Nvidia + Arm

Hi everyone, welcome to 16 Minutes. I'm Sonal, your host, and this week we're talking about the news that Nvidia intends to buy Arm from SoftBank for $40 billion. The news was announced earlier this month, but the deal still needs to be approved. And just this week in the news, it's been reported that some critics suggest the deal may not go through, as reported by CNBC just yesterday. The critics include UK lawmakers, Arm co-founder, others. And as reported last week in MarketWatch, given an editorial in China's state-backed Global Times, there are concerns that the deal could expose Arm to escalating political tensions in Europe and China. We talk about this aspect of the deal at the very end of the episode. Note too that I drop out briefly and come back in there. But for most of the discussion, we focus on the deeper questions of what this deal means for the tech ecosystem and more. Since the premise of this show is to tease apart what's hype, what's real in the current headlines, especially given the longer history of innovation and the broader tech trends. As a reminder, given all the companies cited, that none of the following should be taken as investment advice. Please be sure to see a6nz.com/disclosures. And our two experts today are A6NZ A6&C board partner and former president of Windows at Microsoft, among other things, Stephen Sinofsky, as well as A6&C operating partner Frank Chen, who's also shared some content on our site around AI Inside and more. Separately, if you'd like to learn more about the backstory of RISC, which we briefly mentioned in this episode, from the person who pioneered it, please be sure to check out our podcast with John Hennessy, Mark Martín Casado, and myself. I've included it in the show notes along with the news articles I mentioned. Okay, so to start us off, Stephen, can you give us some quick context on who the two players here are and why this deal matters from your vantage point?

So most people know NVIDIA because they're massively powerful GPUs that all the gamers use that are discrete x86, and then all the use for machine learning and crypto mining and all that. And then ARM are the chips in all of our phones, all of our printers, all of our routers, all of our devices. In fact, the company itself is really just a 6,000-person sort of silicon design and intellectual property company, and they literally manufacture, if you will, a design for microprocessor that hundreds of partners all themselves license, and then a few places manufacture those chips. But Arm itself is a design shop, and it's always been that, focused on this philosophy of reduced instruction set computing and a scalable architecture that goes from processors that cost pennies all the way up to sort of the elite phone processors. So this is a merger of sort of a customer of Arm that NVIDIA does for its automotive and other components and the Arm consortium, if you will.

And just to kind of grossly simplify it, reduced instruction set computing risk shorthand in this context would be low power usage, very high energy efficient type of chip architecture, which classic Intel chips had great performance, but were very slow on that side.

If you were a computer science researcher, you probably argue with the risk is essentially lower energy consumption. And CISC, its counterpart, complex instruction computing, is inherently higher power. If you dig into it, and there's a team at Wisconsin who did it, who said the instruction set has little to do with the inherent energy efficient. It's not like RISC is inherently less energy hungry. It's just Intel and ARM had very widely divergent design goals. Intel wanted to build the fastest, biggest, meatiest processors that were available because they were always plugged into a desktop PC. Or a data center PC. And ARM, they were like, look, I'm just going to design my chips to hit a power budget always. Never consume more than 10 watts, 20 watts, 30 watts, whatever it is. And that was the path that they went on. And it turned out that was hugely important for laptops and then phones. And then they kind of lucked into the biggest growth market of all technology times, which is smartphones.

That's actually— it's super important to understand the context of this deal that Intel has been on this Moore's Law. Like, I'm squinting when I— grinning when I say this, or squinting. Which is ignoring power consumption. You know, like, by and large, the focus has been on, you know, more transistors on a smaller surface area doubling every 18 months. Whereas, as Frank said, like, ARM all along has been about constraint, and just fewer transistors consume less power because they were going in these mass-scale devices where if you're going to make 100,000 or a million of custom manufacturing things you can't afford the hardware. Like if you just want to put a processor to control LED lights, you can't be $2 of the cost and you can't consume more than the light bulb consumes in energy. And it really just happens that right now the innovation is all happening in the billions of devices, not in the 100 million devices. And much more importantly to the context of this deal, which we'll talk more about, is just where Apple has been innovating in the phones. And that's what really changes the landscape of this whole deal. One other important thing to add too is that the landscape of ARM customers is the who's who of all technology companies. And then all the printer companies, all the companies that make networking gear and all of these companies, like every single technology company is a licensee of ARM. Some use it a lot more, others are just, it's become a decreasingly important part of their overall business. So at one end you have Samsung, which is building their own ARM chips, they manufacture their chips. And then you have Intel and Microsoft and Apple have all been and are ARM licensees. Licensees. Now, Intel has been an Arm licensee forever. And it's sort of a weird thing that they've always been there and yet never capitalized on it to the degree. And that was a big internal battle even within Intel.

