Sharp Tech with Ben Thompson - The End of DeepSeek Week: Moneyball for AI, The Future of Compute Demand, Geopolitical Reality Checks, and More

Episode Date: January 30, 2025

Andrew and Ben reconvene to answer your emails on DeepSeek and its implications. Topics include: DeepSeek as the Oakland A’s and Big Tech as the Red Sox, questions about distillation, video game history and coding to the metal, waiting for Silicon Valley products in AI, the future of compute demand and power consumption, and a variety of follow-up thoughts to Monday’s export control discussion.

Hello and welcome back to another episode of Sharp Tech. I'm Andrew Sharp and on the other line, Ben Thompson. Ben, how you doing? I'm doing well, Andrew. How are you? I'm doing all right. Happy year of the snake. It's New Year's over there. Are you a snake guy?

Not a big snake guy. Not a big fan. I am very happy I was born in the year of the monkey. Great animal. You know, just, you know, You find your blessings in all sorts of little ways. And God bless 1980. There you have it.

You know, that was a test and you passed, not a snake guy. So we can continue the podcast for another 200 episodes here. Looking forward to that. And I'm a monkey guy also. I had a stuffed monkey growing up. Always had an affinity for monkeys. In any event, as for the show, we recorded our first episode this week on Sunday night in America, about 11 hours before a truly wild Monday morning on Wall Street.

And in the intervening days, you've written about Deep Seek on multiple occasions. We've gotten a flood of great Deep Seek emails. So the plan for this episode is to just bounce around and hit as many as we can. Does that sound good with you? Yes, I'm going to try to keep my comments short and briefly, get through all of them. Always a challenge. It's good to have goals.

Billy says, is there anything to the analogy that Deep Seek is the Moneyball Oakland A's, having created new efficient tactics that the rest of the industry will now take as standard? If that's the case, what happens when the American tech companies utilize these tactics with infinitely more resources like the Red Sox did for the next decade after that Moneyball A's team? Do we yet know if the richer companies can add layers of, of AI-backed RL that would require more resources, but ultimately produce better models. What do you think, Ben? Great Billy's analogy here.

I like the analogy. It's a good one. Yeah, I mean, part of the reaction that was definitely overwrought is would Deepseek like to have more computing? Yes. Do they have more computing in some respects? Yes. Like this idea that now computing is no longer valuable.

is ridiculous. I think there's, you know, I think there's fair questions. You sort of mentioned the Wall Street angle. Everyone who's like, why are these stocks down? AI is going to take over everything. It's going to, like computing is going to fill all available space. Are correct. The challenge with stocks is there can be a timing issue.

So like, you know, at what point does all that computing get absorbed? Does that extend out the timeline? Does that change what matters for inference? things on those lines. It's always difficult to say what's happening on Wall Street. But there are rational reasons to simultaneously believe this analogy is a good one, that more computing is useful, and yet also why some prices might be down and why some might be up, why particularly the companies that are paying for all of this, right?

So, yeah, this idea that, yeah, then the Boston Red Sox adopted the A's philosophy and paired it with a huge payroll, it's a great analogy. That's why at the end of the day I was emphasizing again and again, this was a tremendous gift to the U.S. And to the industry sort of generally to uncover these optimizations and these ways to be more efficient because ultimately it's going to benefit everyone. And frankly, I'm still pretty skeptical on the this is all, this is, you know, this is all a lie. This is all a thing to just take down big tech. Like the smart thing to do, quote unquote, smart thing, if you're. your number one concern is geopolitical concerns.

It would be to not disclose this at all. Just, you know, keep this knowledge sort of internal. And the fact it was shared, I think, actually buttresses the idea that the details are broadly correct. And again, when we're talking about the money involved, it was for one training run, not for all the experiments, not for all the data collection, not for all the R&D, which again, the paper was very clear about what the money was. four. And it's gone through this filter, this telephone tag online that has, you know, it's gotten pretty war. And I think it's because people want to believe that story. Like, it's a compelling counterpoint to, you know, a week earlier, Sam Altman announcing we need $500 billion for

Stargate. And so it doesn't surprise me that certain people just took the $6 million number and ran with it. And it got out of control and twisted into something that it wasn't. Right, no, for sure. And, you know, the, Dario Amade, the CEO of Anthropic, I know we're going to sort of get to it. But he did mention the, you know, look, this is broadly in line with the cost reductions. We've seen its capability is similar to we did a year ago. Of course, it's in his self-interest to say that and argue that. But I also think that is reasonable.

And the whole thing about, you know, that they stole our data or which is referring to distillation, which we've talked about the last couple podcasts. That's like, yes, that happened. Can I ask, is there any way to prevent distillation going forward? Like, did any of these companies have any sort of defense to that behavior? Not really. I mean, it's just like, I mean, there's, they work to stop API access at scale. You can put rate limits on.

