How Arm conquered the chip market without making a single chip, with CEO Rene Haas
Arm CEO Rene Haas | Illustration by William Joel / The VergeArm’s chips are in your iPhone — but what exactly does Arm do? Continue reading…
One of the more interesting quirks of the modern tech world is that there’s a really important company at the center of it all that doesn’t make anything. But its work is in your phone, in your TV, your car, maybe even your laptop — and the data centers that keep all these things working.
I’m talking about Arm, a chip design company that’s been through quite a lot these past few years. Arm designs the instruction sets for modern chips; Qualcomm, Apple, and Samsung chips are all Arm chips. Arm licenses the instruction sets to those companies, who then go off and actually make chips with all sorts of customizations. This model has been a runaway success.
So today, I’m talking to Rene Haas, the newly minted CEO of Arm. I was curious about some basics: How does Arm make its money? Who are its customers? And was it affected by the chip shortage even though it doesn’t actually make its chips?
Okay, Rene Haas, CEO of Arm. Here we go.
Rene Haas is the CEO of Arm. Welcome to Decoder.
Thank you very much, Nilay.
I am very excited to talk to you. You are pretty much the new CEO of Arm. It’s been six months, seven months?
I am definitely new. It’s been six months, but it’s been a very fast six months. I am not new to Arm; I have been with Arm for almost nine years now, but certainly, this is the latest new role for me.
There is a lot to talk about. There are lots of changes coming to Arm. You have plans to go public, and you just had your last quarter’s earnings. Arm is a fascinating company, though. I think we need to start at the beginning. It is a critical puzzle piece in the entire tech ecosystem. We all depend on Arm. It affects all of us down the line, but it can be pretty opaque. Consumers don’t really have a direct relationship with Arm, and it is mediated by lots of other companies in between. So let’s start at the beginning. What does Arm make exactly?
Listen to Decoder, a show hosted by The Verge’s Nilay Patel about big ideas — and other problems. Subscribe here!
Arm is not a well-known company, as you said, and not well understood, but we like to think we’re extremely important. Starting off, we sit inside the semiconductor world in the semiconductor value chain. You can essentially find Arm technology in almost any type of semiconductor product and / or OEM product. We are in smartphones, laptops, smart TVs. Looking around my desk here, there are probably dozens of Arm processors everywhere. We don’t actually build anything. We do designs for pieces of the product, which is intellectual property. We build that design, and rather than building a chip, we license that design to someone who’s going to build the end product. The primary product we are best known for is the microprocessor, the CPU, which is the brain of almost any type of electronic device.
We don’t build chips but license the brains to someone else who will. Based on that, there are a lot of Arm brains out there. Reflecting on the numbers for last quarter, between all the semiconductor companies and OEMs in the world, 7.4 billion chips were built and shipped with some type of Arm CPU, GPU, or technology inside. Which is a huge number. So we are in the semiconductor value chain, but we don’t build anything. We do designs, and most of those designs are microprocessors.
Who are your customers?
“Almost anybody you can think of is our customer.”
It’s funny, Nilay. In the electronics industry, it would almost be easier to say who isn’t our customer. Almost anybody you can think of is our customer. We have TSMC, Samsung, GlobalFoundries — these are the people who physically build the chips. Intel, AMD, Nvidia, Qualcomm, Amazon, Microsoft, and Google are also customers. Then, when you get to other parts of the world, we have Alibaba, Tencent, and ByteDance. Just about everybody is a customer of ours.
How do you make money from those customers? Are they just licensing designs? Are they buying reference chips off the shelf? Are they paying patent licensing fees? How does that work?
Our business model has two components. We have an upfront license fee that our partners will pay us for access to the technology. That gives them the rights to build designs using our technology. If those designs ultimately find their way into production and into an end product, we then collect a per-unit royalty based on some arithmetic math relative to the contract. So, on a high level, there are two sources of revenue. One is what we call licensing revenue. And the other is what we would call royalty revenue.
If I go to Qualcomm and I buy a Snapdragon chip, do I have to pay you? Or does Qualcomm have to pay you? Or does someone else pay you?
You don’t, but Qualcomm does. In that example, Qualcomm would report to us how many units were shipped that you bought. There is a pre-negotiated royalty rate, and they would send us a payment to cover those royalties.
Samsung makes and designs its own chips for its smartphones, and those divisions of Samsung actually have to contract with each other. If I buy a Samsung phone with an Exynos chip in it — and not a Qualcomm chip — how does that money flow back to you?
Here is the beauty of the Arm business model. Qualcomm uses our technology, and so does Samsung. When a Galaxy phone ships, we are likely getting paid. We are getting paid by Qualcomm if that product went out with Qualcomm, and in the case of Samsung, we are still getting paid. It depends on the licensing arrangements we have with partners, but typically, they are with the semiconductor arms of these companies — no pun intended. In the case of Samsung, if we signed a contract with the chip group, then the chip group would be making the payments to Arm.
Here is the last one at the top of the difficulty ladder. Apple has a very unique kind of license from you called an architectural license. They fully design their own chips, and I don’t think they use a lot of your designs, but it is Arm intellectual property. When I buy an M2 MacBook Air, how do you get paid?
Commercially, it is very similar. We have contracts with companies like Apple, and they pay us a royalty just like everyone else does.