Hmm, interesting.

So let's actually now talk about the deal and what's really specifically interesting about this news. And then we can also zoom back out into where we are in the long arc of innovation. So Steven, you mentioned that there's this whole who's who of everyone who's using Arm. The question I have is theoretically then Apple could have put in a bid to acquire Arm too, and they did not. NVIDIA did. And Apple, I believe, has like an exclusive license to, like, in lifetime, in perpetuity type of license to this. So maybe they don't care. And as we all know, Apple announced their own advancements in their ongoing chip architectures and whatnot. I'd love your guys' first quick take on that aspect, as well as then more broadly, what you really see as the thing that people should or shouldn't be paying attention to here.

This is just an interesting tidbit of history. So first, on the two companies, the acquirer and the acquired, Nvidia started after ARM. ARM was 1990, Nvidia was 1993. And then the other interesting tidbit is that ARM started as a joint venture between a company called Acorn Computer, believe it or not, Apple itself, and a company called VLSI. So they were the three companies that formed this joint venture. And in fact, the original name for ARM was ACORN risk machine. And Apple was like, no, it can't be ACORN. That's too close to a fruit. I'm Apple. And so it became Advanced RISC Machines and then eventually ARM.

That's amazing.

Yeah. So that's just sort of a funny tidbit.

By the way, on a personal history note, my dad was one of, I think, the first engineering fellow at BLSI Technology when we were growing up in the '80s and '90s. Wow.

That's so interesting. I didn't know that.

I think that the interesting thing that I think a lot of people are zeroing in on right away is that Arm was this very carefully balanced ecosystem of component providers and big chip makers and phone makers and even these big giant partners. And Nvidia was, quote, just another Arm licensee. And so now all of a sudden that ecosystem is completely out of whack because an Arm licensee now owns it and is also telling everybody, by the way, we're going to be making a bunch of chips and competing with all of you. And some people have this very academic view Well, that can't happen. You can't have an acquisition where all of a sudden there's channel conflict. And of course, this is hardware, software, and silicon stuff. So the stack of where the horizontal play is, is always moving up. And so you, you don't have these permanent divisions between firmware, hardware, peripherals, software, operating system, and applications. Like the world doesn't stay stable like that for a very long time. And there are all these people, they're looking at the deal as though only one part of this is going to adapt. And what's really happening is Apple, Apple has taken the ARM design and built so many pieces around it and then built a software layer around it. And so if you're one of these ARM licensees and you want to compete, you have to do everything that Apple is doing. Yeah, there's the ARM instruction set, but what about location and GPS and sensors and pulse ox and all of these other things that you're going to start to think of as like basically the processor or the motherboard of a device. And all of those partners have to do something. And so the worst outcome is nothing happens because then ARM just becomes decreasingly critical as a central part of it as everything gets added on top. And in many ways, that's what happens to the PC ecosystem. The part that Intel made became less important. It's starting with graphics. You should get it all in one place. And then places like modems or wireless WAN connectivity, all of that stuff just didn't get integrated in. A statement of packaging, pricing, but also technology. Like when you integrate the modem into the chip, then you get, you know, 3-week standby and all-day battery life and always connected connectivity. And all of that just comes as part of the architecture. And so the world is different now.

And the world changed, right? Because like your phone does not have room for a motherboard. It only has room for a system on chip where all of these things are pre-integrated. Like if you crack open an iPhone or an Android phone, there is this big honking piece of silicon and that rewrote the lines because prior to that, the lines had been, well, there's a CPU and then there's a graphics card, and then maybe you have a math coprocessor and maybe like you have a hardware accelerator for your camera, and they were all separate pieces and the motherboard tied them together. And so it was a pretty big redrawing of the lines of who does what and what vendor and who licenses and who benefits. In fact, Qualcomm basically willed that whole thing. Into existence. Without Qualcomm bundling the modem and the compute and the graphics onto a system-on-chip, maybe we would be living in a different bundling of those capabilities. And so they're kind of like an enabler to today's acquisition.

What's super interesting about that right now is, of course, Qualcomm, because of the way it serves multiple OEMs, is sort of stuck being unable to pull everything together, which is a position that Intel found itself into. When you want to serve multiple customers at infinite number of price points, it gets very, very tricky as to what you include, where, and how to price it, which is exactly a problem that Apple doesn't have.

This has been fascinating and so interesting, especially when you think about the history of computing and where things aggregate and fragment and kind of go through these ongoing cycles and these lines being redrawn. Coming back to NVIDIA and why they are buying Arm where Apple didn't, one of the things that Jensen said in their announcement post is, quote, NVIDIA will bring our world-leading AI technology to Arm's ecosystem while expanding NVIDIA's developer reach from 2 million to more than 15 million software programmers. So what's really interesting here is that they're not only talking about acquiring Arm, but actually putting their technology more broadly into the Arm ecosystem. Can you guys talk a little bit about that for a minute?