But they're, you know, at the end of the day, you're trying to like block. It's like trying to block piracy back in the day. Like, you're dealing with digital business. inputs and outputs that's that's you know and I've heard rumors of like crazy things that folks in China or elsewhere have done to try to distill these models even if they're they are rate limited or their API access is blocked or whatever including you know like I mentioned it in my in my article like pretty exotic things just with chat clients and like scripting them and trying to you know get all the answers out and it works it's like it works like said it works to the benefit

But internally, like no one, Open AI is not serving up the original first run model for users to access. That wouldn't be economically viable. A huge thing, a huge part of the post-GAPGPT launch was they were overwhelmed with demand and they were drowning in costs because like they were just serving up the whole model. And that was actually where a ton of the distillation breakthroughs happened precisely out of desperation and a need to how can we serve hundreds of millions of people economically.

And so distillation, it's just a fact of life. It's not a good thing or a bad thing. But my point here is people are like, oh, they're claiming distillation. That's cope. Well, no, it's true. Like, I think I've been pretty consistent. The U.S. labs are still ahead. Like, to do a GPT-40 level model or a sonnet level model or an 01 model,

months after those models were publicly available, by definition, you're not ahead. And the efficiency breakthroughs are real, but they're efficiency breakthroughs. They're not capability breakthroughs. So there's a bit here where everyone on all sides is, you know, arguing their interests, but that doesn't mean they're wrong. Yeah. Multiple things can be true at once. And I mean, in terms of the meta conversation, I think that the fact that Deep Seek is a Chinese company kind of broke people's brains and drove the reactions here to an insane place.

And I was joking about it with Bill Bishop earlier because we recorded Sharp China a day late this week. It's like anytime there's a big Chinese breakthrough in technology, it goes to this place where you have one contingent of people who are basically, they spend days explaining, why the progress doesn't matter, and it was subsidies, it was IP theft, or in this case, it was distillation. It's all not a big deal whatsoever. And then on the other side of the spectrum, you have these people who treat the breakthrough as evidence of like imminent US decline and changes to the world order. And so it's just like, the truth is usually in the middle. And I think the truth was in the middle on this one as well. Like the reality is Deep Seek is very well funded, not some $6 million side quest.

They were using a ton of American hardware with Nvidia chips. They distilled Open AI's model to develop V3 and R1. And they introduced some really cool innovations that will force competition to be better and smarter going forward. It's not an end of day's situation. And I wonder if, like, Mistral had done this, would this have been seen as a really interesting kind of twist and subplot and not one of the biggest stories in the world for the first couple days of this? week. I think you're exactly right. Like, because they're, you know, on the, I don't want to speak too much for the, the China side. I'm not Chinese. I like, like I did sort of, I think this is

probably going to be very empowering for the Chinese ecosystem, um, would be my guess. But from the American side, I do think there's a real tapping into a real underlying angst where, like, China is crushing us in all these areas. The one thing we have is technology and oh my God, losing that as well, which there's a bit as an American. I'm okay with that. A little bit of urgency might be healthy. That's right. That's right.

Exactly it. Indeed. All right. Well, one more analogy. Mike says, the incredible efficiencies that deep seek displayed reminds me of another era of impressive outputs on inferior hardware. The gaming scene of the late 90s to early 2010s. Video game consoles like the Xbox, PS2, PS3, etc.

were typically launched with hardware roughly similar to a mid-tier gaming PC, but they would stay in market for five to six years, meaning that midway to the end of their lives, the hardware would be very outdated relative to a new gaming PC. That said, games like Uncharted, Gears of War, and Grand Turismo would be released exclusively to consoles, with graphics better than anything coming out on much more, more powerful PCs at the time. That's because console developers were, quote, coding to the metal

and eking out every bit of performance from the hardware, including often Nvidia GPUs, maximizing against their static constraints, just like the deep seek team. PC developers, on the other hand, had to code to a moving target of a host of different PC configurations, different CPUs, RAM capacity, different GPUs depending on the PC, and it meant they couldn't come close to optimizing to the level of console developers. As they say, history doesn't repeat itself, but it rhymes. Thanks again for everything. Go commanders. The future is bright. Stick with it, Andrew. Thank you, Mike. I'm not a gamer or a video game historian. I'll defer to you. Grade the analogy part two. I think it's, I think it is a solid analogy. I mean, it's not perfect, like for lots of reasons. We could pick

apart. But it was the case that you were getting this incredible, this incredible output on old hardware with consoles and PC gamers would be sort of annoyed at it. One of the things that's possible, yeah. Yeah, one of the things that's shifted over the years is consoles have become even more PC-like and there's much more of a standard approach using game engines and less sort of writing to the metal for a specific console. And part of this is just an economic thing. When the cost of asset creation started to outweigh the cost of engine development, it used

to be you spent all your time building the engine. And when you're building the engine, that is like close to the metal. Then it realized actually, no, if we're going to make money on this game, it has to run everywhere so we can sell it to the most number of people that everything sort of got abstracted away. And I think that it's been the case by and large, you're going to get the best graphical experience on a PC for basically every game. And that's been the case for a while. But that was part of the sort of PS4 Xbox 3 or whatever Xbox 1 that like that was the generation where it really shifted. And we've talked about in the past, Sony was ahead of this. They saw this coming in part because developers were increasingly reticent to write to the medal for the PS3, which was actually it was hard to develop for.

