0:00Um, great room. Sam, thank you for coming.
0:02Thank you for having me.
0:03Um, so just curious, did you catch Saturday Night Live this weekend?
0:09Uh, it like a little funny. I love Saturday Night Live.
0:14A little funny. Okay. Well, so that was about the Jeden Wong uh scandal. Um, you a few years ago, almost exactly three years ago, uh, had a situation where you were also ousted by the board. Uh there was a letter going around about Jensen that was I think copy pasted from the one that your supporters drafted a few years ago. Um so I'm curious, do you think he should do you think Jensen should come back?
0:40Um I don't think I know enough to answer that question. I Jensen has long been a hero of mine. Heroes have flaws. I am grateful for the company he's built. Um I do have like personal strong empathy to a situation like this. I remember uh what it was like although it feels like forever ago. Um I but it's hard to imagine the AI
1:08revolution without Jensen. Um and the company he has built took an early bet on us. Uh we took an early bet on him. I have extreme extreme gratitude for that. They also guaranteed I think $ 105 billion for your data center in Ohio. They're also, you know, they're the bank, they're sort of backing the entire industry. Do you think that guarantee is worth as much today as it was, you know, a couple weeks ago.
1:37Um, I mean, Nvidia is like a behemoth of a company. I think people still trust their guarantees very much. But but yeah, I mean Nvidia without Jensen like Jensen and Nvidia clearly very deeply associated uh and I I think the ambition he has brought and the scale has has been quite important to the industry.
1:59So you think there there will be a ripple effect there'll be an impact on the industry. Um, look at this point I think there's so much demand in and it's so clear that like AI is real and people want to use it and the revenues and demand are growing so strongly uh that I think the industry will be fine uh but again I think Jensen has played this extremely special and valuable role as someone to really drive things forward
2:28with great ambition. So anthropic going public this week. Um in a time when I think there's there's a lot of skepticism from investors. Um curious, do you actually want to go public? Like if you could just avoid it, would you?
2:45I think there's a lot of skepticism from journalists. Um investors seem to be pretty bullish on AI. I uh things will go up and down. Um prices will go up and down. You know, there will be periods of overbuild and underbuild. Um there will be periods where something goes wrong or something goes really right. But the the overall trend is that huge value is being created. Um huge value is being
3:12disrupted too. And I think investors seem I just as we were backstage listening to the amount of capital being invested. Investors seem to realize that. Um do I want to go public? We we will go public at some point. I I think there are real downsides to being public. Uh and it's a very volatile time in our industry with very high stakes consequential decisions to make around safety and how society's going to
3:41interact with this technology. Um and there are doubtless things that are easier as a private company. On the other hand, uh I think it's not just like investors that h happen to have access to private markets that want to invest. the world as a whole would like exposure to this and I think they're that's you know directionally a good thing to do.
3:59Yeah. Do you think if so Anthropic is going to go first? Uh that's clear. Once you go public, is there enough, you know, do you think there's a just enough investor demand left or are you going to be sort of the lift to their Uber, do you think? Is that a worry?
4:17And again, I I get that you have to like write something every day and there's got to be like, oh, this, you know, this fight and that one and is this winner and this loser and what's happening here and there, but like the macro trend and the value being created. Um, I am not I am not at all worried about there not being a huge amount. In fact, you know, one way you could look at the last year is there's been way too much investor demand and not enough supply. Um but you know I I think we'll be able to continue
4:44to build we the whole industry just incredible incredible value.
4:49Yeah. Do you So you made these these comments behind closed doors it was reported to Samsung executives where you said there's been some double counting in the industry. What did you mean by that? Can you explain?
5:00Sure. uh when for a long time uh there was basically one chip provider in the industry and uh so when people were counting wafers they could ask companies like ours what demand was going to be um and they could I think get a sense of where end user demand for the chips they could make would be um there are now many people making chips and many of those people think they're going to sell most of the
5:30volume to us anthropic and others. Um, and we have, you know, some limited amount of demand and data centers and everything else. And I think there are many people making chips all talking to the silicon supply chain. Uh, assuming that we're going to buy all of their chips. And, uh, I think the demand, the data center buildout, that stuff is solid. There are, but we're not we're not all going to be able to or need to buy every, you know,
5:59all of our supply from every chip vendor.