Yeah.

What Arm has done a fabulous job at is engineering a route to market where they reach everybody and they're kind of perceived as non-threatening. Whereas Intel might be threatening or Apple might be threatening. And so it'll be really interesting to see what happens when NVIDIA tries to hitch a ride on this route to market, which is, will that work? Is it the Trojan horse that lets them get more pieces of NVIDIA intellectual property into design wins? Or will it not work because like ARM needs to be independent and people don't want other NVIDIA stuff jammed through this channel?

Specifically using that example that you used earlier of Qualcomm had the modem. Well, today Apple's graphics are vastly superior to every other mobile graphics chipset. And so NVIDIA having really robust graphics, they have work to do. It doesn't come for free or anything, but bringing graphics and machine learning capabilities to the chipset, you know, most of the people licensing Arm will value that and be willing to take it. There's only a few people who are going to be like, well, we could do better ourselves. And that is really the thing you hope for. When the Macs with Apple Silicon come out, there are still going to be people swearing that graphics using discrete NVIDIA graphics cards are still faster than graphics on this new Mac, whatever. But it's not going to matter. The difference in power consumption, setup cost, and everything is so overwhelmingly in favor of Apple that the ARM ecosystem needs to compete with that. And so this is such a big opportunity. And the skeptics, particularly the partners who are like, oh, he's going to compete, channel conflict, should all go, oh, wait. Maybe he's going to deliver a neural chip or some equivalent that we can leverage. And that's exactly what the everybody-but-Apple camp needs right now since there's nothing else.

So we've talked packaging, obviously. Where does pricing come in?

Well, one of the success factors of Arm has always been its low price. Now, a lot of that is due to the fact that the chips just didn't do a lot. But by and large, Arm chips are a factor of 10 less costly than the equivalent computer of Intel. And ARM chips span the gamut from, you know, sort of sub-$1 processors up to $25, $30. And you double or triple it depending on how you want to do BOM calculations to get to a device. And so there is a lot of room in there for Nvidia to package more things, because if Nvidia doesn't package it, someone else is going to try. And then they have two things. One, it's going to cost a lot more because there'll be smaller number of units. Dedicated R&D and all that, plus you're gonna not get any evangelism success with developers. And Samsung and others have repeatedly shown that, like, people just don't target the APIs from one OEM, even if they're all flagship phones. And now we have phones and tablets and watches and home devices from Apple all basically running across the same hardware. Their leverage is like the entire industry was for x86, but for just their own devices. And so this cost part of it is a huge factor, and it also gives a lot of room. You know, there's gonna be a lot of opportunity to sell different styles and sizes and feature levels of those chips as well.

So let's tease apart what people are talking about in the headlines. I'd love to you guys help me probe into what might be hypey about the deal or where you think people might be missing the bigger story.

Okay. Well, the hype side of it is coming from NVIDIA is a trading stock. You're getting a lot of hype from the deal-making Wall Street stock price tracking crowd because of that. And then there's a lot of skepticism from sort of the hardcore technology side that sees NVIDIA niched into this world of discrete graphics cards and an ARM customer that wasn't as successful as it had hoped in automotive, as you've been pointing out. And then this notion that, wait, they're buying it, but they're not gonna shepherd the ecosystem that we all value. And the thing that I think is missing from the hype scale is this reality of the landscape has completely shifted. And it's a very bumpy sort of transition when these happen. When Microsoft joined with Intel and created Wintel, which is a name that came 10 years later, the idea of IBM blessing Intel really perturbed the marketplace because IBM used to make its own chips. And then they blessed just one chip. And for years, the articles were always Motorola versus Intel. And then Intel just laid waste to Motorola. And then Apple had to go bet again and bet wrong on PowerPC. And then finally bet on x86. But since the bet on x86, they've been, we should have made our own chips all along. And maybe this RISC-V consortium of a bunch of everybody but The important people can get together and make something, but that's not really going to materialize in short order because there's no motivation from the largest players in the industry to participate. And so I think both ends of the hype are sort of these visceral reactions. And the reality is really ignoring, like, this is just a step function change in basically making low-cost chips that has to happen. Otherwise, only Apple will be making them.

So I want to probe on this specific idea of whether RISC-V allowing open sourcing without licensing could change the game.

