making around where those tokens go. That's what CPUs do.
Chip design can take anywhere from 24 to 36 months, depending on the complexity of the chip. The actual design—the architecture, the RTL generation, if you will, and the mapping of the architecture—is not the largest amount of time. The largest amount of time is in the verification, the validation, the debugging, and the documentation. AI is really good at that.
And if we were to shut it off, it's like being in the 1990s. You've got the internet, and you're now saying, "You know, only internet between the hours of 2 and 4."
After that, go to the library that we have down the hall. It's got all the books that you can look up. It'd be anarchy.
The genie's out of the bottle, right? There's no stopping that.
Rene, thanks so much for doing this with us.
Pleasure.
1. Arm and Chip Supply Chain
Congratulations on the chip presentation at Hot Chips and all of the progress that Arm has made. I think there's an enormous amount of interest from the technology industry and the software industry in better understanding the chip supply chain recently. For anybody who's not super familiar, can you explain Arm's position in it? Then we'll get into more recent topics.
We have 2 positions in the chip supply chain. Our primary business is licensing IP: the CPU core that finds its way into smartphones, data centers, automobiles—you name it. Our customers are the ones who either build the chips themselves, like Samsung, which has its own fab, or, for the vast majority, companies that take their chip designs to TSMC and get them taped out.
In that world—and this is the cool thing about Arm, because we're so broad in terms of the markets that we serve—we see everything. We have a very good sense of what's going on in automotive, data centers, and smartphones. We see the supply chain situation from all angles.
2. Shift from IP to Manufacturing CPUs
We also introduced our first product last March, the one you just mentioned at Hot Chips, the Arm AGI CPU. So now we're in that soup ourselves, from the standpoint that we're also having to figure out how to buy substrates, wafers, memory, and so on. We're up to our waists in everything on the supply side.
Why'd you make the move now? Arm has existed for a few decades. The focus was always on IP, which is effectively designing the way that different chip components are put together. Then you license that out to other people to actually manufacture and incorporate into their designs. Why did you decide to start making some of your own CPUs?
Yeah, it was an evolution from the early days, when we just supplied the IP components—the pieces, the CPU IP, the GPU IP, the system IP, and so on. A few years ago, what we were starting to see was that product cycle times weren't slowing down. Chip manufacturing times were extending. The ability to get solutions out faster was becoming more and more important.
We moved from these individual components into what we called compute subsystems. When we went and did the road show a few years ago, I used the LEGO analogy: essentially, we're providing the blueprint for how you stitch it all together. Demand for that was insane.
Initially, people thought, "Well, people aren't going to want these subsystems, because that's what a chip designer does. Why are you providing that piece?" But it saved time to market and a whole lot in terms of cost and speed. The physical product was sort of the next leap, if you will.
There are certain sets of customers that will license IP from us, and they've got all the capability in the world to build chips based on Arm. There are a lot of companies that want to have products based on Arm, but not all of our customers build products that serve those markets.
3. CPU IP and Customers
Meta was that first example. They wanted a general-purpose agentic CPU. There wasn't anybody out there who could give it to them. They came to us and said, "Hey, why don't we do this together?" That's how we got into it.
How has that landed with the rest of your customer base?
One of the things that we were very careful about was making sure the ecosystem was on board with this, because CPU IP is really only as good as the ecosystem—the ecosystem of chip people, software folks, and people who build around it.
We talked to just about everybody who was a customer and said, "How do you feel about this as a direction we're going?" Surprisingly, we got a lot less pushback than I thought. The reason was that the more software that's available in the wild, whether it's proprietary and/or open source, benefits the broader ecosystem and the customers themselves.
4. Softbank Leverage and Capital Strategy
Whether it was NVIDIA, Amazon, Microsoft, or Google—all people who build Arm-based server chips—they were all on board. I think the ultimate proof point was when we announced the product last March. We had Jensen, Rani Borkar, Amin, and James Hamilton—all the folks from those customers I mentioned—saying, "Congratulations. It's a great thing." So it's been okay.