6:01What does that mean for the industry?
6:02Are people is it over, you know, are things overpriced? What do you think?
6:07Um, again, I'm sure there will be ups and downs. The long-term demand for huge amounts of compute. Um, and and the value that that unlocks, just the incredible stuff that people are now doing at at real scale with this technology, I think that will keep going. Um and you know like any other like any other tech revolution um the long-term trend is clear the forecast any quarter to quarter and year to year
6:36of supply and demand exactly matching. Harder to say it's interesting you say you know they could count on us buying all the chips. I thought you did want to buy all the chips. Um but do you think like one one way of looking at this whole Nvidia thing is now you can. They're not shipping to China anymore. that seems to be um the supply chain is very complicated.
6:55We've been constrained by different things at different times. Sometimes it's been memory, sometimes it's been rack assembly, sometimes it's been, you know, data center capacity. Um sometimes it's been some random component like you're like really we can't make you know we can't sell more services because the world is out of like PCB laminate or whatever. Um, so it's it's just a very complex supply chain um at now at huge huge scale.
7:21Yeah, we also just had this blue wave that was the other news before the the Wong stuff. Um I think it was all but one of these data center referendum bills passed. In other words, people hate the data centers. Um a lot of voter sentiment on both sides of the aisle against AI. I'm just curious like how do you think the industry screwed this up so badly in telling the story about this technology?
7:46I I think if the sales pitch is like we are going to offer you a cure to cancer and cheap stuff and the price of that is you are going to lose all agency and control and voice in the future. You are going to live like a Labrador. Maybe the machines will try to make you happy, but you certainly won't have any ability to do that. and we're going to like rush to hand over um you know control of the future to AI and we're going to like make our make our AI really nice uh and he's going to take good care of you but
8:15that's going to be that. I think it's a terrible sales pitch. I don't want that world. Um I and I don't think most people want that world. There's other big questions about the economics too.
8:26But I I think fundamentally if you talk to people who really don't like AI and you get past the initial things. I think data centers are easy to point to because they're like a big physical instantiation of this broader thing. I think the job stuff it's it's a real immediate term worry that people feel very deeply. But there is this like long-term question about what are people's role going to be in society and what you know what is the balance of power going to look like either between
8:55people and AI or between people that have a lot of AI and a lot of computing power to use AI versus people who don't.
9:02Um and I think this is a place where the AI industry needs to take a very strong stance which is we are aggressively in favor of and committed to people staying in control of the future. um we should not want uh to have a handover, I think that's the current term people are using in Silicon Valley. Uh either gradual or sudden to AI uh you know making decisions about our future.
9:25This should be a tool that people get to use and I think people can kind of tell that not everybody in the AI industry thinks that way and are appropriately terrified. uh and that combined with the economic uncertainty and you know various ways that AI has you know caused cyber issues or whatever um has people against data centers but I think from the conversations I've had that what they're really against is this feeling that man are the people building AI actually
9:55really truly on team human. You said back in uh July 2025 you made this comment about Democrats uh you're politically homeless. Democrats have lost the culture of innovation. Do you still feel that way?
10:09Um roughly speaking, you know.
10:13How does that affect your view on the the the results, you know, what we're seeing now? I mean, obviously I will I want to work with whoever our elected leaders are. Do I still think there is room for like a party that more closely matches the ideals that I think have worked for America and the spirit of innovation for a long time. Um, I think that'd be great. I hope that both parties will move in that direction and I think there are like positive signs in both
10:43directions. Um, but yeah, you you said you want to work with the administration. you've, you know, before. I mean, you could call up David Saxs, you could call up Susie Wilds, can you call up Steven Miller?
10:56I don't know. I'll try.
10:59Have you, Do you still, you know, just curious like when was the last time you talked with Trump?
11:05Um, the China State Dinner.
11:10Yeah. And And what was the discussion then? What did you guys talk about? I mean, you don't get to have like a deep in-depth conversation. That's a lot of stuff. But yeah.
