NVIDIA Takes The Crown, Return of Ilya, Vision Pro Stalls

0:00NVIDIA passes Microsoft momentarily to become the world's most valuable publicly traded company. Ex-OpenAI chief scientist Ilya Sitzkever has a new AI company focused on super intelligence. Apple is managing its ambitions, shall we say, on the Vision Pro. And plenty more coming up right after this.

0:21You're used to hearing my voice on the world bringing you interviews from around the globe. And you hear me reporting environment and climate news. I'm Carolyn Buehler. And I'm Marco Werman. We're now with you hosting The World Together. More global journalism with a fresh new sound. Listen to the world on your local public radio station and wherever you find your podcasts.

0:48Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced format. Big week of news looking at the rundown. We have so much AI to talk about. Also, an update on the Vision Pro. We have a great guest for you. Returning to the show is Reid Albergati, the technology editor at Semaphore. Reid, welcome to the show. Thank you. It's good to be back. It's great to have you back. Another historic week, shall we say, here on the show, because for a minute, really on Tuesday, and right now the market is not helping it sustain this moment, but who knows, by the time this goes to air, it could be back in pole position.

1:23But NVIDIA became the most valuable publicly traded company in the world. We had talked about when this might happen, whether it was going to happen this year, somewhere down the road, whether it was even possible. And now we have it, right? So a company whose product effectively is the chips, the software, and the networking required to train and run AI models became the world's most valuable publicly traded company for a moment. It's significant. Looking at it, this is not a company that people had on their bingo card, so to speak, to take the pole position on the public markets two years ago, a year ago, even in January.

1:59And yet here it is. What do you think this all means? And do you think that it's a bubble? Because that's obviously the question that comes up. Yeah, I mean, I don't think AI is a bubble. I think, you know, they, NVIDIA has been now for like a decade, they've been an AI company, and they're, they've timed this perfectly. And they are now kind of the infrastructure layer, both in hardware and software for AI today. And I think the question for investors is just, you know, can they be disrupted? Every single company, it seems, that's in this field is trying to disrupt NVIDIA, even as they are also trying to sort of kiss up to NVIDIA and get as many of their chips as they possibly can today.

2:39they're trying to figure out how to build their own silicon and they're and they want to replace the software layer the the you know kuda so i think nvidia has everybody gunning for them and you know the i think for a while now the smart money has been saying okay this is nvidia is a great company but it's not it's not quite sustainable um and it's i think at some point they'll be right but who knows how far how far this can go i saw a great meme this week uh where It was like, you know, one of those people said that NVIDIA buyers will eventually become fatigued. And then it was like NVIDIA buyers.

3:13And it's just like a thousand pictures of David Goggins. Just like, it will not be fatigued. But I think that what you brought up is interesting that there are companies gunning for it. And to me, it's like, I don't think anytime soon we're going to see a company like AMD, which is also making chips and software, I believe, to run this stuff. I don't think that they are going to be the ones that are going to potentially cut into NVIDIA sales or even slow them down. But what you mentioned, when you said everyone's trying to do this, my head immediately went to like, are the big tech giants going to be able to make this work on their own?

3:51They're the ones that are the biggest buyers. I believe 45 % of NVIDIA's data center revenue, which includes the chips and the software and the networking for AI, is coming from the big cloud providers and probably Facebook as well. Talking about Amazon, Microsoft, Facebook. If you want to do a Google, if you want to work on AI today, you need to use NVIDIA. But they're all working, I think, to a T on their own chips. And if they're not, if they haven't announced yet, there's probably interest inside the buildings to do it because these chips are expensive. Goes like, what, $30 ,000 to$40 ,000 each.

4:25There's a new generation coming out. It's not just the chips though. right? It's, yeah, they're working on their chips, software networking, but they're trying, they're all trying to figure out ways to essentially make AI development hardware agnostic. And that's a software problem, right? They, right now, everything around, you know, building these AI, it all works better on these, on these NVIDIA hardware, on NVIDIA chips. There's so much work going on now that, you know, is, is honestly beyond my, you know, I mean, it's super, super complex stuff and like, you know, would, would bore your listeners, I think to not to, not to make that, that, that mistake.

