Steven Sinofsky on AI PCs, NVIDIA, and the Future of Computing

Theo Jaffee speaks with Steven Sinofsky about the next generation of personal computing and the growing role of AI-native hardware. The conversation covers NVIDIA’s entry into the PC market, Microsoft’s strategy for AI-powered devices, Apple’s hardware roadmap, and the long-running tension between backward compatibility and platform reinvention. Sinofsky explains why AI may fundamentally change how personal computers are designed, and why local inference could become increasingly important as AI workloads grow. Along the way, they discuss Windows, Surface, Arm processors, Apple Silicon, and what the future of computing might look like as AI shifts from the cloud to devices.

• 0:05Speaker 00

  Having lived through like a half dozen component shortage things, you just sort of wait them out. And you don't let some local max or local men determine the future. This will all correct itself in short order. This world where you're all gated on dollars per token is a thing that's going to move to your own device, which is exactly what happened with all of computing. Anytime there's a resource constraint that you have to pay for, it moves to your device and becomes free. AI introduces yet another opportunity to change that dynamic for the PC to have it be forward looking, not backward looking. And I think this is incredibly important opportunity for Microsoft and for the industry

• 0:50Speaker 00

  as a whole.

• 0:51Speaker 01

  Live in the Situation Room with Steven Sinofsky, who might have been like the first ever guest on MTS back when we were still doing test streams, I think. He was the first person I interviewed on a test stream. He was the president of the Windows Division at Microsoft. He created the Surface program at Microsoft, which we have some very interesting news about today. We're thrilled to have you on, Steven. Welcome to MDS. Welcome back.

• 1:20Speaker 00

  Well, thanks so much. Good to see you. Hi, everyone.

• 1:24Speaker 01

  Yeah. First question would be Nvidia and Microsoft and ARM and a few other companies just announced something very interesting at Computex. What exactly did they announce? And what does it matter?

• 1:41Speaker 00

  Sure. Well, just so folks know, because it hasn't, it doesn't get in the news much, but Computex is this big giant trade show in Taiwan. And it's, it's the weirdest show because it's like this total inside baseball.

• 1:56Speaker 00

  You know, silicon supply chain show. And normally you never hear about it. Like, in fact, I never went to it even. I wanted to, but it actually turns out it was always right around the same time as a big Microsoft sales meeting. So I never went. But you could think of it as the ecosystem show for everything it takes to to build a a computing device of any kind. Totally wild, but Jensen in his keynote last night did this incredible slide where he walked up and down the whole length of the stage, pointing to partners that he was very excited to be there. And I would bet anyone that anyone watching would have no idea of the companies he was pointing at.

• 2:38Speaker 00

  Like, these are, some of them, like, half of the ones he pointed to as kind of being entertaining were were just names of companies in traditional Chinese. And and so you didn't even, like, you don't even know what they are. But it's it's an incredible show. It's just wild because it's such inside baseball about components and peripherals and chipsets and assembly lines and and it's deals and deals and stuff. And every 10 years or so, it jumps into the mainstream, but never like the past 24 hours. Just, you never see that. And actually, it was a lot like, I think it was two years ago, Jensen keynoted CES.

• 3:18Speaker 00

  And I I'd been to 40 CESs and I'd never seen one with such a broad media reach. Yeah, Jensen's like Taylor Swift of the tech industry. So true, yeah. It's an incredible level of, I it just speaks to the awareness of tech and then the awareness of AI and what Nvidia has done. Because, I mean, like, I, you know, it was bigger than, like, CES Xbox was like a sideshow compared to this. Wow. Yeah. And he's huge in China, too. Well, the show is always huge in Asia broadly because most of the companies are there,

• 3:59Speaker 00

  whether they're in mainland or in Taiwan, or in Vietnam, or Singapore. And that's sort of the origin of the show. Like, this show was like, go there and speak in Asian language for forever. So fascinating. So the big announce, though, I mean, look, there's a zillion things going on, but the one that rose to the top was Nvidia announcing what they, they called the RTX Spark Superchip, which is a mouthful, before the show was broadly called N1X. And like, that's sort of the, what it is, is just, it's an ARM CPU loaded with Nvidia

• 4:42Speaker 00

  parallel processing graphics, basically into one system on a chip that has a whole new memory architecture relative to the historic way that PCs had been built. And the target for it are the PC makers. So it's very, very exciting. And if you, the mainstream press, the stock market press, the CNBC going on behind me, like, they all looked at this as like Nvidia entering the PC business, which is, let's call it the mainstream chip business. But, which is so weird, because, you know, long before, in the start ages, which is now,

• 5:22Speaker 00

  we're talking about 2011, we actually announced Nvidia making PCs and making the Surface computer, the very first one. And it was a Computex big announcement, and it was a CES announcement, and it was, you know, I, I remember that very vividly, the partner slide had Nvidia, and Qualcomm, and Texas Instruments, and, and all the chip makers and PC makers. So it has a ring of familiarity at, I would imagine, like, one one hundredth the scale. All right, I sent you guys, you could throw it up, but I sent you the, the tech meme reel from that day when we announced the stuff. And it was still a pretty significant role. So you'll, you'll be able to throw it up at some later date.