11:20So, you know, I can't help but think about you had your own spy drama, right, with this, you know, person putting a hidden uh a hidden microphone in your office. I think you've talked to employees about the the danger of, you know, people pretending to hit on them and potentially being agents of a foreign government. Um, you know, I'm just curious like has this changed how you interact with you have a lot of investors from the Gulf. Do you view them differently because of all this stuff?
11:50I we have now had I think more practice than is maybe obvious from the outside about living in this world where it is clear that AI is an extremely important thing happening and that people are looking for an edge on in all sorts of ways whether it's you know putting a camera in one of our conference rooms or trying to take our engineers out for a few drinks and get them to tell pre-training secrets um or whatever.
12:18ever. So, I think we have been living in this reality that everybody else is now kind of very publicly and suddenly coming to grips with. Um, and I'm sure we've been tricked sometimes and I'm sure that our operational security and overall security has not been perfect, but we have had time to adjust to this. the the a thing that we talk a lot about uh when it comes to safety in which I put this bucket of of
12:48stuff like the I put security in the broader bucket is being pragmatic centrists. Um it it totally does not work to I think say we're going to just build AI in absolute secret in a lab.
13:02We're not going to ever trust anybody else with it. We're not going to bring the world along. We're not going to iteratively deploy this technology and get the world to, you know, be able to use it. Um, but also we're not going to pretend like this is like any other technology. And there are no moments where we have to do things maybe differently than any kind of company or any industry has come before. And that can range from things like being very careful in our external interactions to saying, you know, at some point this technology may get so powerful so quickly that we really have to change
13:31all of our our plans uh to be able to continue to make the safety guarantees we want to make. And um that thing we talked about earlier I think this principle of people staying always standing control of the future uh that's going to require some very different it may require some very different actions as these models get so capable. People talk a lot about recursive self-improvement. Easy to say those words. I don't think most people like really have wrapped their head around what that would mean and what we might have to do. Um so I think for all of our
14:00interactions we try to like have the weight of what is happening and what we are doing on our mind in every way and then act as these pragmatic centrists where you have to kind of like engage with the messy reality of the world.
14:16Since you mentioned RSI are you there yet? Is that happening? Um I think like many other things including the term AGI it is like very hard to have a clean demarcation and say now we didn't have AGI now we do. Um are we developing new models faster than we would without using our technology? Yes. Are we pushing a button and just saying hey model like design the next version and rapidly
14:45self-improve with no human oversight?
14:46Definitely not. um we could be pushing more towards the RSI spectrum than we are, but I don't think we could safely do that right now. And again, I'm not I I think this like question of if we make all decisions first from a principle of saying people need to be in control of the future forever. There are going to be a lot of follow-on implications from that for things like how quickly someone should be willing to push onto the R.
15:14So, you're going to hold back on RSI a little bit because of that. I I think that holding back on RSI is a tactic. Uh I think RSI can be a good thing.
15:23Certainly using this research, using this technology to do better and faster alignment research is a great thing. Um but I think you should model us as we will make our decisions um from a place of being unwavering on people needing to stay in control of the future and sort of this pragmatic centrist view of safety and interaction with the world.
15:45Meter uh had this report I'm sure you saw about Anthropic saying that they they've doubled the speed of their capability development on specific tasks compared to like a couple years ago. Are you also seeing that rate of increase?
15:59Um I don't know of a measurement like quite like that. Uh and and I think it'd be like it it is hard this these things are still jagged enough that it's hard to say like overall we're doing this whole class of things two times as fast. I'm sure you could find some things we're doing much faster than that, some things that are much slower. The blended numbers I think are always harder than they seem. Um, we are speaking of Meter, uh, we are working with embedded evaluators now, uh, including meter. And
16:28I think that this principle of having people getting like an outside opinion on the safety work and the sort of robustness of these systems is going to be increasingly important and valuable.
16:42So, let's talk about products. I use dots. I find it incredibly useful. Um, I was looking at the terms of service and I was sort of wondering like where's where's my data? Like where's this memory being stored and it's somewhere on the cloud. Um, currently I can't change it or you know edit it in any way. Um, and I'm just curious like it worries a little bit worries me a little bit about like lock in. would you allow people to sort of have a mechanism to
17:09export their dots uh memory to other so we do for chipt memory um and that's like an easier structure the the I think in principle yes people should be able to take their data with them the dots version of this is harder than the chatebt version of this like dots really can build up like a quite sophisticated understanding
17:37of your life. Um, so I don't have like a ready-made answer. I'll say I I agree with the principle in general of like people people should have their data and be able to do what they want with it.