Where are you in the learning cycle as a business now, selling physical chips? That feels like a lot of new capabilities.
Yeah. To deliver a product, we're obviously a fabless semiconductor company, right? We don't have a fab, and we have no intention of building a fab, but we fit in that ecosystem. That means you need supply chain operations people. You need to work with TSMC and Samsung, as I said. You need to work with Samsung, Micron, and SK hynix to get memory allocation.
On the engineering side, you need a lot more capabilities. You need back-end people, layout people, implementation people, bring-up labs, physical stuff. We didn't have a lot of physical stuff, which was kind of the beauty of the original business.
I remember discovering that Arm had a 98.5% gross margin.
Yeah, kind of beautiful.
I don't think I've seen that otherwise. Yeah.
I came from NVIDIA before I came over here. Most of my career was in the chip world, and I remember coming to Arm in 2013 and thinking, "No inventory, no RMA, no scrap—what's not to like?"
5. AI Adoption at Arm
So we had to add a lot of those capabilities. We have a lot of people on the leadership team who've come from that world. I've got executives from Broadcom, Qualcomm, and NVIDIA. I worked for NVIDIA, so we have leadership that's done this before at other companies. We've been able to build up that muscle pretty quickly.
How have you approached AI adoption? We were speaking earlier about news from OpenAI today about Jony Ive and a new chip that they designed. Their claim is that it was a very fast time to market, and part of that was using AI tooling to design chips faster. How much adoption have you seen there?
I know other companies have also talked about things like adopting formal verification at Amazon or elsewhere for their training chips. I think the chip world is starting to evolve in terms of AI usage, and I'm curious about how you've done that at Arm.
Personally, I'm a huge believer in AI as a utility that's going to help productivity for every single industry. It's going to be the great leveler in terms of companies that can get started super quickly. For industries—whether it's healthcare, infrastructure, or robotics—every industry is going to use artificial intelligence as a utility, full stop.
Since I'm such a believer in this, of course we use it very heavily inside Arm. On the non-engineering side, we're using it all over the place. But on the engineering side, we've seen a huge benefit.
You mentioned verification. Chip design can take anywhere from 24 to 36 months, depending on the complexity of the chip. The actual design of the architecture—the RTL generation, if you will, and the mapping of the architecture—is not the largest amount of time. The largest amount of time is in the verification, the validation, the debugging, and the documentation. AI is really good at that.
I would say we probably have 80% to 90% of engineers today inside Arm who use it on a daily basis. And if we were to shut it off, my analogy I give to people is that it's like being in the 1990s. You've got the internet, and you're now saying, "You know, only internet between the hours of 2 and 4."
After that, go to the library that we have down the hall. It's got all the books that you can look up. People—there'd be anarchy.
6. US Manufacturing Protectionism
So the genie's out of the bottle, right? There's no stopping that. There are certain things that the tools still aren't that mature at. One of them is RTL generation, and then physical design and implementation in best-in-class systems. That's simply because the models are trained on what's available publicly, and a lot of that information is quite proprietary.
That being said, there's a massive opportunity between the ecosystems and everyone in the industry to make that better.
It’s only going to get better.
Have you been fine-tuning models to try to address that gap, given the proprietary information that you’ve offered?
We’ve been working with model makers around that. Absolutely. And I think that’s a big, big opportunity. One of the things I’m proud of at Arm is that, given our business—our core IP business—we probably have the richest IP portfolio, both in terms of not only the IP but also—and this is the killer—the documentation, the test benches, and how you build the IP.
I’ve worked for chip companies in the past that have said, “Why don’t we license this IP that we’ve got? It’s really, really valuable.” And then you get into, “Wait a minute. There’s no documentation. There’s no explanation. No one’s ever going to be able to use this.”
It’s unusable and it’s untestable, right? And if it’s unusable and untestable, it’s actually untrainable. And if it’s untrainable, it’s not usable for AI.
I think we have some built-in advantages based on our business model that’ll allow us to really be able to take advantage of the tools as they get better.