5:06Um, I probably shouldn't use that term, but like, we're not going to get into all the details, but basically if you can, if you can figure out how to, to allow AI models to be trained and run on basically any piece of hardware, then you don't necessarily need the most powerful chips. You could just buy more of less powerful chips, right? And right now you can't. Right now it makes economic sense to buy the absolute most powerful chip that you can possibly buy and NVIDIA makes that. but that may not be true a year to five years from now. It may be that you could take, you know, it may be that you could take 10 billion raspberry pies and put them together and string them together and train AI models.

6:00Because if you solve the sort of this interconnect problem, right? The problem of like the loss between connections of chips right now, you want as few chips as possible right now because you lose so much. by connecting them all together. And so then you want the most powerful chip. Does that make sense? I mean, and that's just one example. It's so interesting talking about how this is a software problem. One of the most interesting tidbits I had after spending a full day at NVIDIA on Thursday of last week at the headquarters, which was very interesting, was that they have, I think, they definitely have more software engineers than they have hardware engineers at the company, which I think would sort of boggle people's mind.

6:45Like people think the surprising thing about them is that they design the chips, they don't manufacture them. I think the real surprising thing is that the true edge they have is, I mean, of course their chips are better, but the software and the networking as well, talking about that loss problem is world-class. And that's why it's been so difficult for companies to come up and grab that lead from them. Yeah, I mean, I think that's the other point. Yeah, for sure. They were smart. I mean, they started working on that and building essentially that moat because they knew that the hardware, you know, could, could become commoditized.

7:14And they started building that, you know, years and years ago. Um, so I, I have a lot of respect for them. I think it's a, it's a, it's a great company. They are a very, very impressive company. And here's like a few stats talking about like, people have talked about this NVIDIA FOMO, right? Where it's like, obviously it has elements of being a meme stock and the rewards that some investors have gotten for sticking with it are just unbelievable. So this is from a Bloomberg report. By the way, the stock today is, or stock right now, I'm looking at it, is$3.29 trillion. Not bad. It's up 170 something percent this year.

7:51Its data center business rose 427 percent this year from a year earlier, which is nuts to 2.26 billion. And that's 86 % of its sales. That's all the AI stuff. So it really is, I mean, it does a lot. Like you see that there's a real breadth that the company does, but AI is really pushing it forward and has 80 % of the AI chip market. But here's what I want to talk about with the investors that have stuck with it since the beginning. This is from Bloomberg. If you bought NVIDIA at the IPO, your return is 591 ,078%, including reinvested dividend. That's nuts. And this is how difficult it's been to stick with them.

8:31So this is from the Bloomberg story. Longtime investors in NVIDIA have had to stomach three annual collapses of 50 % or more in the stock. Sustaining the current rally will require customers to keep spending billions, a quarter on AI equipment whose returns on investment are so far relatively small. So first of all, if you bought Nvidia at the IPO and stuck with it, kudos to you. You are a steel stomached investor. But it also is interesting because who says, you know, it's almost like the past is prologger or whatever they say. and maybe this is the case again where you know we don't have huge rois on ai uh spending right now and that could eventually i mean i'm not thinking i don't think it will happen this year or next year or the year after that but that might eventually be what cools this off and not competition what do you think yeah it could be i mean to the chip market it's just always cyclical there's all it's just always a boom and bust so you know there will be a bust i mean it's just there's no there's no way around it i mean i think the question is the smart question is more like than what happens after that is there another you know do they just keep soaring after you know these inevitable dips right and one of the things p i thought was smart just to look back and compare it to like when cisco was at one point the most valuable company comparison yeah i mean it's that's sort of an interesting one because they're more on like the infrastructure layer and when when the internet was relatively new.