And they started just writing for Xbox and then porting to PS3. And they're like, crap, for many years, people would write specifically for the PS2 or in the sort of the PlayStation before that. Now they're doing this new approach. If we want to have differentiation, we need to buy up all these studios, get exclusives, and then people buy our whole thing. So it's actually tied to this discussion we've had before. Again, it's not a perfect analogy for different reasons, but the idea of constraints driving tremendous innovation in terms of efficiency and things on those lines is absolutely the case. And by the way, I didn't expand on this fully. but if you do the calculations of the amount of compute they had,

their model efficiency was so high with their mixture of experts approach, particularly in training. One of the big innovations was it was thought that mixture of experts made training much more complex and difficult and less efficient, but you got the efficiencies in inference. They came up with a new way, particularly of making sure every expert got trained fully and housing that sort of counting within the experts

sort of a bias factor instead of trying to calculate it all at the top end that actually got them more efficiency in training, they still had excess compute, like, relative to what they needed. And that included them hand programming

some of their compute units to manage communication sort of more efficiently. That's crazy. The nerfing, like the nerfing worked. Like, it really did. And it's a testament. And this sort of part of the whole discussion, right? We can talk about distillation.

We can talk about distillation. the fact they are still, if not years behind, but months behind, that should not at all diminish. This is incredible engineering work. Like, it's, it's, it's, it's really amazing what they, what they did. And, uh, and again, it's a blessing to everyone that they released it and made it broadly available. Indeed. Okay. So related to engineering work, Adam says, I've really enjoyed the discussion of the deep seat kerfuffle this week. And when discussing it with a buddy in IT, I was reminded that what's happened fits perfectly into established practice and philosophy in the big tech world. Ben has repeatedly mentioned that because of constant advances in compute, the hard

work of optimization has almost never been worth it. Why spend valuable time making something run better when next year's chips will run it good enough and the opportunity cost of optimization is more features. My buddy drew a parallel to the bloated operating systems we have now and inefficient applications. We have utterly godtier hardware, but often the subjective user experience is no more responsive than it was for early PCs. I'm starting to wonder if the efficiency chickens are coming home to roost, have decades of skyrocketing compute capacity resulted in a software development culture where optimization is not given sufficient attention. Yeah, this is a really good point. And I saw a couple other people sort of mentioning this and

sort of putting it in my face that I've made this optimization point. I think it was made most concretely in an interview I did with Kevin Scott of Microsoft. He was making that point. But I was, you know, in broad sort of agreement with him. Yeah, I think there's two points. Number one, I think there is a bit where it's sort of like the comparative advantage question. Like you should always choose your comparative advantage. But the problem is there's a time element to that.

And there's a learning curve that is abandoned that. only comes, you know, we're seeing this with the U.S. and China, like the U.S. can't build stuff, right? Yeah. That's downstream from rational decisions made at the time that accumulate and lead to a wide diversion that ends up being a problem. In this case specifically, I think what's coming to Roost is it's not that we, it's not that it's not logical to spend $100 million or whatever it is to build the future models. if you really believe that AI is going to have this astronomical economic impact. There is a question, though, at what dollar figure does it make it feel like we miss the boat, right?

Yeah. And I think that, you know, I mentioned in an update this week, Google and how Google built up their infrastructure simultaneously to developing their product and completely revolutionized the whole server area, which used to be these bespoke machines dominated by Sun. And what they did was they basically made it, no, we're going to run on x86 like consumer grade hardware. Now, over time, Intel responded to that opportunity and made server grade x86, but still, drastically cheaper than sort of what came before. We're going to assume that it's going to break. We're going to build systems that are resilient and handle stuff breaking down all the time. And it's going to be actually much cheaper and much more scalable.

and that was foundational to their success. And so I think there's probably a bit here where we have probably gone too far in the don't worry about optimization. There probably should have been more dual tracks where one team isn't worried about optimization and another team, that's all they think about. And that's probably been a mistake, but I totally own the, I never said this before this week. sort of critique. And if anything, I've said the opposite. It's a very good point. Well, and as far as the challenge going forward, are you saying that's a muscle group that needs to be developed over time among today's engineers? Like, can you just flip a switch? This is the reason why I, like, again, I'm all over the place on Google. I can never decide.

I usually go the wrong direction. So always take my Google analysis with the grain of salt. But this is, I wrote a piece last year about Gemini 1.5 in Google's nature. I was talking about the million token context window. And there's almost certainly an aspect of how that works that is downstream from Google infrastructure and the way they optimize things and organize things. And they design their systems from the ground up to work with their software. And they do it in a scalable way. And Google is still the best at this.

They have the best data centers. They have the best infrastructure. They have their own chips. If anything, this is the part of Google that's still the most performance. It's still working. I mean, like, yeah. So it's a big reason for if you want to be a Google optimist, I would have this very high on the list that at the end of the day when it comes to serving the world, Google is still the best. They can do it the most economically.

They can undercut anyone on price, particularly if models are commodities. Again, my concern with Google is if they're really good at it, the chief entity being hurt is Google search. And so, but but from a setting that aside, the sort of close. classic disruption question from the infrastructure perspective, they're still the best. And to the extent that does matter, that's good for them. Okay. Well, related to all this, culture-wise, Dan asks, Andrew and Ben, at one point in Monday's podcast, Ben discussed what deep seat could mean for the LLM product overhang. In other words, all the products we should eventually see if LLM progress stopped today.