7. Changes in Chip Time to Market
Really exciting. How much do you think—if you were to extrapolate out, this is a little bit of an uncertain question—but if you extrapolate out 2 to 3 years, and all the tooling that’s likely to come in AI, and the ability to fine-tune models against some aspects of the design that you mentioned, do you think that 24- to 36-month cycle shrinks to a year to 6 months? Do you think it stays roughly where it’s at? I’m a little bit curious about how that really impacts these cycles and time to market, because that has pretty dramatic ramifications.
In terms of the clock speed of the entire industry.
I don’t know if it’s 2 to 3 years away, but 5-plus years—
Can you go from an idea to a GDSII file? A GDSII file is the file that you actually send to the fab to go get built for certain designs. Quite possible. So it takes that whole design piece out of the way. It takes that whole piece out of the way relative to the verification. So I think for the more straightforward designs, quite possible.
Now, if you go into the tool and say, “Design me something that’s 10% faster than Vera Rubin, 20% cheaper, and 30% more efficient on this model,” you’re not going to be able to press a button and have it happen right away. But I think in 5 to 10 years, our industry is going to see some amazing differences relative to how chips are designed.
How does it change—I’m sure you had some prediction of this—but how does it change the way you look at the business, given there’s just a big diversity of large players and new players that all want to have their own chip designs now? The Veras and the Gravitons of the world all use Arm. It’s a big step up for them, but it’s a big diversification of the customer base, right? That can be only good.
Oh, absolutely. I think what’s going to matter, back to the earlier discussion we had on supply chain, is understanding the supply chain impacts—how all of that gets built and put into ultimate end products. I think that’s going to become a much more important muscle as we go forward.
Said another way, there are a lot of really great young companies today doing AI chips—well-known companies getting tons of funding, innovative designs, et cetera—selling into an industry where the capital requirements are just massive and the relationships with memory vendors are incredibly critical, or the relationship with substrate vendors.
So companies are going to have to be much
More access to a 3-nanometer line, a 16-nanometer line, an advanced packaging line—all of it.
Yeah. All of that. And I think that is not going to stop in 12 months; it’s not going to stop in 24 months. I think we’re going to be in this constrained environment for 3 to 5 years at least. So long as the transformer is the unit of energy relative to how you generate AI training and AI inference, by design, it is very compute-intensive. It’s very memory-intensive.
If you think about that, that’s going to drive a lot of demand for supply chain acumen, which then goes back to people who’ve got great ideas on chip design. They’re going to need a lot of other things just to be able to get access to capital, wafers, and everything you just talked about.
8. Data Center Buildout Bottleneck
We’ve just had a cascading series of things that have been the bottleneck to more compute for the AI industry. 2 years ago or so, I think it was packaging and packaging-related items, and then eventually now people talk about how it’s memory and things like that that are, in some sense, limiting to certain systems being built at sufficient scale. Do you have a view of what the next sort of bottleneck that’s coming is?
I think building out data centers is going to be a bottleneck. When I say building out, if you look at all the projects being done today, not a lot of them are ahead of schedule or needing less labor than they thought, right? And then, when you layer on top of that a lot of buzz coming from different parts of the country in the United States relative to slowing down data center development or putting restrictions around it, I think that infrastructure buildout could be a headwind relative to everything going on, which may be, in and of itself, okay.
Because if infrastructure buildout was not a headwind, I think capacity for wafers and capacity for memory probably would be a headwind. So you’re going to see a number of different governors, if you will—not governors of states, but different things that are going to throttle the growth of this.
Just expanding for a second, I’ve been on a bunch of panels, and I get a lot of questions about the AI bubble and when it’s going to stop. Setting aside the valuation bubbles, which are a stock market index component, the bubble in terms of, “Are we oversupplying relative to demand?”—not even close. And I think, again, that’s because the demand is insatiable, just given the way these models work.
And infrastructure buildout, access to wafers, access to memory—all of that’s combining.