• 6:07Speaker 02

  Yeah. So in what way is, is this laptop more like AI-native?

• 6:14Speaker 00

  So the big thing that's really changed is that the the compute burden has shifted from the CPU to the graphics processor and then the associated neural processors, TPUs and those chips. And and that's the thing that really changed. It's not unlike, you know, 15 years ago, what had changed was the bulk of the interesting processing had moved from the CPU to the GPU just for rendering. Right. And where we are today is just an extension of all of that. And and it's the difference, it's just insanely important. Because now that's the compute that that we think everybody wants to do. Now, a lot of people will look at this and go, well, I use ChatGPT.

• 6:59Speaker 00

  I do it on my MacBook Air, my Chromebook, my phone. I don't really need like another thing. The the problem is, and this might be a word you you guys have used or heard before, but the problem is tokens. And so the problem is that everybody is gated by the consumption of tokens, which cost money. Whereas you can't get them if you're trying to use them for free. And so the interesting thing about this device is how much of compute can it move to your local device? Where you basically have infinitely free tokens? And and that is incredibly interesting and super important. Now, of course, you've all seen this run up to where we are today

• 7:42Speaker 00

  with the stacks of of Mac minis. Yeah. And everybody running their agents on minis. And so why did they do that? Well, there's a whole bunch of stuff about privacy and sandboxing them. But the primary thing was if you just want to let something roll for three days while it figures out your best travel itinerary, you really don't want to end up with a $10,000 bill. So instead, you buy, you know, three minis and let it crank away with like each mini putting something in isolation or whatever. And so, but if you just fast forward, you know, six or nine months, it's abundantly clear. And I say that as a predictor of the future, not like as a it's obviously intellectually clear, but like

• 8:24Speaker 00

  it just seems to me that this world where you're all gated on dollars per token is a thing that's going to move to your own device. Yeah. Which is exactly what happened with all of computing. Anytime there's a resource constraint that you have to pay for, it moves to your device and becomes free. And that it just it had I just don't imagine it. I don't know how it can't happen.

• 8:49Speaker 01

  Sure. So for someone who wants a more AI native device in like a year when all of these products have shipped, you think they're going to want like an Nvidia Spark laptop? Or do you think they're going to want to stick with like MacBook Pro? Or the rumored MacBook Ultra, which is supposed to come out later this year, early next year?

• 9:07Speaker 00

  Well, this is just, I mean, this is the huge thing. And the way that that I think this can play out is, well, of course, you could play it out in like essentially status quo, which is, you know, the Fortune 500, you know, 80, 20, 70, 30 rule will be able to just fall to Windows devices running Intel or maybe Spark devices running ARM, but running with a Windows operating system. And then, you know, the cool people, the bosses, the elites, or whatever you call it, running their MacBook Pros with Chrome or Safari, just connecting things and phones. But but there's another path where it becomes incredibly important

• 9:50Speaker 00

  to run highly optimized AI stack of software on your device. And whatever that stack is, is going to get optimized for a particular hardware base. And that's a thing we've seen over and over again. Now, where we are right now is just so interesting because we don't have enough information to know where things are heading. At the announcement last night and the press releases and the commentary, you know, Microsoft made it clear, much to my surprise, which we can go into, that the Nvidia stack of CUDA will be available and supported and part of this Spark. Now, there are a lot of ways for that to become true. It could be a download that just runs.

• 10:34Speaker 00

  It could be a thing that's installed, pre-installed on a Spark device. It could be a thing that's part of the OS and updated with Windows Update and administrative permissions and all of this other stuff. It could be a whole range of things that I still, nobody knows yet in terms of public announcing, how they're going to do that.

• 10:56Speaker 00

  The same thing holds for Apple. And, and you know, today on a Mac, you can run all the models locally and stuff like that. You can't really do that for, on a phone. And so, interesting question is going to be, what is Apple going to do when, at WWDC with respect to the CUDA APIs? Like, are they going to be native? Are they going to be a thunking layer? Lots of stuff could happen there. Are they distributed? Is it an App Store app? Is it an OS component? Nobody has any idea. Now, for both companies, the past is very interesting. And most people didn't live through this, but Nvidia has always been an outsider to the personal computer industry.