Can anyone get the kind of license that Apple has where they get to use your IP but design their own stuff completely?
An architectural license basically gives the rights to a company to build what we call an Arm compliant processor. Anyone who does this can make minor modifications to the microarchitecture — how it is actually physically put down on the chip — but what they can’t do is modify their CPU in a way that doesn’t allow it to run Arm instructions. This is really important because, at the end of it all, we have to maintain software compatibility. If anyone is running an Arm processor, whether it’s something that we built or something that a partner with an architectural license has built, it has to be compliant to running Arm software.
We don’t have many architectural licenses because it’s hard to do. We build really good CPUs — though I am biased, of course. It is very hard to build a CPU that is different or better than what we do while still being Arm compliant. There aren’t many people who do that. There used to be more, but now it’s a smaller number because (a) it’s hard to do and (b) it’s hard to find the people to do this. These are hard teams to build. Most companies look at it and say, “If I am building an SoC (a system on a chip), and I only have so many precious engineers to differentiate my product, then differentiating the Arm CPU is probably not the best place to spend my energy.” Better places are around areas we don’t do, like a camera, modem, or IO.
When Apple or another architectural licensee ships a product, do you have teams that go and validate that it is running the Arm instruction set and that they haven’t broken the rules? Or do you just use the honor system so that you don’t piss off Tim Cook too much?
We have a set of requirements and a compliance suite that we test against to essentially verify that what they have built is Arm compliant. We run tests against that to see if it can run Arm instructions and code, that the compiler isn’t broken, or that somehow you’re not able to recognize the right instructions. The short answer is that we do have a set of compliance tests for anyone building an architectural license-based design.
I wanted to ask these questions to close the loop of this company that sits at the center of almost every modern chip — save the Intel and AMD CPUs that people might encounter — because it is somewhat opaque. I think that loop closing is where you actually validate that this is Arm stuff that works in the Arm way. I think most people don’t understand that, so I wanted to ask that set of questions and get a feel for the business.
These are very important questions. One of the things that has helped us be so ubiquitous in our history is the fact that, whether it’s an Arm implementation or an architectural implementation, they run software targeted for the Arm Instruction Set Architecture (ISA), and it just doesn’t break. There have been a lot of casualties in the CPU graveyard where they have allowed extensibility, meaning they allow customers to add custom instructions. While that might sound innovative and cool, what really makes a CPU architecture relevant in the long term is having developers know that it’s going to run. If a developer is writing a piece of code for an OEM and Arm is embedded, the developer doesn’t want to know, need to know, or even be in the position to know that they’re designing a thermostat with Bob’s chip inside and Bob has some extra instructions. You need to take advantage of that because they may not know whether another OEM device has the Arm chip that has those instructions. Leveling the playing field and making sure that the software dataset looks the same is really important. Our founders did a fantastic job of adhering to that and making it quick-principled. You can see it has really benefited us now.
You have a lot of customers that are all pretty full-throated competitors of one another in various ways. I have talked to a lot of executives in positions like yours. They all strike me as doing a lot more political work than engineering or research work. How is that split for you? Do you think that you’re a politician just keeping everyone on a level playing field? Or are you deep in the weeds of processor design?
We are definitely deep in the weeds of processor design. I mean, that’s what we do. At the end of it all, we spend a lot of time and energy on developing these CPUs and the software ecosystem, solving partner problems, and really ensuring that the products are leading edge. One of the things that is a bit different about us is the fact that we deal with everybody. We have to maintain the consistency of how we manage our relationships with partners, and that is really around the access to technology, the access to bugs, and the access to people. The world relies on Arm. As you have said, we are a little opaque. I can say on a podcast that the world relies on Arm, and someone listening will respond, “I rely on you guys? I don’t even know you guys!” We are very serious in how we manage those relationships with our partners, though maybe not so much the politics of it, to make sure that fairness is at the top of what we do.
Let me give you another simple example that you can laugh out the door if you want to. It strikes me as instructive. A couple of years ago, we talked to the people who run HDMI, which is the industry standard for how you plug PlayStations into TVs. There is this one feature of HDMI that doesn’t work well where your one remote is supposed to control everything. Honestly, they threw their hands up in the air and said, “That problem is too hard to solve. No one wants to take it on, and everyone will just screw with them. It’s never going to get solved.” Politically, I understood why that was the answer. That’s the lowest stakes of all, right? I just wanted to plug my DVD player into a TV and make the remote control work, but the politics of it were effectively insurmountable. Your politics are way harder stakes, and you’re not an industry standards body. You are a company that reports earnings, that wants to IPO, that has to grow those earnings for shareholders. How do you manage that kind of tension?
We are known as the Switzerland of the electronics industry. We don’t try to pick winners.
We really try to stay as neutral as possible. We are known to be the Switzerland of the electronics industry, which is not a bad parallel. We don’t try to pick winners. We are involved in the ecosystem of ecosystems. If you start at the lowest level of the semiconductor chain — GlobalFoundries, Samsung, TSMC, Intel, all the people who build chips — you have to work with all of them. We have to make sure that our technology is going to be able to be built on every semiconductor process in the world, which requires investment across all of those partners. Then, all the way up the stack, when you think about Android, Linux, Windows, and all the major operating systems that we support, we have to make sure we are there, too.