You mentioned that, and I think a lot of companies are learning today that strategic use of the cap table, access to capital in an era where you either need to consume a lot of compute, or you need to put a lot of CapEx into the ground, or you’re just doing big technical projects like coming up with CPU IP. You run SoftBank Group International. You have this one dominant shareholder. Arm itself as a business is just like a beautiful cash-flow machine from the outside, right? I’m sure you think a lot about the leverage of SoftBank and how to use that well. What advice do you have for entrepreneurs navigating these CapEx-intensive industries from where you sit?
One of the benefits we have at Arm—a publicly traded company, yes, but with a very, very large single shareholder—is that I have lots of informal investor meetings with my chief shareholder all the time about this. We have a big advantage in that there are a lot of things symbiotically we can do together that can help Arm advance its initiatives by having SoftBank as our largest shareholder, and we look to be very, very innovative around that.
To your point in terms of young companies, I would say that strategic partnerships are incredibly important early on, whether it’s with people inside the supply chain, people in private equity, the banks, the banks themselves. It’s a different game now. On one hand, semis are kind of back because you now have a wave of semiconductor startups. There was a long time where that was just not happening—investment in the industry—and now we’ve got a lot.
But access to capital is going to be the gate for them in terms of how they get through that. So I think getting much more creative in terms of how they work with the ecosystem is going to be super, super key. And we at SoftBank, that’s one of the things we look at very strategically: companies that we can bring into the portfolio that we can help, that we can provide a combination of either a backstop and/or, if you think about SoftBank, what we just announced—we, being SoftBank—SoftBank Neo, which is our intent to become a neocloud.
In that world, we could become a home for these young companies that have chip technology that, in other worlds, they’d have to go up and figure out how to get a design win at Microsoft or Google. We can provide a lot of interesting avenues for that.
9. Softbank Portfolio Overview
Can you talk a little bit more about the portfolio things that fall under your purview at SoftBank? I know, as mentioned, there’s Arm, and then there’s this broader suite of things. So I’d love to hear—we’d love to hear more about what else you’re responsible for. And we had some specific questions for some of those as well.
The way to think about it is SoftBank Group, which is headed in Japan by Masa, has a lot of different operating companies underneath it. One of the largest ones is SoftBank K.K., which is essentially SoftBank Mobile. Inside the US, there’s a lot of investment activity going on with SoftBank Group International. There’s SoftBank Vision Fund.
But increasingly, a lot of the strategies that we’re trying to pursue around SoftBank are helping with the strategies that Masa talked about publicly at his shareholder meeting in Japan, which are around robotics, OpenAI, infrastructure, and Arm. I probably have my eyeballs on a lot of stuff, to be honest with you, in terms of helping Masa really realize the execution of that vision.
So, yes, I’m leading the direction of Ampere, Graphcore, and another company called Stack AV that’s doing things around autonomy. But maybe a better way to think about it, Elad, is that I’m in the room for a lot of discussions that Masa is having and helping him formulate that strategy and, more importantly, helping execute it.
Mhm. How has being part of the SoftBank Group, or working with all these different companies—even SB Energy and the broader ecosystem—changed your point of view on what you can do with Arm?
Well, one thing it does is give us a huge bird’s-eye view relative to where the broader industry is going, whether it’s around infrastructure, capital, or energy. But you can imagine it could also provide a home for our products, right? It doesn’t need to be the home, but it certainly can be a home, which is also a big help.
10. Robotics Opportunities for Arm
When we think about the verticals that SoftBank’s involved with—robotics, energy, data center infrastructure—and then you look at the products that Arm has, the only one we’ve announced so far is the Arm AGI CPU, you can start to connect the dots and say, “Gosh, there could be some very interesting opportunities for Arm.” That doesn’t necessarily mean that we’re getting into the broad merchant-chip business. We could just be doing products simply for SoftBank.
We’re a couple of years into serious efforts in more generalized robotics at this point, right? If you compare it to about a decade for LLMs, there are increasingly interesting demo results from companies on generalization across tasks and environments, more robustness, maybe even in-context learning, but not widespread deployment quite yet. First, would you agree with that characterization?