It reminded me of a discussion he had with Nat Friedman and Daniel Gross in June 2024, particularly this tidbit from Daniel who said, obviously, hopefully there will be better breakthroughs and the models will become easier to use. But you could also take today's models and make awesome stuff out of them. And why that hasn't been happening, by the way, despite the best attempts of this podcast, I think is a very deep and interesting question. What is going on with Silicon Valley in general? and is that innovation pipeline from research breakthrough to iPhone waning a little bit? I don't know.

So I guess my question is, Dan says, the same as Daniels. What is going on with Silicon Valley's innovation pipeline? Why aren't we seeing awesome stuff? Is it simply that LLMs are such a massive innovation that it'll take time to productize, as Ben has theorized before? Or is there more to it? Culture, big tech dominance, inertia among potential customers. It's, seems like the theme has come up several times now. And at some point, we need to ask what's wrong, either with the technology or the people supposedly leveraging it to build stuff. Ben, do you have thoughts?

I mean, I think it's a little bit of all the above, but all the above goes into the, the fact that historically, this stuff takes time, right? The internet, the World Wide Web was in the early 90s and you had the dot com era, most of a lot of companies of which didn't work. and then suddenly it did. And same thing with the PC. It takes longer than you think. And I,

again, why is it because of culture? Is it because of customer inertia? Is it because of all these bits and pieces? Yes, it's because of all those things. But also it hasn't been that long. Just to be clear.

Exactly. It's only been, it's been two. Now, Transformers were 2017, so that's been a while. But even that speaks. to it. Transformers were invented in 2017 and

it took a while for folks to even start utilizing them. And so you had GPT 2 and you had GPT3 which when we started that podcast series with Nat Daniels, when GPT 3 was out, it was before chat GPT and

we're like, no one's building products around this. This is actually amazing. And then chat GPT comes out. It's like, oh my, that was the wake-up call to everyone. But to your point, that was only slightly over two years ago, November 2022. And just the history of technology is that you can have tremendous breakthroughs sitting in plain sight. And they can actually also not just be sitting in plain sight, the transformer was,

they can, like ChatGPT, have everyone awake and scrambling. And yet it still takes time. And if anything, the larger the change and the larger the transformation, the more time it actually ends up taking. And this is a, it's kind of a unsatisfactory sort of excuse, if you want to call it that. But a lot of things about human nature and humanity are sort of unsatisfactory excuses. And what we can do is just look at historically, it takes time to figure this out. A lot of the initial things are you just take the old paradigm and slap on AI and see if that works. And actually, again, my favorite analogy, you can put advertisements next to you can put advertisements,

to text on the internet, like the newspaper, that didn't make any money. You had to come up with the feed. That's what actually made money. And Facebook invented the feed in 2006, like 13 years later after the sort of web was invented. So this stuff, it just takes time. Now, are there questions about Silicon Valley's ability to innovate and all these bits and pieces? Sure. There always are.

But I think that is actually part of the broader critique. Yeah, I mean, it's unsatisfactory because it would be so much more satisfying to have a take on like the screwed up culture in Silicon Valley and the complacency that pervades across all these companies. And I think it dovetails with some of what we were talking about earlier with like when you have just unlimited resources to throw out a problem, then maybe you lack the forcing function to optimize. and work on efficiency and really innovate in useful ways. At the same time, though, as alluring as that story is, it's literally been two years. And so I have to sort of take a beat before I hold anyone's feet to the fire on that one. And the fact the matter is just the chat clients are awesome.

Like they're, and you, this is another human nature point. The usability of these chat clients is basically, gated by your willingness and capability of thinking of things to do for them. Yeah. Right? Like I was writing the deep seek FAQ article. I initially started putting every question as an H4 header in Markdown, which is four hash marks.

And I'm like, no, this is not going to work out while I should just make them bold. Do I go through and I could, you know, figure out a way to do it or whatever? No, I drop in a chat, GPT, say change all these H4 headings to bold. And it does the whole thing for me. Right. Like it's just like. And it's not a huge thing. but that saves you what, probably like 20 minutes, 25 minutes or something?

Yeah, just in general irritation or getting someone else to do it for me. I guess someone just lost their job, my text formatted. So I mean, like even internally to each of us, we all have massively untapped efficiency gains that is downstream from our lack of imagination on how we can use this stuff. And so that applies at the micro level and it applies at the macro level. I think that's true. I would also say that the people I know in tech love using these on a daily basis, LLMs, that is. And I actually have more and more friends in just mainstream society that are using them pretty frequently. But I am waiting for somebody to productize it in a way that makes it more idiot proof. I mean, that's what I thought Apple was going to do with Apple intelligence, frankly, is make this so easy that it becomes like shorthand. for everybody. And then Apple intelligence has just been an absolute disaster. And so at some point, I think there will be even more adoption that we've seen thus far.