We really try not to play favorites in terms of advantaging one versus the other. I’m not sure how we would do that at the end of the day, but what it means is that we stand on the shoulders of a pretty big ecosystem. I say it’s an ecosystem of ecosystems because there are design tools, fabrication processes, software operating systems, and middleware. We often work with interest groups, and we create them. We don’t really work so much with standard bodies, but we do work very heavily with all the players that I mentioned. We make sure that we understand everything they are trying to do from a roadmap standpoint to ensure that we are going to be as compliant as possible. At the end of the day, we license technology to someone building a chip. That person building a chip wants the broadest opportunity that they can have where they build it, have software that runs on it, and control how it ultimately goes into production.
That is a very idealistic answer. I want to come back to it in the context of Arm being a publicly traded company because I think that might change the dynamics a little bit. Let’s hold off on that for one second, though. I want to get through some of the additional basics of understanding Arm first. How many people work at Arm?
The last head count number I saw was probably around 5,800, but with contractors, it is probably north of 6,000.
How are they structured? Is it all chip design, or are there lawyers, too? Are there patent attorneys and chip designers in a 1-to-1 ratio? How does that work?
It is definitely not 1-to-1 lawyers to engineers, I will tell you that much.
Some companies are definitely 1-to-1 lawyers and engineers.
We are nowhere close to that. We are mostly engineers, most of them in the UK. Arms headquarters is in Cambridge, a couple hours north of London. We have a few different engineering sites across the UK, France, and different parts of the Nordics. We have a couple design centers in the United States — in Arizona and Texas. We also have a fair amount of design engineers in India: Bangalore and Noida. Most of our employees are engineers. Our legal department is pretty darn small. If I just have the thumbnail ratios, the proportion of engineers to lawyers is pretty large. We have done a very good job of understanding how our licensing model works and how to protect their intellectual properties. We don’t have a huge legal department, but we do have a lot of engineers because these products are really hard to build.
Who reports to you? How is your team structured?
I am CEO, and I have the classic chief financial officer, chief people officer, and head of legal reporting to me. We are organized around business units, so we have a vertical line of business structure. We have automotive, internet of things (IoT), infrastructure, and client. Those GMs report to me. I also have a chief architect, our head of sales, and our head of engineering that also report to me. As I’m listing it, there are a lot of direct reports, but that is what the team looks like. It’s mostly people oriented around engineering in the business, with the classic functions around people, legal, and finance.
Here is the classic Decoder question that I always ask everybody. You have been at Arm for nine years, and you have been the CEO for six months. How do you make decisions? What is your framework?
I’m going to have to cheat a little bit and say that I had dinner with Tony Fadell, who I know you interviewed a while ago. I like his analogy of opinion-based decisions versus data-based decisions. I would say that the more white you get in your beard and your hair, the more comfortable you get with opinion-based decisions. Pattern matching starts to come into play. History repeats itself because humans repeat themselves. We are humans, and humans make the same successes and mistakes.
I’m a mix of opinion and data, and the more experienced I get, the more I rely on intuition. Experience helps. For those folks who are football fans out there: why is Tom Brady still playing quarterback at 45, even though, physically, he is playing with guys half his age? Because it is hard to fool him, and he’s seen it all. Our business tree is far more complex than professional football, but it helps a lot. The short answer to your question is that I rely on both, but I probably move faster now than I did when I was younger. Opinion helps me a lot more than data because my gut is more intuitive.
My understanding of Tom Brady is that he doesn’t eat any tomatoes and eats avocado ice cream every day. Is that on your list?
He is kind of like Benjamin Button. Every time you look at him, he looks younger and younger. I do have weird dietary habits as well. Folks in the UK will know, I’ll have yogurt and granola every day for lunch. My assistant there will almost preemptively say, “I’ve got your yogurts in the fridge. They have your name on them.” I have seen this from other leaders — not that I would put myself in that category of being super successful — but part of it comes down to reducing the number of decisions you have to make. I have always found that it helps me, personally.
“I wear the same attire. I eat the same food. It’s stuff I like, and it’s one less thing to worry about.”
I wear the same attire. I eat the same food. It’s stuff I like, and it’s one less thing to worry about.
I have definitely found the same to be true with myself over time. It helps that I pick really cool clothes. That’s been my trick.
I would not say I’m there.
Let’s talk about a big decision now. Arm was famously caught up in this SoftBank Vision Fund turmoil, where SoftBank raised billions upon billions of dollars for something called the Vision Fund. They invested in a lot of companies, bought Arm outright — which is a big deal — and then tried to sell it to Nvidia when the Vision Fund got a little shaky. The industry basically lobbied as hard as it could against that deal, saying, “We do not want Nvidia to own this core CPU technology processor, this design technology. That’s Qualcomm. That’s Apple. It’s down the line. Don’t do this.” The government said they would block the deal, and SoftBank backed off.
You arrive as a new CEO. You say, “We’re going to take the company public.” That’s a big flip. Your predecessor was adamantly opposed to taking this company public for that reason we were talking about earlier: the pressure of having to grow revenue puts the model of being a neutral and fair provider to everyone at risk because you could cut special deals to increase revenue. Those pressures will arrive, but that is your decision. How did you make that decision?