Yeah. Broadly speaking, I think whether it’s humanoids or dedicated machines to do certain levels of tasks that can be retrained, it’s going to be enormous. Robotics 1.0 was a purpose-built industry: You had pieces of machinery designed to do a certain task, and software optimized for that task. If a brand-new automobile line came up, or some different piece of equipment, and the robots weren’t well-suited for that, you’d rip up the line, et cetera. So, as you can imagine, the barrier was pretty high.
Getting to a world where the robots can learn based either upon being trained or what they see, and then when you combine that with being able to design something mechanically general-purpose enough to take advantage of being reprogrammed, and then you layer on top of that the cost coming down, you look at it and say, “Oh, my gosh, what will it not be able to do?”
It’s almost like something out of The Jetsons, right? A lot of things will ultimately be done by robots: construction, infrastructure, service, and security. Right now, you see a lot of stuff on Instagram or TikTok of Olympic races with robots, et cetera. I don’t think anyone’s going to have any interest in watching a sports league of robots. There may be an enthusiast class who might be interested in that, but the broader utility is going to be around a lot of human labor tasks that will ultimately easily be replaced by robots. There’s no question.
A lot of the hypotheses people have about the form factor of robotics tend to split into 2 or 3 camps. One camp is that they’re going to be humanoid, or roughly the human footprint, because so much of the physical world is already designed that way, and the tooling is designed that way, so you can just slot robots right in. Others view it as there being much more specialized, task-specific form factors. Do you have a hypothesis on that?
I think it’s both. There are a lot of jobs and work tasks that are optimized around a person being 6 feet tall and having arms of a certain length. But I think it’ll be both, and I think the fact that they’re going to be smart and can learn—and, to answer your earlier question—Arm is going to be everywhere.
We have a tremendous amount of technology, from a real-time sensing standpoint, around microprocessors that will be out at the fingers. They can do perception and sensing. That’s all going to be Arm-based. Today, whether it’s NVIDIA or some of the work that Qualcomm does, most of the brains—the brains that you see in the humanoids—are all running on Arm today. So I think, for us going forward, the robotics industry will be powered by Arm.
Are you seeing any early indications? I mean, you have this great seat, to your point, where, given the ubiquity of Arm in a lot of these different types of devices, you can kind of see the future before others in terms of where adoption is happening or where shifts are happening from a technology perspective. Are there specific pockets that you think will be most likely the early adopters of robotics that you’re starting to see some signal from?
I think it’s still a little bit early because the business models have not actually been figured out. The cost of robots is so high, right? Because the cost of robots is so high, people buying the robots themselves—that’s a tough model to get people’s heads around. Does it actually replace people? So I think costs need to come down, and the business model needs to be ultimately vetted, because other robotic footprints tend to be things like automotive, certain surgical robots, or—excuse me—distribution centers. There are a few very bespoke applications that I think account for most robotic sales today.
That’s why I was a little bit curious.
Distribution centers, for sure. I mean, that can ultimately go completely automated, right, even to the ultimate delivery.
To me, loosely speaking, a truck that has autonomy is a robot of sorts. So around factory automation and delivery and distribution, that will be one of the very first areas to be automated. No doubt.
There is increasing debate and, very quickly, policy or EOs around supply-chain controls and usage controls around both robotics and chips and data centers. Sorry, I’m going to throw export controls in there—so 4 types of controls. All of these controls are relevant for you now, either from your end-customer perspective or as a relatively new entrant—you’re going to own the end product and have a supply-chain organization of your own. What’s your stance on how protectionist—I realize it’s not an American company, but you do a lot of business here—how protectionist the US or the West should be about manufacturing of chips, creation of data centers, and robotics? What are your overall stances here?