• 11:41Speaker 00

  It's always been an add-on. So, on the PC, if you ever wanted to use an Nvidia graphics card, you bought the card and you downloaded drivers from Nvidia. Or before that, they came on a CD or a floppy disk that came with your graphics card. And so, for 30 years, this whole thing was like, do you have the latest Nvidia drivers? Where do you get them? And we went from, you know, getting a new CD, to getting a new DVD, to FTP, to downloading them from the web. It was whole cycle. But it was never a first-class part of Windows until we fixed that in Windows 7 and got them on Windows Update and all this other stuff. And the APIs on a PC to do graphics, you could always just download the Nvidia library

• 12:25Speaker 00

  and call them. But the official Windows APIs were DirectX. And they just did the same kind of thing, just completely differently. That the X is Xbox. And so, Microsoft was all in on the DirectX APIs. They were a huge part of Windows release called Windows Vista, when they first got integrated, and then Windows 7 Forward.

• 12:49Speaker 00

  Then there were the Nvidia APIs, which at first were just the Nvidia APIs, then they became CUDA. Then for graphics, Nvidia embraced this open thing called OpenGL. And then Apple went through the same exact thing. On the Mac, you could download drivers, you could install an Nvidia card, but the APIs, and then they supported OpenGL for a while, but they always wanted you to use their own stuff. And the phone did away with all of that, and Mirable did away with all of that, and it was all in on Apple. Now, the good news for Apple was the native graphics were just outstanding, and they've always been great.

• 13:30Speaker 00

  On the PC side, Intel was so far behind that it just kept pulling both Nvidia and ATI/AMD to be, you know, what you used if you used Photoshop, or made movies, or were just graphics, graphic intensive. And so, in the next few weeks, we'll know what Apple is going to do for these APIs, and more importantly, the modders themselves and the runtime. I mean, Nvidia has an enormous investment in the open source models and tuning them through their hardware. And the ecosystem has done a great job, as evidenced by the Mac minis, of tuning

• 14:09Speaker 00

  those APIs for the Mac, but that has nothing to do with the phones. And so, that's an operating system difference. And the number of phone people is large, and as we know, the hardware is the same. Now, it's not quite the same, blah, blah, blah, amount of memory, all of that stuff. But it's very interesting to see the details of Microsoft and then what Apple chooses to do.

• 14:35Speaker 02

  Right. So, like, obviously, we are seeing like a memory shortage, right? And like, so what do you expect the cost to a consumer of this kind of, like, you know, very AI-native computing device to cost?

• 14:52Speaker 00

  Well, certainly, for the, having lived through like a half dozen component shortage things, you just sort of wait them out and you don't let some local max or local men determine the future. It, this will correct itself in short order. The history of it, whether it's been DRAM or hard drives or processor shortage, all of these things, we've had them come and go, or even smaller components. So I'm not worried about it at all. I mean, obviously, you know, if you're thinking that you need, you know, 96 or 128 gig for a standard consumer device versus, say, the eight on a MacBook Neo, there's a huge difference. But also that will change in the models too.

• 15:37Speaker 00

  Like, right now, the models themselves are all tuned to run in hyperscale data centers. And every month, it seems like there's a new paper that says, oh, we cleaved this giant thing off of the inference pipeline, so now we don't need nearly as much memory. So that all will get fixed. Not even an inkling of concern I have for that problem.

• 15:59Speaker 01

  So another thing that was just announced yesterday was, last time we talked, you were very, very excited about the MacBook Neo for Apple as like a category-defining product. Dell just came out with the new XPS 13 that is, they say slightly better specs on, slightly better specs than the MacBook Neo, and it's like $100 more expensive. So the Neo is like, what is it? It's 400 for students, 500 for everyone else? It's $599, $699. $599, $699. Yeah. Yeah. 499, 6599, I think. Yeah, the XPS 13 is $599, $699. Yeah. MacBook Neo is 499, $599.

• 16:46Speaker 01

  Yeah. So what are your takes on this?

• 16:50Speaker 00

  Well, first, you know, kudos to Dell. Like Dell is just on incredible role and Michael Dell is just legendary CEO and read his book, his, his second book that I, that that came out during the pandemic, I think, or right after, right before. It's fantastic. But XPS 13 for a very long time was sort of my go-to laptop when friends and family and whatever would ask for one. It is like the best laptop. And then it took a little bit of a dip and went in a funky way on, on design. And it is actually back with a vengeance now. And so XPS 13 is the laptop

• 17:33Speaker 00

  to get. Now, this latest one is an attempt to build on that same chassis using Intel. I, I, you know, 30 years, I can't keep track of the names.