17:46But it do you think people will be sort of worried about that? It's like a when you talk about this this thing living in the cloud that's like got a full understanding of my life. It just seems like I want to sort of own that. I don't know. I don't want that to be out in the cloud and I have no control over it. I can't even see it. Well, I think you want some way to be able to see it, to modify it, and to take it if you want to take it. Uh, exactly what that look like. It's easy to think about, you know, and someone said to me recently, I thought it was a good insight that the
18:15biggest lock in the entire tech industry is your um your iPhone camera roll and your eye messages. you just like that is over now 20 years of our lives and it's there and you really like you don't think about it because you can kind of sit there and look at it and probably theoretically you could export it but you're not going to. Um and but that's at least like you can just say like I have 100,000 photos I'm going to move them somewhere else feels theoretically plausible. Um, if there is just this like
18:44subtle understanding in the same way that if you have like a really great assistant that understands your life and your preferences and that like oh he normally wants this but in this weird circumstance he wants this. Those are like hard rules to write down. I mean maybe maybe dots can do it in a way that person that people can't. Um and again I agree with you on the principle. I just like haven't thought before this about like what the actual button in the product version of this looks like.
19:08Well, you never want to fire that assistant. That's for sure. Right. So that's yeah good good customer lock in there. Um okay so speaking of consumers and everything you got in some hot water talking about computer RAM uh the other day. Um you know you said I think it's a bad allocation of resources for people to have all this RAM on their home computers. Um you know people feel like I should be able to tinker with this stuff. I should be able to use it on my
19:36you know run models on locally etc. um that was a big part of like the PC internet revolution was like these home tinkerers who do cool stuff in their home computers. Do you think that we're like we're missing that in this era of cloud everything in the cloud? I I want to be clear. I want people to have a lot of RAM. I think everybody should buy a lot of RAM, be able to buy a lot of RAM for their own computer. I think this is good. I think the home tinkering is great. Um unfortunately a few years ago we started saying like the
20:05world needs to build a lot more memory supply chain in particular. We said the whole chip supply chain, logic, memory, packaging, whole thing. We said in particular there was going to be a huge squeeze on memory coming and you know like someone quoted me as saying the world needs to invest $7 trillion in capex and for like I was in much more hot water for that than the RAM comment. Everyone was saying this is ridiculous.
20:31This is impossible. This is insane. The world will never do this. This is ludicrous. this is, you know, there's no there's no way there's this much demand in the world and the world did not build out that much and now we have seen DRAM prices. I don't know exactly but maybe they went up like 5x 7x something like that. Um I think it would be much better if the world just had a lot of memory.
20:54um that comment which I think got misinterpreted but there what there is like a true part of it which I will stand by is that if we're going to be in this memory supply shortage then in some sense whether or not we'll decide to do it because I think there are other trade-offs like people having their own you know at home inference the optimal thing would be can we share this memory and have it working 100% of the time rather than like sometimes off in someone's edge device uh again optimal
21:22use of this limited supply maybe is not the reason to make the decision but that's what I was trying to say what what's the argument I mean we have like you know the CEO of ASML here right like what's your argument when you meet with these people like because they're saying I can't do I can't increase production really uh no I think now people are increasing production like I I think we will see production get ramped up a lot it just it would have been good to do it a few years earlier do you think they did sort of drag their heels on that they didn't they didn't
21:52believe that the this was going to happen. The way that someone explained this to me is that being in the memory business has been approximately the following story for the last several decades, which is you invest $20 billion and you make $19 billion and that's it.
22:12And so and like you know there's booms and busts and some years you make money but on the whole um it has not been a great business and now all of a sudden it is. But uh I I like the way that I've explained it is the way that it's been explained to me is that the industry has just been through too many booms and busts and so people are nervous to or they were nervous to really make a very bold capex commitment.
22:39So on the data center point, do you think there is all this opposition? Will the data centers still get built or is this going to really hamper your ability to to do the buildout?