9:59You know, they were building all this networking equipment and it was the hot, you know, the hot commodity. And then eventually this stuff becomes commoditized, right? I mean, it's just, it's the way it goes. Yeah, but I think that kind of building off what you're saying, it seems like a pretty good bet that what we're seeing in AI is not gonna slow down in the next decade, at least. And so if you're NVIDIA, you're really sitting pretty right now. And Cisco is still around, doing fine. yeah we love webex um no i'm just kidding i hate that thing but um but yes disco still alive they so they are an advertiser with semifore so you know i guess i'm so you're gonna yeah shout out webex okay i won't uh make you shit talk your advertisers no i'm just i'm totally joking i i i there's a there's a there's a wall there just for your listeners just to be totally clear yeah but But before we move on to what Ilya Sitzkever, the former OpenAI chief scientist, is doing, for listeners, this is obviously a huge story, like the NVIDIA story.

11:04There's almost like so much more to talk about. And we will be doing that next Wednesday when Michael Batnick from Ritholtz Wealth, who's been a longtime holder of NVIDIA, the second half of the Compound and Friends podcast, along with Josh Brown. Batnick and I are going to get on the line Tuesday to talk about what underpins and what the NVIDIA trade and the risks there. And we're going to have an entire NVIDIA stock-themed show for you on Wednesday. So stay tuned for that. I know there's a good chunk of you that are going to like it and a bunch of you that are going to really want the hardcore technology episode.

11:37So don't worry. I'm also interviewing Nick Bostrom, who's the AI doomsday philosopher who now believes we might see AI Utopia just an hour later. And that will be on the feed within the next couple of weeks as well. So a little bit for everyone. Okay, speaking of a little bit for everyone, the guy who basically helped build OpenAI Ilya Sutskever, is launching one of the weirdest, and I'm sure this means he's never going to come on the show, but I have to say what I believe. And that is that it's one of the weirdest companies I've seen launched in a long time. So it's called Safe Super Intelligence Inc.

12:12It's a startup. This is from The Verge. It's a startup with one product, and that is creating a safe and powerful AI system. I think that The Verge is really underselling it. They're the only product they will ever release according to Ilya is super intelligence which you know for just to give the definition it's AI that's smarter than humans and not just smarter but way smarter because it just multiplies its intelligence once it hits human level intelligence. I'm just going to say straight off here's why I think it's weird. This is from an interview that he gave in Bloomberg. Since Geber declines to name safe super intelligence safe super intelligence's financial backers or disclose how much he's raised.

12:53I got to say, I got to be strong here. I think that's total bullshit. You cannot come out to the public and say you're creating a company that has one product, safe super intelligence, and not tell us who's given the money. I think it's a completely unacceptable way to roll a company out. It undermines credibility in the project itself. this notion of dark money is is i think terrible we should know who stands to make money from the building of super safe super intelligence or super intelligence itself and if you're not going to say that sorry i can't take your project seriously what do you think reed so you're not an investor i take it oh no definitely not no um we obviously passed on the deal when i think i'm kidding i think you're no it's an interesting point that it's it's not the direction i thought you were going to go with it.

13:42I, you know, think you're making kind of a moral argument here, which is this is such important technology that the public deserves to know, right? And more than that, I think that if you're promising that you're going to create safe super intelligence and you're going out there being high-minded talking about how like you're not going to have the pressure to ship products because you only have one product and you'll release it when it's ready and you won't when you won't. And by the way, like this is the thing that you have to be safe about, that you have to make sure that there are guardrails, that it doesn't turn us all into paperclips, you know, wink, wink, you cannot get out there and not tell us who's funding this.

14:17I just think that's, that's unacceptable. No, it's an interesting, I think it's an interesting point. I mean, I was sort of, I read that and was kind of like, well, why? Like, why are you not to, is it because you don't actually have any investors? Is it because it's Dustin Moskovitz and, you know, you know, it's not, people are trying to avoid, there's a controversy around open AI, maybe, you know, maybe the investors don't want to seem like they're, you know, taking on open AI or something like that. I mean, there's probably some interesting reasons around that. I think the market will probably take care of this problem, right?