We announced the change when the Nvidia transaction basically came apart at the end of last year. After I took over in mid-February and our fiscal year ended in March, we were finally able to talk about our financial results, which we refrained from doing for a while. We were very quiet during the Nvidia period. When we announced our revenue for the year, we had a record of well over $2 billion. We had done $2.6 billion — we had never done anything over $2 billion in the past. We had an operating margin of nearly 40 percent, but people thought we were losing money because we were very quiet. If you fast-forward to this quarter, it was even higher than where we ended last year: north of 50 percent and $700 million revenue, $450 million being in royalties.
Several of the calls I had with analysts and reporters were like, “Where did this come from? What kind of funny math are you doing? Is this some pro forma equation?” But actually, we knew we were doing okay. Not long after SoftBank bought Arm, we reorganized and created two business units. I took over what was classic Arm, and we began to pivot toward other markets. That pivot was not only a business model but also products. We knew a couple things in the data center were happening. We knew TSMC was getting really good at processing, and we knew we were making good headway on software workloads. We felt if we advanced our investment into some specific instructions, such as SME and SVE — these are vector extensions for specific workloads on hyperscalers — we could make some hay in terms of the hyperscaler market.
Just to be clear for everyone, that is the cloud computing market. That is your Google cloud, AWS stuff.
That is your cloud. AWS is a significant partner for us. They announced Graviton2 along with some fairly eye-popping numbers relative to the 40 percent improvement in price performance over other architectures. So we have diversified our business by not only developing different products but also by addressing it through different parts of business model strategy. We knew our business was going to be okay. All of the financial results you are seeing now, which are terrific and the team has done a fantastic job on, really come from work that was done a few years ago. You’re not going to be able to see royalty results on units happen overnight. We develop IP, and that IP has to go to a customer. They have to build a chip, and that chip has to go into the product. The product then has to be qualified. All of this can take three to four years. We feel good about where we are going. We also feel really good in the areas that we have been investing in, such as cloud, automotive, and IoT. These are large secular growth areas that I think we are very well positioned for.
Let me push on that. I spend a fair amount of time on CNBC; I am a tech journalist, and it’s a business channel. I’m always struck by how much CNBC is for investors when they end every interview with “what is going to happen next quarter?” I appreciate that. It’s their job, and they are really good at it. But here, you are describing multiyear bets, right? You made some bets two years ago that are paying off handsomely now. Do you think that quarter-to-quarter investor pressure will change how you operate the company? Because that seems like the risk there.
I think any time you are a public company, that is just part of how the world operates. I can’t talk too much about what life would be like as a public company, though. I am under pretty strict orders to stay in my lane on that.
One of the few lawyers in the company is about to run into the room.
Exactly, so I can’t really say too much about that. I can say we are pretty confident in terms of the secular growth we are seeing in the markets we’re involved in.
You just expressed some confidence in automotive, IoT, and cloud. The cloud market exists, and it’s growing. Arm has made that move, and it definitely seems like it’s going to happen. I talk to a lot of car CEOs, and we spend a lot of time on The Vergecast talking about smart home stuff and era things. Those markets aren’t quite there yet; they haven’t really made the turn. Is that something that you can drive as the CEO of Arm? Or is it more like, “We are just going to have the designs and products ready for when cars actually become network computers on wheels”?
When people think about electronics in the car, they automatically go to autonomous driving that you see with crews and such. That is just one dimension. Think about what goes inside an automobile and the amount of processes in a car. You have your instrument panel, which is all digital and completely computerized, and almost all of that runs on Arm. The power train, anything involved with the mirrors and the brakes, all of that is quickly moving to Arm as well. There are a number of things happening in the car that are moving toward Arm that have nothing to do with the instrument panel or the autonomous aspect.
In the power train, you have many of these older electronic control units (ECU) that were standalone units. They might have an old proprietary microcontroller in them, and they don’t communicate with the other parts of the car. All these ECUs are being redesigned. Inside a car, there are maybe 50–70 ECUs. There are lots of them, and each one of those might have had an old proprietary microcontroller that had no connectivity, no memory management unit, that didn’t talk to other parts of the car. So the car becomes a highly networked piece of equipment. Then you add on autonomous and advanced driver-assistance systems (ADAS), which is an area for Arm to grow.
I think, for us, the car is a bit of a sandbox of multiple technologies: instrument panel, power train, power train / ECUs, and autonomous. Autonomous, by the way, is a huge opportunity for us. Going back to the data center arena, what really matters in the car in terms of compute is performance and efficiency. You can’t have a server in your trunk running off an EV and be successful at it. Some of the cars today, that is kind of what they are. They are kind of like a server in a trunk. That is going to get better over time, so we are very positive about the automotive market. We have been growing very fast there.
So are you going to take some of your engineers and say, “Go figure out ECUs, an engine controller, or a body control module that will work across cars so that people at Nvidia can come and license that and sell it to Ford”?
That is already happening. What matters there with these CPUs that go into automotive? Well, efficiency, power, and functional safety all matter. You have to have all of the redundancy to make sure that this can run in a safe way. Some people do it in software, through compute libraries, but most would prefer to do it in hardware because it is more secure and effective. We have developed automotive processors and graphics processors that have functional safety embedded. We weren’t doing that in the past. We would basically roll out a general purpose thing to use wherever you want. That was one of the things we did a big change on in the last number of years. Automotive is going to be a very large market for us.