22:51Um I think data centers will get built.
22:53Exactly where exactly how you know what what the things that make sense for the communities they're in or whether we kind of put them further out away from communities, that stuff I I think wins and public sentiment will change over time. Um it wasn't that long ago that everybody wanted a data center in their community for the jobs and the kind of industry and you know like it was now clearly they don't people like put them way out in the desert. Uh maybe people want them
23:21in their communities again. Um I I personally have like a lot of sympathy to not wanting to have a data center in a community but I also think there's a lot we have a giant country. We have a lot of space we can do this. And I think people really want to use AI. People also want a lot of the other things that data centers provide.
23:39People like the internet and like watching videos on their phones or whatever. Um, so I think data centers will get built, but clearly the current way they're working being built and the engagement with communities is not working for people and you know the ecosystem will adapt. Do you think this it's a concern? I people talk about this buildout being sort of funded by debt and that it's all a house of cards. Do you do you think the debt is a problem?
24:06Um I have never seen the actual revenue growth including great margins show up for any revolution in tech that I have witnessed in my career. If it were not for that uh then I think it would be a problem. And I also think that you know these are like smart people on the whole issuing the debt. Um there's going to be like in any moon, there's going to be a lot of foam on the top that maybe needs to come
24:35off. Um there will be dumb things here and there, but but the fundamentals of not only transformative things happening in business today, but business is now coming to us and saying like I have seen the light. I see what's going to happen here. I can see what this does to the cap products and services I can offer in the future. um that just seems
25:02extremely clear and compelling. Now, is it possible that the buildout can get ahead of how quickly the economy can digest it? Yes, that could happen. Um I don't see that level of recklessness happening right now. Um the and I also think we're just we're underpricing we're underweaiting what can happen with the next generation of products. You know we had the chat bots then we had
25:30the coding agents. I think in the last few months we started to see the beginning of AI for science and what can happen with using a lot of compute to discover new math new physics. Um, I think we're going to start to see what doing a lot of compute can do for chip design and other some of the most valuable kind of engineering projects we have. I think we're going to see what a lot of compute can do to do a lot of the kind of wrote
25:57work inside an organization. So, and all of that has great value and I expect it will keep going.
26:03I mean, you announced actually just this morning that you're going to expand free compute for academic researchers. Um, you mentioned these math experiments. You've you've solved these historically difficult math theorems. Um, I'm curious what you think the impact of those, you know, particularly the math discoveries.
26:22Like you're you're invested, for instance, in nuclear fusion companies in, you know, companies that are partially reprogramming cells, like really cutting edge science. Do you see immediate impacts or will it be more of a long-term thing?
26:35Some of both. Uh generally I'm I'm I'm a believer that maybe the most important driver of long-term human progress comes from better scientific understanding. And sometimes that happens that very quickly changes uh society and the economy like you know we figured out how to make language models and very quickly the economy changed. Um, and sometimes it's like a fundamental discovery in physics and it's going to apply in a super
27:05important way to a new material or whatever, but that will take decades longer. So, I think it will be a mix.
27:09But what we're hearing from scientists is so compelling now about how much more productive they are. And not just that they can work faster, but they can discover things that they just weren't going to be able to figure out otherwise. We heard this a lot from mathematicians. We're now hearing it from physicists, chemists, material scientists. um and a world where we can get 10 years or 100 years of science done in a year. Um and then the ability to
27:36compound that into valuable products and services and technologies we all use. Um I think the world can get so much better if we think backwards 100 or 500 years and what it was like to live then. I think we are very grateful for the scientific discoveries and the engine to get them usable in people's hands we have had since um and I think in the notistant future I hope people will look back at 2026 and be like they had it really rough and the main reason
28:04will be the scientific discovery and the compute and the AI that is being used to and the robots whatever to make our lives uh just much more enjoyable higher quality more fulfilling higher agency to live. Um, so we were thrilled to give AI uh to the scientists and I think that the next couple of generations of models are really going to surprise people on the upside of how much they can be an engine of scientific discovery.
28:29Well, I think we're going to end it on that optimistic note. Okay, it's a good one. I really appreciate the time. Thank you very much.