So, putting my American-citizen hat on for a moment—and Arm, as you said, is not an American company; our headquarters is in the UK—we have a lot of employees. I wouldn’t say half our employees, but maybe 30% are in the US, 40% are in the UK, and maybe 30% are in Asia. We’re a global company, but with a huge US footprint. As an American citizen, and someone who grew up in semiconductors, I remember in the 1980s, when the US was the leader in semiconductors and Japan Inc. started to get very, very aggressive in terms of memory pricing and essentially taking a lot of market share. The US started something called SEMATECH back in the day, which was really about how to refortify the American semiconductor industry. I thought at the time that it was the right move, and there was a lot of energy around that.
The internet hit, SaaS companies were all the rage, and people kind of forgot about semiconductors being a strategically important asset. But I think it is critically important for the United States to have as much of that technology inside, on US soil. I would say the same thing to the UK, just to a lesser extent because of the scale of the UK. But when you think about the size of the US market and the criticality of semiconductors to what the US does—whether it’s Intel, whether it’s Micron—I think we need more US fabs. It’s critical for national security. It’s also critical for diversification of the supply chain, so I’m a big believer in that as a strategy. I think it’s really, really critical.
As far as the export controls go: “We’re going to limit the chips because we don’t want China to win the race,” end quote. My personal view is that it’s an infinite game. I believe, first, in terms of the race, that there’s not going to be a winner. The race is going to be over. But you could get to a situation where a lot of the critical technologies are not US-based.
And that’s not going to be a good thing, right? People say, “Well, the cost will go down and goods are cheaper.” But ultimately—and I’m a big believer in this—for both national-security and economic reasons, you want to be at the forefront of technology because it drives innovation, but it also drives ecosystems. If you think about the US auto industry in the 1950s, post–World War II, when Detroit was the center of the universe, you had spots across Wisconsin, Ohio, and Illinois—whether it was Firestone or Bridgestone—Bridgestone is Japanese—or other companies in that ecosystem that fed into it.
11. Data Center Backlash
Data centers are kind of the same way. People look at data centers and say, “Oh, it’s a big Costco box, and there are 2 cars in the parking lot, and all of that is being driven automatically, so there are no jobs.” I call BS on that, because if you think about energy, liquid cooling, and all of the things that make the data center better, those are all jobs that can be created and done here. So I think, as a national policy, it’s incredibly important for us to be investing, A, in the United States, and B, making sure that we stay in the lead.
On the data center side in particular, it seems like a lot of the actions being taken to try to prevent future data centers feel more coordinated than not. I know it’s phrased as grassroots efforts, but it seems like there’s some coordinated function there. Do you have a hypothesis as to why there’s been this sudden, unexpected outcry about data centers from certain corners?
I think there is a fear—we may have chatted about this a bit earlier—that AI means job loss, and job loss means all these implications.
Do you think that fear is well grounded? Because somewhere it seems like it’s only creating jobs.
No, I don’t think it’s well grounded at all.
I think the electricians’ labor union specifically said, “Please don’t ban the data centers. We need these jobs,” very recently.
Completely. That’s a great example, right? Because here’s one where there may have been a stigma to being an electrician. It’s not necessarily viewed as a highly educated job, or you don’t need a PhD. It’s a highly skilled job that requires a lot of training and certification, and you need tons of them to do this kind of work. That’s very critical to the data centers.
So I think, to your question, part of the backlash is just fear. There’s a fear that my jobs are going to go away. The AI boom, for good or for bad, has benefited a lot of people, and there are a lot of people who have had no benefit from it, right? There are a lot of Americans, just again on the American political scene, for whom it’s tough to make the mortgage. Their paychecks haven’t gone up, and now they’ve got this AI thing that looks like it’s going to make things even harder.
So I think the data centers become a bull’s-eye, unfortunately, for all the things that could be bad about AI, which I think is just people holding up fake, tainted water and claiming that it’s ruining the water supply. I feel like there are other things that are just being made up about data centers as a way to try to create fear.
For sure. Unfortunately, it’s become the boogeyman for a lot of things.
I think it’s also pretty clear that there’s organized media influence around these issues as well. But I think you can have all 3 separate points here, including yours, Rene, which is that there are benefits from the construction of essentially a rapidly growing new industry that can create new technology and new jobs and create external wealth for the communities around them. But it’s on the industry to communicate that.