• 17:43Speaker 01

  Or three.

• 17:44Speaker 00

  Panther Lake, Python Lake, something Crystal Lake. I don't know what it is. It's some lake. It's Intel names are always places I've never been.

• 17:52Speaker 01

  And Panther Lake.

• 17:54Speaker 00

  Panther Lake. It's, it's the Intel race. There's a lot of excitement about it. It does integrate some of the AI compute stuff into it. But it's not going to be the target machine that the, the PC ecosystem wants to sell. And its, its capabilities are going to be different from the ones that they do want to sell, which is different than the Neo, which has the capabilities that need to be targeted. So the Apple, the, the Apple hardware line has a lot of homogeneity in it in terms of capability. The PCs can become really hit or miss. And that's always been the difficulty in the PC ecosystem is even when there's a winning machine, it's not the

• 18:38Speaker 00

  one that when you walk into Best Buy and say, I need to buy a computer, help me, Mr. or Mrs. salesman. And then they just direct you to something based on spiff and current ads or whatever. So we'll see. I'm sure it's a quality machine. You have the tech people on X talking about, you know, taking sides. It's either the Neo killer because it has an HDMI port or whatever, or it's like embarrassing to the PC ecosystem because regardless, it still runs Windows. Those extremes are stupid.

• 19:13Speaker 00

  Um, you know, these, both the Neo and this machine are targeted at, at just people who need a computer.

• 19:21Speaker 00

  I think in five years, people who need a computer will also need a computer that runs agents.

• 19:26Speaker 01

  Yeah.

• 19:27Speaker 00

  But the hardware, software world will be unimaginably different in five years. So this conversation has no relevance to the product lines that will be available in five years.

• 19:39Speaker 01

  Hmm. So wow, it, it, uh, it goes up to 32 gigs of RAM and a terabyte of storage. Uh, it starts at eight gigs of RAM, which is like not great, I guess. Five, 12 gigs.

• 19:52Speaker 00

  That's not a good, that's not a good number for a PC.

• 19:55Speaker 01

  Yeah.

• 19:56Speaker 00

  The Mac will do eight. The PC is, I, you know, like, I hate saying it, cause, and I truly had a bunch of these over the past month, six months or so, but I, I spent a lot of energy with our team on getting the memory down to, to two and four gigs at the time. But eight is, is, is gonna be hard. Now, Windows is doing a lot of work right now on that. So we'll see where that goes. But right now, if you, if you, if you ask me about what PC to buy, I would send you to a 16 gig PC.

• 20:29Speaker 01

  Hmm.

• 20:30Speaker 00

  It takes work. It takes like techie work to get it down to be eight gig reasonable. Like uninstalling a bunch of stuff, playing around, stuff that you wouldn't, you shouldn't tell anyone to do.

• 20:41Speaker 01

  Like which PC would you recommend? Like a specific laptop? Dell XPS 13?

• 20:47Speaker 00

  I would get, I would get a Dell XPS 13.

• 20:52Speaker 01

  What do you think about the Surface lineup? Right now?

• 20:56Speaker 00

  Well, here, okay, like I'm obviously not objective about the Surface lineup.

• 21:01Speaker 01

  Hmm.

• 21:04Speaker 00

  I think that the PC eco... When we designed Surface, and I wrote like 80 million words on this, which everybody could go see on hardcore software. I'm not going to replay them here about what we did wrong and right. But originally, Surface was envisioned to be this platform discontinuity in PCs. It was going to be the move to mobile chips, ARM, and mobile firm factors, like a tablet. We shipped it as a convertible tablet, but as a tablet. We actually did an Intel x86-based Surface, and at the time, we called it an "objection handler", and it was to handle the objection of things you didn't like about the ARM-based Surface.