Right. There's probably going to be some situation, you know, maybe we're going to end up with more sort of classic founder series of someone solving a problem for themselves and they make a product out of it and ends up being a huge company. Right. You know, like the, there are more myths to be made. Yeah. Right now it depends on your volition.

And soon enough, it will be. sort of second nature for all of us. Right, but that was the thing with computers, too. And the web, frankly, it depended on your volition, right? Like, like the first people using PCs were eager to figure out what to use them for. And the first people on the internet were out there finding obscure communities. Are we talking about BBSs sort of back in the day? Carolina basketball message boards, absolutely.

That's right. That's right. So we'll get there. So this is a long rant in response to a rant you had on Monday's episode. Okay. I'm going to pour myself some coffee here. That's right. Get comfortable on your end.

William says, at the end of the latest episode, Ben dropped some absolutely abysmal takes after correctly explaining and discussing R1. The best take on Twitter that I saw was essentially everyone sees R1 and then doubles down on their preexisting views. Ben was no exception. When discussing, why are they called OpenAI, her, her? You have to remember the context in which they were founded. There was not a new AI lab being founded every week at that point.

The primary player was Google's DeepMind. Have you read the DeepMind papers? There's barely enough technical substance to do anything with. They didn't even allow external researchers to use the models at scale until AlphaFold. Open AI was open relative to Google. Google. And now the window for what it means to be open has moved. If you want to say, just want to jump in. Okay. Just want to jump in. Great point. Very, very well put. I'll continue. Okay. Plus one to William. We'll see how it does in the second half. If you want to say

F anthropic or open AI for slowing down progress, what the hell? Where is the Iyer for Google? The whole reason we have any of this is because open AI was founded. They are the reason this stuff exists. Moreover, the take about holding back GPT2 is also terrible. They did that to figure out, what is the minimum we should do to make sure this can't be misused. They now have a process that they, and Anthropic, follow when rolling out new models. They produce model cards that rate CBRN risks. That's good, even if you don't think it's possible for these models to have meaningful CBRN risk. And if DeepSeek drops the weights of a model that does have high risk in one of these areas, it's useful for us to have language to discuss it.

From a business perspective, from a business analyst perspective, the industry leader doesn't benefit from open source or open weights. This is part of why Google didn't open source their infrastructure the same way Facebook did. It was a strategy credit for Facebook. The same is true today. It's a strategy credit for Facebook to release their Lama models. It seems like Deepseek is interested in open source, not from a business strategy perspective, though, So that's pretty interesting.

It may be that as long as Liang one Feng is in control, they will release the weights to all most, in parentheses, of their models. So, Ben. You don't need to frame it. This is, like, first off, William, fantastic e-bell. Like, like, the rant deserved a rant. Okay. And you delivered.

So very good. I already gave you credit up front for the open eye versus Google. Great point. To jump to the end, the business analyst perspective. Yes, correct. It is, you see from the distillation bit, what's the easiest way to avoid distillation? Just get the weights, right? And I would also add on from a deep seek perspective, it makes perfect business sense to do open weights and open models for the same reason it made sense for Facebook to do open infrastructure.

And even from the counter argument to they should have kept the secret so the U.S. would know about the efficiency gains. The benefit is they did just spread this knowledge to everyone in China. who is dealing with severe chip constraints. So, like, all this bit sort of makes sense. So I actually grant almost everything in this email. I will just explain the source of my rant. That was not a business analyst rant. That was a, did it confirm some of my takes?

Yes. And my takes is it has deeply, there's this tension between, using the language, there's several things going on, using this language of safety and the world ending possibilities of AI, my frustration is to the extent that is a real concern for you.

And to be clear, I'm not flipping about this. I completely grant the concerns that's part of what drives my irritation. It is detrimental to your cause to expand the definition of safety to mean a large number of things which we may find objectionable, but which are not safety sort of oriented.

So when it talks about like, and that's why I went to like the GPT2 sort of notes talking about like misinformation or or or biases or whatever it might be, again, those are legitimate things to have a discussion about. But the way it manifests an actual discussion is you push back against that and you're accused of not caring about the world ending. And that's the Mott and Bailey sort of aspect of this argument, where this is an argument of fallacy where you go out to the Bailey and you make sort of a point, someone objects to that. You retreat to the Mott and say, what? You're talking about the world ending?

You don't care about this? You don't care about safety? Yeah. And this might be motivated by my own personal position, which is, I promise you, every time I write about this, I get a lot of pushback about me not caring about the world ending. And so there is absolutely personal frustration about this sort of conflation that happens. Number two is the overall structure of Open AI. And I've looked for this for ages. It might have been like some conversation.

I don't remember. But there is a real element of even this board structure and this idea. If you really believe that AI is going to control the world, implicit in that with this structure is we will control the world. Right. Right? Because the board's going to make the decision. And we will decide what to release and when.

Yeah. Right. Exactly. And so there's just a, if you take these arguments seriously, then I have real concerns. Right. And now, to the extent this, the first point William made, double down their existing views. Yep.

Absolutely happened. One of my existing views all along is that this is number one. If this does become dominant, I'm very worried about Sam Altman deciding the course of the world. This isn't a shot at Sam. It's a shot at any one person having that degree of power. I feel like that's a reasonable stance. Yeah. Number two, I don't think that's going to happen.