We are talking about the future. Let’s bring it back to the present for a second. We have been in a chip shortage for a long time, which may or may not be coming to a close. Intel and Nvidia just had pretty bad quarters. Intel is saying they are actually going to raise prices. Where do you sit in the chip shortage? Is that something that affects you right now? Is that something you see coming and going? You are kind of divorced from the actual hardware piece of this.
We are divorced from the standpoint that we don’t build anything, but we are very linked to it in the sense that our royalty model is linked to how many components people ship. For sure, we have an eye on it in a very big way. Back to the earlier discussion, we are pretty diversified in terms of the end market. The other thing that is happening is that more and more CPUs are being used in these SoCs. Where application processors for mobile phones might have used one CPU, the cluster for compute is now nine or 10 different CPUs. You have 10 CPUs, and that is just the apps processor. Then you take into account the touch sensor or anything with the display or camera. We are seeing that, even though units have been softening in some markets such as smartphones, we have been shielded by it from the standpoint of where our product goes. Broadly speaking, semiconductor shortages and boom bust — that is the nature of our world.
I do think this one is a little different. First off, it’s not all technology, and it’s not all areas. IoT is strong, industrial is strong, cloud is strong, and different nodes are strong. Sometimes the older technology — 14, 28, 40 nanometer — is kind of hard to get. All the devices in these complex systems require a mix and match. You only need one thing to not be ready to have a problem. With COVID and the world not as flat as it was three or four years ago layered on top of that, it doesn’t take much for one supply chain to get messed up. Next thing you know, we’re short on a product like toilet paper. And one would say, “How are we short on toilet paper?” This is an aside, but I remember reading into that, trying to understand a bit more about the problem. These toilet paper factories had a very fixed set of equipment that did commercial toilet paper for industries and businesses or did residential toilet paper. When we all went home and stopped going into the offices, shopping malls, and movie theaters, the demand for commercial toilet paper versus consumer went out of balance. That is why the toilet paper thing came about.
I did not.
He was on here. He told us about the toilet paper thing, and then we talked about chip and LCD manufacturing. He said they had too many SKUs. They were selling too many kinds of toilet paper, and he told them to reduce it. That solved a huge part of the problem.
Right. So now, think about a car that needs all kinds of diodes, capacitors, resistors, and thermal sensors. It’s not just the fact that three nanometer is hard to get and a fab is hard to build. It’s all of it. You can tell by the color of my hair that I have been in this industry for a long time. I have been in business cycles where people start putting the brakes on R&D and slow projects down. They don’t do new things, and they stop innovating. I am not seeing that this time. I think that is because of the digitization of everything — the super high demand for electronics products and how it changes our lives. There is so much innovation going on. What is our indicator of that? Licensing and new design starts have never been better for us. It’s broad across all markets.
What you are talking about is one of the reasons that R&D is picking up the pace as much as it is. The world isn’t as flat anymore with the promise of globalization being reconsidered broadly across the globe. Here in the United States, we just passed the CHIPS Act to really incentivize chip manufacturing and all kinds of design investments. TSMC is building in the States. Intel just broke ground on the new fab in Ohio. Do you see that factoring into, “Okay, all these countries know we are heavily reliant on a handful of fabs that are in Taiwan. That is a geopolitically hot area. We need to move these critical dependencies into our countries.” Is that something that you play a part in, or is that something you are just watching from the sidelines?
We are definitely involved in helping the dialogue. Wherever we can help by talking to political officials in any country to amplify the need for this, we do. It’s not just an Arm issue; it’s an industry issue. A single point of failure for anything you are doing is not a good thing. COVID exposed a lot of things, and people had their eyes opened, like “Oh my gosh, this is a real problem.” I give a lot of credit to folks who were driving the CHIPS Act activity because I think it is very important.
“50 years from now, there should be world-class fabs on every continent. We shouldn’t have to worry about geopolitical concerns.”
It wasn’t just important for the United States, but 50 years from now, there should be world-class fabs on every continent. We shouldn’t have to be worried about geopolitical concerns for something that is just like oxygen for how the world operates.
One of the realities of how the world operates now is that the bleeding-edge process nodes are mostly controlled by TSMC. That’s five nanometers, three nanometers. They are way ahead of it, and they are in control of it. That is theirs right now. Very few other people can compete on that level. That is your smartphones, that is the bleeding edge of tech. The older process nodes — 40 nanometers, 14 nanometers — that go into cars are deeply constrained, and no one is going to build those fabs anymore. That is not a good investment. How do you see that playing out? You have this tight constraint to TSMC at the bleeding edge. Everyone wants to build capacity there, but no one wants to build capacity for the old stuff. At the same time, all the governments in the world are like, “We just need capacity in our countries.” What is the dynamic there that you see?
Full disclosure: I am not a manufacturing expert. But what I can tell you is that we see a lot of work being done relative to how you can transition certain fabs for new work. It is exactly as you said. People are usually going to be a bit averse to plowing a lot of money into end-minus three fabs. You start to look at, “Can I convert fabs that we are building to a certain process technology and move that to being a logic fab? Are there memory fabs that can be converted to logic fabs?” But it is a very complicated problem because there are only so many factories in the world and only so many people that know how to build these things. These companies are public companies. They have to make money. It’s a very complex matrix.