• 21:49Speaker 00

  You know, like that it didn't run existing software, it wasn't compatible with old software, or whatever. And so, my heart and the strategy for ARM was always to introduce this discontinuity where, look, the world is different now. The hardware world is different now. The usage scenarios are different now, and portability is different. People want better battery life. They don't want fans. They don't want viruses, and all that other stuff. Didn't work. I moved down to San Francisco area to chat to founders and stuff. And what Microsoft did was sort of basically abandon ARM for the next eight years or so,

• 22:31Speaker 00

  and focus on the objection handler side of things. So, all the surfaces that followed were, in my mind, a niche product, because they were just like different Intel PCs that weren't super important to me. I mean, I had brought each one of them, but they weren't important to me. AI introduces yet another opportunity to change that dynamic for the PC, to have it be forward-looking, not backward-looking. And I think this is an incredibly important opportunity for Microsoft and for the industry as a whole. But the way it's different, like it was in 2011, in that it's mobile chips and the scenarios are different,

• 23:16Speaker 00

  even more so. 80% of the typical PC buyers are just running browser-based compute, and they just want the... They just want the keyboard, the form factor, and they like Macs because they don't wear down over time. They have all their battery life for real. They have the viruses and malware are a whole different game. It's sort of this sealed case that we used to call it. PCs that did move to ARM also thwarted all of the Windows APIs, which was the thing we chose not to do. So now the new PCs running ARM are just the old PCs with the same viruses,

• 24:00Speaker 00

  the same problems with fans, the same, you know, lack of quality over time. Like, you know, the classic Windows thing is, oh, you could just go edit the registry. Well, if you have an ARM PC, you could still go edit the registry and you could still hork your PC totally and then you're screwed. And so I don't, I just don't think that backward looking is the thing. Which brings us to ass night and all the X comments on the SPARQ laptops. And everybody immediately jumping to two things. First, Nvidia announced that they're all going to run all existing Windows programs, which of course just follows from Microsoft's strategy of converting Win32 to ARM,

• 24:44Speaker 00

  which wasn't hard. We'd already done it. It was just opening up the dev tools and the ability to load the apps and things, which we disabled for ARM because we wanted to move the ecosystem forward to a new OS API. But then the other, the other part of this is, is just how, how you, you spin the whole thing in terms of backward compatibility. And then they said, oh, it runs every single app of all time. It's like, yeah, but you don't want to do that. And more importantly, the second thing is you don't need it anymore.

• 25:19Speaker 00

  But all of the enthusiasts are going nuts because they see it as Intel being replaced by Nvidia, which is conceptually true, except not really, it's just an alternative. And what you're going to see in the marketplace is just sort of this price comparison. And Intel and Nvidia are just going to drive the prices to each other. And only one of them can really afford the battle. But that doesn't change the value proposition for consumers, which is what they really want is to not have that backward compatibility. They just don't know it. If they got a PC without a fan, that you couldn't edit their registry,

• 26:01Speaker 00

  you couldn't break it, you couldn't just go into the system folder and delete stuff. All of these things that you don't even think about on a Mac anymore, and you don't even think about, you can't even think about on a phone. You don't want them on the PC. And so it's tough for me to see Microsoft sort of embracing this. Because I mean, I understand, like, if you want to sell the enterprise, you have to run that VB app from 2003, but that's not, you don't need to do that. You could just put it on a server and remote, remote into it. You could put it in a VM on a x86 machine. There's a million ways to do that. You just don't need to run it on the machine that you want to run your agents on. And in the short term, everybody is going to be running terminal anyway.

• 26:46Speaker 00

  And these agents in today's agents are all headless anyway. That will change too. But right now, that's the big, we're at this fork in the road. And Microsoft has already said the direction that they want to take it, which is they just want Nvidia chips to do all the things that Windows has always done, which always tests where customers. And Jay, you say the customer's like, yeah, but those registry editors and admin scripts and stuff, really screwed us up. And like, we know, we know this. So it's tough for me to see.

• 27:18Speaker 01

  Yep. Well, we're really excited for the Nvidia Spark laptops to come out later this year. We'll see if it can replace my MacBook.

• 27:26Speaker 00

  Yeah.

• 27:27Speaker 01

  I'm not, I'm not sure.

• 27:27Speaker 00

  We'll do a review.

• 27:29Speaker 01

  Yeah, we'll do, we'll do a tech review on stream.

• 27:30Speaker 00

  Well, I have the, I mean, you could see what it's going to be like if you just have the, um, the, the Spark that you could get from Dell today, that's the, the mini device. I have one of those and it's incredible. I mean, it's just, it's, my, and there's a tower as well.

• 27:47Speaker 01

  All right, well, we'll take a look at that. Um, Stephen, thank you so much for joining us.

• 27:51Speaker 00

  Sure, thank you, guys.

• 27:55Speaker 03

  Thanks for listening to this episode of the A16Z Podcast. If you liked this episode, be sure to like, comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X at A16Z and subscribe to our Substack at A16Z.substack.com. Thanks again for listening, and I'll see you in the next episode.

• 28:21Speaker 03

  This information is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z. Such advertisements, companies, and individuals are not endorsed by AH Capital Management, LLC, A16Z, or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.