In a sense, that's sort of somewhat orthogonal to the debate, which is open source or not doesn't change the degree of difficulty. Unlike other open source things which grow organically, this is a known target, which is all of the Apple ecosystem. And by ecosystem, I mean think about Tim Cook and that Apple silicon slide that he showed at WWDC with all of this constellation of things. So the A14 Bionic, you could open source like part of it in the RISC-V. It's sort of the instruction set part of it, which is the core part of the ARM chip that is licensed. But like, that's not the hardware.

You can't ship an instruction set. You've got to ship a product.

And then they could take the layouts of the chip and open source those, something that China has tried to do. And you could do all of that, but in the 2 to 3 years it might take to go from nothing to like a chip design, Apple will have added 6 more other coprocessors on the SoC. And, you know, people can hem and haw at any given one of those. But the problem that you have with competing is that developers just get all of those, and they can call of those APIs, and they can use them all. And so anyone competing, it's not just can you make equivalent hardware or just the important part. You have to then build an ecosystem of software around that. And that's the part where to me, a consortium like RISC-V, it's sort of aiming where the puck was in a lot of ways, which is, you know, that was the battle when everything was not Apple. Like if you just look at it across Qualcomm, Samsung, and a few other people, you can sort of imagine that getting some traction at some point, like an open software foundation kind of thing. But if I were like deeply embedded in the Arm ecosystem, I would be in a panic, not because of Nvidia buying Arm, but because of Apple's announcements since WWDC. Like that's where my head would be.

Okay, so now let's talk quickly about practical considerations, both for startups or broader implications for the industry. What are the mindsets or anything that builders and people in this space should be aware of when it comes to analyzing, watching, building in this space?

So you don't really behave any differently as a Silicon Valley entrepreneur. You're building iOS apps and Android apps and web backends and so on and so forth. This doesn't change your life that much. The interesting thing to watch will be the geopolitics of this whole thing because Arm is based in the UK and NVIDIA is based in the United States. China, of course, has embarked on a long mission to basically have Chinese technology all the way down the stack, right, from the software to the operating system down to the chips themselves. NVIDIA and Arm already can't sell to Huawei. And so what does that mean in the long term? For me, that's kind of the most interesting thing to watch is sort of what are the geopolitics of this deal and who owns the future of computing as a result?

I think the geopolitics are super interesting because China has already for multiple generations, they want Chinese Wintel and they tried and failed with PCs using what was called Red Flag Linux. And then they had an x86 project. And then they thought that they could do the same with mobile, actually made a massive amount of progress on making handsets, but primarily using the ARM ecosystem and to a domestic market. And then the Huawei thing really got jumbled into the geopolitics. And so they're gonna keep going after this challenge. And because their domestic market is so large, it could very easily sort of create a unique offering, much like Japan had in the PC era It's very likely that we're gonna see like something that isn't ARM emerging in just China that is native. And they sort of have a mobile OS now, sort of an Android without the store. And so they might just continue down that path. A little bit more actionable for the founders, I think, is just this notion that, yeah, you're gonna build an Android app and you build an iOS app, but Apple's innovations are making the divergence of those much more significant. And this notion that you just get a team and they do a cross-platform solution across— the apps are gonna get increasingly less good on those platforms because Apple is just adding more and more capabilities, whether it's Apple Sign-On or how identity works or all of these things. And so it speaks to the need for the not Apple ecosystem to sort of converge on APIs at a higher level. Otherwise, they don't stand a chance at all.

Yeah, that's a really good point. Like, if you were skeptical of your cross-compiler, you know, build once, deploy to multiple ecosystems, like, be more skeptical. It's getting harder and harder for that to work, right?

And this idea of cross-platform is so appealing, but the problem is, is that your customers don't usually experience both platforms. And so basically what you're hearing is that the Android people will get a bad experience and the iOS people will get a good one. But you won't really know that because people don't use both and don't compare them.

All right, I'm back.

Sorry about that, guys. I was rebooting my internet connection. I just dropped off. Okay, so bottom line it for me.

My bottom line is it's easy to look at a static worldview and then just say this is overreach or too big a deal or too disruptive to the world. On the other hand, Jensen's an amazingly agile leader, and frankly, the mobile chip market needs to come together in some way to compete with Apple Silicon. Or what we think of as flagship devices will very, very quickly start to look very limited when they don't come from Apple.

Yeah, if you think about it in the lifetime of A16Z, which is just 11 years old now, there have been 3 big, big architectural shifts in computing. One is we went cloud native, and then we went mobile first, and then most recently we went ML inside, which is all serious software got machine learning inside. And all 3 of those trends have been hugely important And this deal is basically the sort of culmination of those three big trends: cloud native, mobile first, and ML inside. And the combination of NVIDIA and Arm, it's the personification of those trends. It will be very interesting to see how they grow together.

Thank you for joining this segment.

Thank you very much.

Thanks for having us, Sonal.