Thirty years ago, Japan was world-class and had logic fabs all over the entire country. That has essentially wound down to almost zero. Japan is looking at ways to reinvent that to get more fabs back online. I think you will see more of the CHIPS Act type of work that was done in the United States across other countries. Korea is mostly one fab, but I think it also has a tremendous amount of capability. It is a complex, scary problem, though.
How dependent is Arm on TSMC? It feels like a lot of companies are recognizing that they are really dependent on one company. It is the industry leader for a reason, but there is a lot of dependency there. Do you think about that dependency as you are designing new processor types or architectures?
We work really closely with TSMC. I would answer it by saying all fabs are important to us, going back to my Switzerland comment.
There’s your politician, right?
Yeah. There’s the politician. Seriously, they are all important. In an ideal world, we have lots of people who can build all the leading-edge technology. Generally speaking, in smartphones and certainly in the data center, that is usually on the leading-edge processes because people are really trying to squeeze out the most performance that they can — less so in automotive and industrial. Any fab doing a leading-edge process is massively important to us.
On the flip side, Intel famously was an integrated designer and fab company for the longest time. They basically got rocked by the TSMCs of the world. They have a new CEO now, and they are breaking that open. They are saying, “Okay, we are going to open our fab to other people.” You just said Intel is a customer. Is that something you are working with them on? Like, are you saying, “Okay, you are going to do fab services for other chip makers. We have Arm designs, and you better be good at making them”?
I’ll put my politician hat back on for a second. We would love TSMC to be building lots of Arm products, which they do. We would love Intel to do the same.
Are you optimistic about Intel being able to do that?
I think Pat [Gelsinger] has done a lot of the right things. He has a hard job, but at the same time, I think he has done some very good work, and we are very open to doing more with him.
This is going to be another politician’s answer, but I have to ask. On the CHIPS Act side, the bill passed, got signed, and everyone is excited. Intel immediately comes in for a lot of criticism because it announced it’s cutting capital expenditures and increasing its dividend instead of plowing it into fabs. Broadly, do you think that is the right move? What kind of timeline should we be looking at for fabs in the United States?
I won’t second-guess Pat’s decisions on how he spends his money. Back to the comment you made, we were talking about a single point of failure. I think we need to move fast and get these facilities started as quickly as possible. We need to have them online as quickly as possible and build more redundancy and have less of the exposure to single points of failure. I think we need more fabs.
What kind of timelines do you make decisions on? It seems like, if a bill passed today or someone announced a new fab today, we are not going to see it for five years. How do you think about your timeline of decisions?
I tend to look that far out. I’m still learning six months into being CEO that we have lots of responsibilities. One of the biggest responsibilities I have is to think about problems five years from now. With the way our business model works, the time that we design a product to the time we see revenue is quite long. We are thinking about where the opportunities are, where the investments need to be made, where the threats are, and where the puck is going relative to system and software design. We think about those things all the time, and we talk about them all the time. Most of my mental energy is spent thinking about what the world will look like in 2025 and 2026, as opposed to next quarter.
There are some long-term bets there that are much more focused on the consumer, rather than replacing the server in the trunk of a car. The hottest trend right now as it pertains to Arm taking over things is that Apple shifted from Intel to Arm-based processors, their own M-series processors in Macs, and that has been an enormous victory for them. They have increased the battery life and the performance. These are now basically the best laptops you can buy because of the chips in them. Is that something Arm can help the rest of the industry do? Or is it, “Here are the designs. Hopefully Qualcomm can make a chip that lets Microsoft and Dell do this as well”? Where do you see that relationship to the industry from your perspective?
I think there is more that we can do and probably will need to do going forward. It’s not so much because we need to help the industry but because these products are really complicated to build. Let’s talk about the architectural license for a second. There was a belief at some point in time that you needed to take an architectural license to build a better CPU to compete against Arm. The IPC instruction per clock of a microprocessor is only a microfragment of what can really tip the design. You think about the memory subsystem, the interconnect, the size of the caches, how you interface into other parts of the SoC, and making sure the SoC is in a multichip package where you have a die-to-die interconnect.
There are some customers that are really good at that stuff and can figure it out, so they don’t need a lot of our help. But increasingly over time, there are going to be sections in the industry that will benefit greatly from Arm helping them in doing more. It is an area that we look at very closely.
“It’s very clear that the demands of doing systems on chips are just going to get harder.”
There is opportunity going forward, and it’s very clear that the demands of doing systems on chips are just going to get harder. To get the kind of performance around the product that you just mentioned, it’s going to be more than just throwing the thing over the wall and hoping it works out.
From your perspective, it kind of doesn’t matter who wins as long as they are all Arm chips, right? If Apple takes 100 percent of laptop market share, that’s great for you. If Microsoft, HP, and Qualcomm can figure out a competitive Windows on Arm laptop and they take 50 percent of the share, you still get paid. Does it just feel like no matter who wins, you also win? Are you thinking, “Man, we have to get better at this because Intel is still taking a big chunk of the potential royalty pool from us”?
The way to think about it is if the products are shipping with the Arm ISA, then it’s a good thing for us. If there is an alternative device being used, it means it is not using the ARM ISA, so that is less good for us. When I was in my previous role, I had a little moniker under my name on the website that we did with the PR team that basically said “wherever computing happens, Arm will be there.” You could stand back and say, “Oh, it’s my thermostat or my microwave, Polycom, smart TV, laptop” — whatever is it. Yeah, that’s Arm-based. That’s pretty good.