14:51Like, I think the other thing he said to Bloomberg, or maybe it was in his statement, was he doesn't want to get into this competitive rat race. I think that was the quote. I mean, I just, I don't see what investors like... Yeah, this is the quote. He says, it will be fully insulated from outside pressures of having to deal with a large and complicated product and having to be stuck in a competitive route race. Sorry, go ahead. Right, right. So it's like, yeah, let me invest in this one, you know, super genius AI researcher with maybe not a lot of business experience. And let me, you know, he doesn't want to be competitive.

15:27It's only probably the most competitive industry in the world right now. um i just just the idea of making a return on this investment seems very low to me and i think if he wants to develop super intelligence he's going to need to recruit the very smartest researchers in the world and take them away from companies that are guaranteed to make them very very rich and then join this company that doesn't want to be part of the competitive rat race. I think it's a tall order. So I think your moral point is a good one, but it sort of will be taken care of by the market here. Wait, hold on. Let's dig into this.

16:10What do you mean the market is going to take care of it? I think, no, he's not going to be able to recruit the talent or the investors and the whole thing is going to fizzle or it will just be a forever passion project of his and he'll get some enough funding to kind of just like get by and from people like you know maybe it's i'm just totally guessing i i haven't done the reporting on this but like dustin moskovitz jan talon the skype founder maybe eric schmidt's foundation one of his foundations are just going to give him a little bit of money and and and he can sort of work on super intelligence for the next 25 years you know and right meanwhile so is open ai and anthropic and you know yeah every other big google every other big tech company or he builds the ais that help him do that you know do the research and he just disintermediates the need to recruit i mean i think that's unlikely i think you're probably right but i guess where i'm going i basically it's so interesting like you're starting from the beginning and sort of charting like the path and i'm starting from the end which is like all right let's say they do it who stands to benefit from that financially because if you do develop super intelligence and you have investors right it's not self-funded it's like okay so like now we're gonna learn who's gonna like who's gonna make money what exactly are you worried about it's gonna like it's like north or something or i mean no no i just i guess i question the um veracity of the claim that you're going to create safe super intelligence if you're also doing it on behalf of investors and you can tell us that it's safe but it's also like it's not it's not i think we got to deal with this like i'm saving humanity's rhetoric and to me the rhetoric is more honest if you're like i'm doing this on behalf of some people who are trying to make a financial return.

18:01And then we can start to have a discussion about what your aims are in the real world versus this make-believe world that it seems like he's in. Well, now you're getting into this other issue, which is like, you know, what's this safe? I mean, every company is saying they want to build safe superintelligence. You know, Anthropic, OpenAI, Google, who else? I mean, those are the big three probably right now. I mean, they're all saying we're going to build safe super intelligence, right? And ultimately, they all have commercial motives, right? And I think at some point, you know, safety becomes very subjective.

18:40We don't really even know what safety means at this point. I mean, I think you basically said that. So I think I would just, I just sort of put them in the same bucket as everyone else. That's just sort of, you know, saying, yeah, we want to build, we want to build this stuff safe, but at the same time, we're going to go as fast as possible and try to make everybody rich. It's just, you know, I trust you. It just doesn't, you know, it doesn't seem plausible to me. I think at some point, like I was talking with someone the other day about this. they were saying they were at this, this, it wasn't public.

19:16There's a meeting recently with a bunch of people from all the big foundation model companies and they're all talking about AI safety and these employees are sort of, you know, like, like talking about how, like even criticizing the companies that they work for, you know, saying, well, I don't think they should have done this and that. It's like, yeah, but ultimately like those types of people are not going to be the ones with power at these companies i'm not you know not to like pick on any one company but they're all kind of the same like ultimately those companies are meant to make a profit and when if they develop something that is incredibly valuable like quote-unquote super intelligence it's coming out whether it's safe or not i mean they'll try to make it safe i'm sure because that's not good for business either but like that stuff is you know it's we're we're gonna see it right Right.