Yeah. On first principles, whether it was smartphones, the internet, personal computers—fill in your favorite technology—there is no downside from being the leader. There’s just not. This is maybe the most important point: There’s just no downside from being the leader.
There are second- and third-order effects that you may not like, but to be the laggard, you are having the entire script dictated to you and everything that comes with it. Look at other parts of the world that are just not the leaders in this space. Economically and socially, they’re left behind, and governments carry the large tax burden of it.
So if you’re on the wave of some technology innovation—and I would argue, to some extent, AI is a little bit of the final frontier of what can be done with silicon intelligence—of course you want to be in the lead. Of course you want to be driving that, because the benefits for society are going to be enormous.
12. Arm Outlook
What are you most excited about in the coming year or 2 for Arm?
Being in the center of all that. Honestly, I think we are—I feel fortunate every day that we are in the heart of all of this, and that we can be a participant in that ecosystem. We can help drive the innovation. We can be involved with leading companies developing leading products. We’re right in the middle of it all because, A, all the AI needs some level of compute. That’s what Arm does, and that compute needs to be power efficient. That’s what we’re really, really good at. So all those roads lead through us.
13. CPU Opportunity
I’ve been in this industry my entire career, and I’ve spent a lot of time thinking, “What’s the next product we’re going to need? Do people really need another tablet, and does it need to be 8.9 inches or 9.2 inches?” Now, the abundance of opportunity for innovation is so great with AI. I’m super excited and feel blessed to be leading a company that’s in the center of it all.
The need for chips is driven by massive change in workload, right? And we have continual, massive change in workload, so there’s no better place or time to go work on chip designs and sell to all the people working on that innovation.
My understanding of the CPU opportunity in this era is like 2 core pieces, and then future devices and robotics as well. There’s the CPU in the rack—this is the Veras and the Gravitons of the world—and then there’s the use from an agent perspective, like sandboxes and agents being able to use all of the software we already have, API calls, tools, et cetera. Do you have any guess as to the scale of opportunity? Both of these things are growing, but am I missing things that you guys are really excited about from the CPU perspective?
From the CPU standpoint, when the data center thing was exploding—let me back up. When ChatGPT had its explosion and everything was all about the accelerators, I think there was so much focus on, “No matter what the question is, the answer is the accelerator.” There’s no computing problem that’s ever been invented that doesn’t utilize and can’t utilize the microprocessor. It is the heart of everything. All roads lead through it, around it, past it, et cetera, et cetera. You look at fundamental system design, and you have to have CPUs. They just don’t go away.
They were a little bit forgotten as this accelerator thing took off. But what then became very obvious was, as more and more of the data was moving away from training—training is obviously very important—to RL, reinforcement learning, to inference, the use of the tokens, the use of the information, of course something has to do the orchestration, arbitration, and decisions around where those tokens go, right? The token factory just generates all these tokens. It’s literally: Where are the trucks that are going to take the tokens away and give them to the users? That’s what CPUs do.
Until something’s invented that says the CPU has gone away and we’re now doing it through some other mechanism, which has yet to be defined or invented, the CPU is going to be doing just fine. There’s going to be a ton of demand for it, in addition to the accelerators that generate the tokens.
The way to think about it is as a system, which again goes back to memory. Of course memory is needed, because in a computer’s von Neumann architecture, or computing architecture, you have a CPU, you have some accelerator, whether it’s a floating-point accelerator or a GPU accelerator, and a memory system. The design hasn’t changed. I think some of the focus kind of moved around.
For Arm, that applies whether I’m talking about a data center, an automobile, a robot, a phone, or wearables. In fact, as you get to the smaller footprints where more and more AI is going to take place, that’s going to be a sweet spot for Arm, because the CPU is table stakes anyway. You have to have it to do all the things that are required in the edge device.
But now we have an opportunity with our instruction set architecture to do a lot of things where you just can’t put a 50-watt GPU on your head, right? You’re going to have to do that AI processing somewhere locally.