I think there's going to be multiple models. And the best way to, this is very cliche, I admit, but I believe it, the best way to fight against this, to defend ourselves against this is to have more AI, not less AI. If we could go back and put this all back into Pandora's box, sure, that's fine. My belief is that we crossed that Rubicon a long time ago. It's going to be out there. It's going to be open. And R1 emphasizes that point.

This is the whole basis of the chip discussion, right? It's like China's going to get this. They're going to, and they're going to get enough chips, and they're going to figure out how to build stuff over time. And you're fighting a losing battle that, by the way, is just the entire, like, wrong mindset of playing defense instead of seeking to innovate. Right. Which, again, totally reasonable to disagree. I loved our discussion on that last time, in part because I feel like I could flip around and argue your point just as well.

It's a discussion that needs to be, needs to be had. But yeah, this is a prior of mine, which is that actually AIs are going to be broadly available, given that if you think that this is, if your number one concern is truly safety, then and you're saying we're a nonprofit, we're going to do this for the good of the world, then you should be open sourcing it. Now, to his point, from a business analyst perspective, of course they should not be open sourcing it. But that means they're being big hypocrites. If they're big hypocrites, I'm going to yell at them.

So again, I think when it makes all reasonable points, that segment, again, It's a podcast. We get a little spicier on the podcast than necessarily I do on trajectory. But that's a reaction to what I perceive as the hypocrisy in this position that if you take a very sharp analytical edge to it looks like trying to capture power

for a very small coterie, coterie. I can read that word. I can't pronounce it either. Somebody in emails let us know. But my question is, would open sourcing open AIs models and open the weights, that is, that would help because it would disperse the innovation and allow other Americans to innovate on it? Is that right? It's like it's like nuclear weapons and mutually assured destruction, which again, would we be better in a world with zero nuclear weapons? Well, actually, it's an interesting question.

It is an interesting question. I mean, on one hand, I was pretty annoyed by the end of Oppenheimer because they made it sound like we unleashed some horrible era on the world when in fact the nuclear bomb was invented and there's been more relative peace than at any time in human history. Exactly. Exactly. This is like the whole thing about the U.S. China economies being so intertwined. Like, yeah, this is really a bad thing.

And they're like, well, maybe it's the only reason there's not a war right now. Right. Like, no, with the U.S. China relationship is mutually assured economic destruction. Like that, and like there is in any, in a convention. conventional world with conventional weapons with and and sort of non-integrated economies to the level we are now, I think almost without question there would have already been a war over Taiwan. Right. Like so like like so there now the problem and the very reasonable pushback is okay. If you want to take my arguments put back my face, you might be right in the short to medium term. But if you think about the long term, all we're doing is setting us setting ourselves up for a truly world ending destructive scenario. Yeah. And my response to that is, you might be right. And it's the same thing with AI.

It's a lot of knock on wood right here. We might be right. But the problem is AI is out there. It's a real thing. It is going to be diffused in my opinion. And given that, the answer is not to try to put it back in the box. The answer is to run in the exact opposite direction. And again, from a business analyst perspective, Williams totally right, opening eyes right to close down, to not be open.

just save me the sort of, save me all the self-righteous sort of posturing. Yeah, exactly. That's what I was responding to. That's exactly. That's what I was responding to, just to be clear. All right. Well, one more bit of Sam news. Ali says, did you guys see the tweet celebrating Sam and Sotia's rapprochement?

Guess they're open to a soft landing now. We got lots of emails about this specific tweet. Do you have any? The selfie? Yeah. Do you have an official comment? On the sand and satcha selfie. There's a bunch of funny comments.

I think my favorite one was something to the effect of when your parents come to tell you it's time for dinner after they've been arguing in another room for six hours or something like that. Feels right. Great stuff all around. No, look, Microsoft is looking like is looking amazing in many respects right now. Like they have the right of first. They still have full access to all opening eyes. APIs. They have the right of first refusal to the compute. And they're also sort of letting themselves off the hook for pursuing the funding of these leading edge models that get commoditized

within months. And so like that seems like a reasonable place to be. Now again, the open AI bet, which again is reasonable for open AI. And the soft bank bet, which Masayoshi's son is not reasonable. So I guess it makes sense for him and whoever ends up funding him is no, actually we're going to get there first and it's going to dominate the world. And it's going to be the economic returns to one model is actually going to be a thing. Now, again, I think the evidence is tilting towards their models being a commodity, but it's still reasonable to bet in the other direction. Is it, though?

Is it reasonable if you, because, like, we just talked about how at the top of the show, like, is there any defense to distillation? No, it's reasonable if you have a, if you have the, the, the appetite for the risk that entails, right? Like, maybe there's only a 5% chance. but the upside of that 5% is astronomical. It's a venture capital type of bet where it's probably not going to be the case. But if it is, the returns are so large that it's worth putting money towards it. Again, this is why the debt part doesn't really make sense to me.