20:09And there's like a whole slew of news effectively along this line this week. So here's another story. So this is from the information. OpenAI CEO Sam Altman says company could become benefit corporation akin to rivals Anthropic XAI. So basically, the story says Sam Altman recently told some shareholders that the AI developer is considering changing its governance structure to a for-profit business that the nonprofit board doesn't control. And this is interesting about benefit corporations. I think this is like a pretty good structure. So benefit corporations are legally protected from minority shareholders who might otherwise sue the company for making decisions that don't prioritize shareholder returns.

20:51I don't know that seems like a more healthy way to do a company but maybe that's a question for another day but it does seem like like on the theme of like going from the world of make-believe to the world of real life like actually this move from open AI would be a more of an acknowledgement of reality we have billions of dollars from Microsoft we are product focused even though we have research we do want to grow and maybe this is the beginning of an acknowledgement of that Yeah, maybe. I think it's still profit motive. Even Anthropics, by the way, their sales pitch right now when they're going out and selling to companies, they're all competing to sell their API access to corporations.

21:37Their pitch is like, we're the safe option. We're the safe version of this. But also, it's a for-profit company as well. I mean, yeah, it's a benefit corporation, But, you know, they're raising billions of dollars and they're racing just as fast as anyone else. And I think, you know, yeah, it protects you from like the minority shareholders. But I don't see that as like changing your overall motive to make money. Like you've got a lot of employees there who were there because they know they're going to get rich. And, you know, that's kind of like I think that's the major thrust still, even with those.

22:14I agree. I think that like it's an admission that that effectively even though it's like this benefit public benefit company it moves might move if opening I moves from this like kind of weird nonprofit where like the CEO can get tossed if the board members don't like you know the way that he speaks to them to you know effectively like a more mainstream corporate structure. I think that's good for open AI and probably good for everybody else because it's like all right we're sort of dealing with an entity that knows what it is versus diluting itself into what it should be. Right. I also don't think when we talk about super intelligence, I mean, it's really kind of this sci-fi concept right now.

22:51And I don't think it's going to look the way people think. You know, it's not like one thing that becomes this all knowing super intelligent, you know, AI. I think there's gonna be lots, AI is just going to infiltrate everything we own. It's gonna be lots of different types of AI models that are specific for different use cases. And, you know, it's not, I don't think, I don't think it's like a winner takes all. Like I came up with super intelligence first and therefore, you know, I rule the world kind of thing. I don't, I don't see it going that direction. Yeah. I mean, probably not, but this is the way that a lot of people are building is that they want to build, I mean, this is sort of the line of thinking, right?

23:32They want to build AI that builds more AI and accelerates the progress. So, I mean, there's amazing stuff coming out. I'm not sure if you're seeing this. I mean, it's mostly like secondhand that I'm hearing it, but just the ability of this technology to code itself is getting way better. And are we going to get to the point where like most code is going to be written by machines versus humans? It seems likely, and if not imminent. But you're right, that's not super intelligence. Like super intelligence, you're right, is more sci-fi kind of. I mean, if we live in a world with super intelligence, that will be a little bit insane yeah and i don't even i just i still don't think people like really i mean they use terms super intelligence these term agi everybody kind of defines this stuff in different ways you know i think i think it's going to get a lot better and a lot smarter but i don't think it's like this i don't view it as this like imminent threat to humanity um no same here existential way i mean i think it's a net good for humanity i think i think we should be excited about the way this technology is developing generally and we should pay attention to safety and all that stuff i mean it should be important but i wouldn't expect any for-profit company whether it's a benefit corporation or not to be the one to sort of decide what's safe or not it's that's exactly like the more companies talk about are aware of what they actually are the better and here's another one where actually this is this is less it's an interesting story it's less coming to terms with what it does and more just like this pragmatic move that's happening within Google and how it's being registered.

25:13So this is from Bloomberg. Google DeepMind shifts from a research lab to an AI product factory. And this is from the story. According to people familiar with the lab, some researchers are frustrated with having to follow roadmaps that they feel have been imposed from above. Some of its biggest advances before the change, when they actually brought DeepMind and Google Brain together, came from small teams that banded together informally. And there's a feeling that this effort leaves less room for experimentation and the pressure is already leading to a sense of fatigue. It hasn't helped that Gemini has suffered one snafu after the other.