You just said wherever computing has happened. But where computing happens for most people is their laptops and desktops. I have heard about Windows on Arm for 13 years, and it has not happened. Are you just content to let Apple eat the industry? Are you saying “we have to push on this” because there is someone else’s instruction set — namely, Intel’s — that is still dominating in that one extremely huge industry?
We are very aware. Just backing up for a second. I was at a previous company working on Windows RT. Back on your 13-year journey, I was the GM of that product line. I was with Nvidia before I joined Arm, and that was my product line. We have made massive progress on this. I think we’re like the borg that’s not going to stop. The tipping point is really there, Nilay. The very first press interview I did on the Windows RT laptop, I remember meeting with a set of reporters. The first question they asked me was, “Does it run iTunes?” We were like, “Ah, shit. No.” It did not run that.
It did not.
We didn’t have a port, and Apple wasn’t helping us. Fast-forward to 10 years later, streaming audio is what’s there. No one really cares about the apps. If you think about all the apps that run on a PC or on a mobile phone, it’s kind of hard to say “okay, what are the native apps that don’t really run on Arm?” They are almost all there. Stay tuned for the PC thing. We are making great progress. You can see by the other ecosystems laptops what the capability is. I think that was a wake-up call to the industry in terms of what the potential of the capabilities might be. We’re going to keep at it.
That wake-up call is due to a very rich company spending a lot of money on a custom riff on your instruction set, right? Those are their chips. Are you saying we can, as Arm, develop products that are as efficient in performance per watt as the M-series chips and then sell those designs to Qualcomm? Or is it down to Qualcomm — which I believe also has an architectural license — doing it themselves?
I am very confident that the Arm implementations can deliver an extremely compelling performance to watt story. I have 100 percent confidence in that. You don’t need an architectural license to do that.
What is the timeline?
I can’t give you a timeline in terms of forward-reaching stuff. Again, my team of two lawyers would kill me.
I’m trying, man.
You are trying. You’ll just have to stay tuned.
I’m doing my best. Your job is to dodge. My job is to keep asking.
You’re doing a good job there.
The other side of that is with laptops, desktops, and even car makers. I go and sit in a Rivian, and I get told about the GPU and Unreal Engine driving the car. Graphics performance has become a key differentiator with all kinds of products across this industry. You have new GPUs with a great name, Immortalis. Previously, they were called Mali. Apple is not using your GPUs. If you think about the big GPU vendors, they are not really thinking about performance per watt, are they? With a high-end Nvidia card, you just need a nuclear reactor in the basement to run that thing, and that’s great. It looks great. How are you going to win in GPUs?
We are actually the No. 1 shipper of GPUs on the planet when you just go by units. I remember that, when we were in the midst of the Nvidia acquisition, [CEO] Jensen [Huang] made a key point of that. We are going to stay true to our vetting — performance per watt matters. We are not going to venture out and do hundred-watt TDP GPUs and try to play in that space. We are going to be in the one-watt range and try to optimize in that envelope. Then there are areas where you could start to do more things in terms of multiple GPUs.
We are starting to see people looking at doing machine learning extensions and potentially doing those kinds of things inside a GPU. It’s kind of interesting because the GPU can benefit from some level of AI and machine learning to do shader drawing in a much more efficient way. At the same time, you can start to work or move ML workloads onto the GPU. There is a lot of interesting innovation that is going to be able to take place on our GPUs, and we are investing heavily in that area. To be clear, we are going to stay in an area where performance and efficiency matter.
Yes, you ship a lot of GPUs. It’s part of the SoC package for a lot of your customers. The performance per watt, from your customers that have their own GPUs or their own GPU extensions, is ahead. Is that the same deal where it doesn’t matter at the end of the day because they are shipping an Arm Instruction Set Architecture and you are still getting paid? Or is it, “We have to be competitive here because otherwise they will just move off entirely”?
GPUs are a little different than CPUs because of the fact that the APIs are extractive. Performance per watt really does matter a lot. We have done a lot to level the playing field, and with Immortalis, I think we are now ahead in a number of areas. The other thing that is very critical with these GPUs is efficiency in terms of interface between the CPU and the GPU. That is also an area that we are investing in heavily, but we have to be on our toes in the GPU market. There is no question about it. It’s very competitive, as you said. You have people doing their own, and you have other third parties doing products. At the same time, it is an area we are very focused on, so we will invest. I think this machine learning AI vector going on with GPUs in the future is a huge opportunity for us.
GPUs are a big investment. You have to spend a lot of money to compete and win the designs. Is that the sort of investment you are making because, over time, you can win those products back from the other custom GPUs and extract more of a rate? Or is it that, just to be competitive at all, you need to have competitive GPUs?
I go back to machine learning and AI. If you have a heterogeneous compute system, where you have a CPU, GPU and NPU, I think there will be a point in time where the compilers are going to be smart enough that they can potentially point to parts of the code that could run better on a GPU versus a CPU. If you think about the entire cluster as an entire subsystem, I think there is a long-term benefit to that. For us, particularly when we are designing a system on chip, we are trying to do all kinds of significant tradeoffs. Sometimes I don’t want that workload to run on the CPU because it is going to be sucking up unnecessary power and I might have transistors in the GPU that I might be able to use if I’m not drawing.” We believe the GPU is a pretty critical strategic component. Over time, we think it’s going to be critical in SoC for things other than just drawing triangles.