This feels like an equity funding sort of situation. But if you're Open AI, I get why you would go in that direction. And if you're Microsoft, I get why you want. And again, it's not like they're cut off from Open AI's models. They still have leverage that they have exercise. size to make sure they still have access. So, well, they're literally putting on a good face for the world this week. And I enjoyed the selfie and I can't wait for future updates on the Sam and Satya relationship.

Yeah, just glad the parents aren't arguing anymore. Oh, man. It's time for dinner. I'll tell you what. I am more open to open AI that I have been in the past, in part because I use chat GPT on a daily basis at this point. I will say if there's one person in tech who is most likely to be the subject of like eight different documentaries that are sold to rival streamers at some point in the next

five to ten years, Sam Altman is like the runaway favorite to be the subject of all those docs about corporate abuse and corporate chaos. It's going to be great. I mean, yeah, it's very reasonable to assume and arguably to make the case in a has happened in episodes in the past, that Sam Altman has an icarus sort of fetish. He will always fly too close to the sun. So we'll see if it, we'll see if it happens in this case. The adventures will continue. Peter says, Deepseek said their final training run of V3 was accomplished with a cluster of 2048 GPUs, a far cry from XAI's apparent 200K cluster,

assuming that compute needs are unchanged, but are utilized using smaller clusters. than previously assumed. What are the impacts on things like networking, power distribution, liquid cooling, and other areas that have been gating factors to creating large clusters? Ben, do you have thoughts there? I think that could probably be like a 45-minute conversation,

but what comes to mind? Well, I would say with Deep Seek, there's this report, I think Dylan Patel originally said, like, back in November, they have 50,000 hopper GPUs. So number one, that was misconstrued to be H-100s. H-800s are also hopper GPUs. The difference is the constraint memory bandwidth.

So first off, the 2048 was almost certainly that's the biggest they could go given the memory bandwidth constraints. Because you have to communicate all these CPUs and keep them in sync. So would they have preferred to do a much larger bit? Yes, they couldn't because of the Nerf memory bandwidth. So is bigger going to be better? Almost certainly. Why are, like I think GPT4 is like somewhere, I'm 24,000 or something? was trained on.

Maybe I can't remember what the exact number is. Maybe that that's even high. The constraint isn't that they didn't have more GPUs. It's that the communication overhead becomes overwhelming. The whole thing falls apart sort of the larger that you get. And so a huge part of this expansion is increasing that capabilities. It's a big part of Nvidia's sort of moat is they're the best at building these. It's not just the core AMD chip is faster than the core Nvidia chip.

Nvidia is just a gazillion times better, number one, at programming for them because of Kuda, but then number two, they're so much better at tying all these chips together. That's right. And so deep seek would have been better on more hardware. Number two, again, this was just the final model run. You get to a final model run that is so cheap and runs so well because you've done a gazillion other experiments and runs along the way. So their researchers are using a bunch of GPUs all the time to even get to this bit to do the final run. Number three, they've been serving inference at shocking lowly low prices, which is arguably the biggest indicator. Like, they completely destroyed the pricing model for inference in China like a year ago.

So that's where most of their GPUs are probably going, people accessing it and needing to use it. A lot of these huge data centers are for people to actually run inferences, not just training. The other thing about this, and I should have made a bigger point about this, I think, in my FAQ, is one of the big implications of the power of distillation, number one, and number two, these reasoning models and their ability to generate chains of thought that are useful is the capacity for AI to train AI is the, the, big barrier is with around GPT4, we got to the, we've surfed the whole internet, right? How do we get more data? The synthetic data question has always been the opportunity and always a question, is it going to work? It looks like it's going to work.

And so you think about it, if you want to get, you know, do the question of, we talked about this the first podcast of the year, is aggregation theory dead. Like, is there a marginal cost to using AI that's going to mean that the way we thought about internet economics is going to change? Well, the counter has always been that regular LLMs are getting way cheaper and they're really good at a lot of stuff. You're not going to use a reasoning LM for everything. But the other thing you do a reasoning LLM for is you can generate that many more answers to that many more questions, a basically infinite number. Especially with reinforcement learning gets into it.

And so you just, you're generating more and more and more synthetic. data, which it's not a direct distillate, and you use that to train regular LMs, which just poop out the answer. They don't think about it. And suddenly those are way more capable. And so you think about it from that perspective, the amount of potential knowledge in the world is infinite. We're talking about bitter lesson, brute forcing the creation and acquisition of all possible

answers in the world, right? Like, and that is basically... So there will be infinite inference demand as we go forward here. Well, this is, is this training or. inference. This is inference for the purpose of training. So the self-contained need for inference, and this is where the $100 million open AI thing makes sense. If they see this is the path, we have a reasoning model that generate chains of thought that can then be used to train regular vanilla LLMs. Their potential appetite for compute is actually larger than ever,

because they want to train the reasoning model, then they want to use. use that reasoning model to generate all this synthetic data to train the base model. By the way, once the base model is better, you get a better reasoning model so you can generate more synthetic data to make the base model better. We have entered this virtuous cycle, which to the credit of the people worried about capital S safety, like AI taking over everything, this is what they predicted. The AI is making the AI better. And we are in that era.