25:50So do you think this is a wise, let's say, let's take this story at face value. Do you think this is a wise strategic move from Google to sort of take its like world-class AI research house and make it more focused on product because i mean it seems like there are some rumblings from inside that like if we're not like kind of wasting time and experimenting i mean i'm putting wasting time in quotes right but if we're not wasting time and experimenting then there's a less of a chance that we'll be able to develop the next breakthroughs that will put google the next leap ahead so where do you stand on this one well i mean first of all to take the story at face value I think is, I'm not saying, not criticizing the story.

26:28I mean, John Victor wrote this story for the information a few months ago as well. It makes sense. And I think there's probably a bit of confirmation bias when you go out to report a story like this. I mean, you and I have both done this. Like we, you know, you want to do that story about how the company feels. And it's like, well, you have no way of surveying every employee at DeepMind. And I think you could probably write this story, you know, based on whatever, maybe 6, 12, 20 employees and still not really capture like how everybody thinks. I've met people at DeepMind who were working on things like, you know, AlphaFold and those wasting time sort of projects that you talked about that turned out to be amazing.

27:18And they're working on product roadmaps now. And they're super excited about it. And I think there's a, is it wise to do that? I mean, A, does, you know, does Sundar even have a choice, right? I mean, this is now like, he's going to get fired if he doesn't go full steam ahead. And like, you obviously have to take your smartest people and put them on your smartest people in AI and make them work on these AI products now. But there's something interesting about like, you have all these people who probably would otherwise be in academia working for this deep mind company within Google. And it's kind of this like ivory tower within Google and it's great.

27:58They're doing great stuff. And now you're like, actually, now I want you to make products. Some of them are probably going to be really good at it. And you're going to end up with some really interesting stuff by basically like letting these people loose on something that they've never really done before. And some of them are probably going to be annoyed about it. And maybe they should go back to academia, you know, and make less money. That's a hard, that's, that's a hard thing to tell your, your, your family, right? Like, I'm gonna go make less money back in academia to pursue my passion. But, you know, I mean, academia is traditionally where that kind of research happens and has happened and probably, you know probably should happen um so i in a way it's kind of like to me it's kind of like restoring the order in the universe where these companies are like okay we've been just like vacuuming up all this academic talent and having them work on these projects and you know now we're going to actually be companies and develop this stuff was that even coherent at all i don't know absolutely but i have a counterpoint for you okay so it's so i think that it sounds logical to say go to academia to do this work but academia is so far behind in terms of kind of like bringing our conversation full circle in terms of having the hardware software and networking to be able to do this yeah um this was a stat that kind of astounded me a while back but um i i feel kind of cruel doing this but i'm gonna make you guess how many gpus do you think stanford has stanford university how many do they actually have yes oh the stanford nlp group the nlp group like they're they're their natural language machine learning group i don't know i mean i know they they probably get a lot of help from from the private sector there like i i know stanford has access to some of these clusters so i guess it's probably like how do you really define it but i don't know they probably don't own that many i mean i i don't even know if they have their own cluster to be totally honest yeah so they they have 64 64 and that's like nothing compared to 650 000 that meta is going to have by the end of the year yeah yeah so which makes sense and that's why you know you've you know the the i'm sure you're well you've probably talked about it on the podcast that this you know research cloud the national research cloud right um no talk about it a little bit actually it'd be good for listeners well there's a pilot there's a pilot program to to build essentially resources for academia and startups, essentially like a gigantic cluster, supercomputer that you can, you know, and it's nowhere near what it needs to be.

30:37It needs to be funded by Congress. It needs to be much, much bigger than it is. And this is happening in the UK. It's happening in Europe. There are these national labs. And hopefully we'll get there, right? I mean, I think it should be, like government should be on par with with the private sector or even surpassing it when it comes to building these things right um if you look at like how all technology has been developed going back to like you know nuclear weapons it's just it's been massive government projects and so that's i mean it is that is an absolute real issue that you're talking about and a problem that I think people are working on and hopefully will be solved.