Maybe zoom out just a little bit. I feel like I can identify the GPU competitors. They are obvious in one way — I mean, one of them just tried to buy you. The CPU competitors are much harder to identify. Who are your competitors?
On the CPU side?
I think there are probably only two choices. It’s biased, but I don’t even think there are really choices. From an ISA standpoint, you have x86. If you want to build an SoC on x86, there are only two companies in the world that will do that for you: AMD or Intel. Then there is RISC-V, which is a completely different part of the stratosphere in that it is open source and lots of different versions exist. RISC-V is an extensible processor, which means its strength is its weakness. In my opinion, the extensibility will lead to fragmentation. I think that has hurt them in terms of getting any kind of software ecosystem. It’s really hard to look around and say, “What is a major software ecosystem that runs consistently on a RISC-V processor?”
There is none. So, where do we see RISC-V today in a system on a chip? It’s in a deeply embedded part of the chip where the external programmer doesn’t know it exists. Here’s an analogy I can give. If you put something in your microwave and hit start on the 30-second timer, that display is probably Arm because there might have been a little piece of open-source application code that ran on it. The timer that actually turns on the oven and turns it off after 30 seconds is probably RISC-V. Those are really the only two options out there. I mean, underneath all that are smaller companies, but those companies really haven’t had much traction in market share.
Do you see a world in which Arm and RISC-V are just completely obsolete x86?
That’s a hard thing to call. x86 has been around for a long time and has a very large installed base. I certainly wouldn’t want to do anything to disparage what they are doing. I think the challenge with x86 is the fact that it only comes from two companies. That, in itself, is a bit of a limiting factor in terms of how wide it can go.
Those two companies, Intel and AMD, have their own fabs. There is TSMC as well. There is that other manufacturing piece of the puzzle, where it seems like those manufacturers over time are going to say, “Look, we are really good at Arm. We are going to get good at RISC-V. We don’t want to be good at x86 anymore.”
I would even argue that they are already really good at Arm. It’s going to get difficult. There is no doubt about it.
Are there any other competitors on the horizon like RISC-V? I mean, RISC-V is another one of those things like Windows on Arm. I have heard about it for a decade, and it seems like it’s coming up in some of these applications you are talking about, but it doesn’t seem to be scaring you very much at all.
It’s all about software. At the end of it all, it’s about having a rich developer ecosystem that is able to tap into writing software. For us, we think there are 15–25 million developers who write on Arm. There are some that know they are writing on Arm because they take advantage of the instructions set, but there are others who don’t because it is abstracted away. The compilers and open-source libraries are all there, which makes it just easy to run.
Case in point: the smart TV. If you are trying to write a new menu for a smart TV and trying to link into the version of Android that is essentially the kernel underneath, that is Arm. That has all been optimized for Arm. It’s a big advantage in terms of self-reporting. The libraries all exist, and the optimizations have been done. Let’s just say you tried that on RISC-V. If company A adds 17 new instructions to make their RISC-V thing look a little bit different, the developer is not going to know that. How is the developer going to even take advantage of that? RISC-V probably ends up being reduced to the lowest common denominator.
We had Qualcomm CEO Cristiano Amon on the show a couple months ago. I asked him the same Windows on Arm question. He kind of pointed the blame at software developers. He said, “Look. Microsoft, Adobe, and the other big independent software vendors have to get on board. They have to make this stuff fast.” You just said it’s all about software. Do you spend time with those companies saying, “Look, you have to shift your focus to the future, which is Arm”?
When you say those companies, do you mean the Microsofts or the Adobes of the world?
Yes. We spend a lot of time with both. He’s not wrong in that, ultimately, that is where the battle is to be won. You have been watching it for 10, 12 years just like I have. It’s much closer than ever because not only is there a lot more work being done on the native apps but also a lot more of these native apps are already being written for Arm.
Look at the two different OSes in the world. That other OS is using their name in code. Look at all the Microsoft apps that just run on their system. They have all been ported that way so all those Microsoft apps run on your phone. We are not that far away. I think when there are a lot of fairly good CPU products on the market that will be competitive with the other guy, the tipping point will be there. To your point, that product proved you don’t have to compromise performance in a form factor with game-changing battery life.
I always ask people what’s next for their companies, but I am going to ask you a more specific question. You announced that Arm would go public in March of 2023. That is coming up soon. Are you still on track?
Unfortunately, we are now in the part of the process where I can’t say much.
All right, fine. What’s next for Arm?
We are going to keep investing in the areas that I mentioned. We think there is really strong growth there with some of the themes that we talked about earlier: those complex systems on packages and complex designs. We are looking really hard at the ability to do and provide more to the industry because I think there is a huge opportunity there.
Rene, thank you so much for coming on Decoder. We will have to have you back. I want to do a full hour on Windows RT because I lived through that from a very different perspective.
That will be PTSD for me.
This was great. Thank you so much.
Decoder with Nilay Patel /
A podcast from The Verge about big ideas and other problems.