And that era is going to demand a basically infinite amount of compute. Okay. So the reaction is overwrought. Again, I think there's a legitimate question about timing questions, right? Is there a bit where, you know, all this demand, particularly for inference, if we get way more efficient, might not be consumed. But the potential for that much more compute needed for training and inference and inference that goes into training, is a huge thing. And so, you know, there's going to be a massive demand for power distribution, liquid cool.

Yeah, right. And all this stuff goes into, yeah. It's all part of it. Right. Yeah. And by the way, there's been such a flood of stuff coming out of the Trump administration. It's hard to keep track. But one thing that that sort of slipped under the radar is there's going to be executive order for creating power that is off the grid. That's basically tied directly to these data centers.

and a dramatic loosening of, this is basically what we called for on this podcast a few months ago. There needs to be the capability for dedicated areas where you can cut through all the red tape and you can create power, and the power can be tied one to one to the data center. The data is very predictable how much power it needs. And it's completely off the grid.

You could build these anywhere. You could build at the middle of the desert, right, with a huge solar farm if you wanted to. The problem with solar is it's intermittent and at night it goes away. And so you need a lot of batteries because you need continual power. This is why nuclear is a natural fit

for these data centers. But you could build that and then just have a fiber optic line that sends the data sort of somewhere else and you don't even need to all the grid stuff you can completely bypass. So the appetite here is larger than ever. Great.

There's actually an unlock of appetite. Okay. That's why, sorry, just one more. I know I'm monologuing a little bit, but that's why my initial write-up on R1 was about the reinforcement learning, the synthetic data, and the distillation aspects.

Because that's the unlock. That's the real takeaway from R1. All this other stuff is kind of a distraction. No, exactly. And I came to R1, having read your piece, and then seeing the hysteria the last week or so, it was all a little bit befuddling to me. I do think you nailed it with the comp to the Huawei 7 nanometer chip,

where it was just like a lot of people who didn't follow AI that closely, we're digesting this news and freaking out about it. But the synthetic data and what that suggests for the future is a massive deal. So strap in, it's going to get pretty weird here as we all continue to learn the bitter lesson. As speaking of the Trump administration, though, a couple follow-ups on the export control conversation. Two questions from Thomas. The first is, how would no chip controls on China have impacted the supply of GPUs for American companies? It was always said that the restriction on compute was how many GPUs, TSM, and Nvidia could make.

If China had also been in the market, how would that change the amount of GPUs that could be delivered? That, I have no idea. I'm just curious whether you have any take on how that might have shaken out. Well, the question is, did it shake out at all? China seems to have beginning plenty of GPUs, number one. You had the H-A-800. A lot of GPUs going to Singapore.

Singapore is buying more. GPUs than they have electrical capability to power, right? Like we know where those are going. And I think this is all a very fair question. And I think one of the more powerful critiques of invidia that you can leverage is that invidia does not operate like an American company, which again, that's kind of the standard in tech. You're just sort of global and you sell to whoever you want.

But, you know, has there been a situation where Nvidia can basically decide who gets their chips? Yes. Have they leverage that to maybe not give the big tech hyperscalers as many as they want because Nvidia realizes that's a danger to them in the long run that they, you know, they don't want to give too much power to end up in a sort of monopsony situation? Yes. Has a lot of that gone to startups like CoreWeave and things on those lines that Nvidia is also invested in? Yes, has that entailed selling a lot to China when Amazon or Microsoft would have been happy to buy those chips. Also, yes. Invita's hands are not clean in this affair. Yeah. And I'm not necessarily worried about Nvidia, but it's pretty funny that Nvidia is the company that has spent the last several years

fighting like crazy to be able to sell their products into China. And in the past two months, they became the subject of an antitrust investigation in China and then had DeepSeek use nerfed Nvidia chips to make a model that somehow wiped out 20% of the stock's value. So, could be a sort of chickens coming home to Roos situation. But again, I think Nvidia is going to be fine, all part of the adventure with that company. Thomas Part 2, I remember hearing about how Smic had stolen IP and people from TSM and ultimately had to pay out a settlement to TSM because of this in 2009. So the Chinese people and government have had a focus on semiconductor manufacturing capabilities

for a long time related and as noted on Monday's episode, a huge blocker on the leading edge has been the lack of EUV machines. So I'm wondering, does Ben think this is also a bad idea and that all current machines and chemicals should be sold into China? What do you think? I was curious too whether that falls into your take. Well, I don't know what my take is, to be totally honest.

Like the, I've always, like you go back to ZTE, and Trump banning them from chips. And I was immediately very uneasy. And I was uneasy for these questions, which is you're setting up a long term. Like there's a card you can only play once. You're setting up a long term problem for U.S. leadership in this area. The game theoretical consequences for what this means for TSM in Taiwan is very problematic. Like I'm not saying it's going to lead to war, but it's in.

increases the potential for it. If China's dependent on TSM, that's a very good reason to not invade Taiwan. Right. If they're cut off from TSM, their cost-benefit analysis is fundamentally different. And so, again, go back and read what I wrote eight years ago. I've been very uneasy on this point all along.

And so simultaneously, the EUV bit, the reason why, again, I have so many hats on this, including a deeply self-interested hats. I'm trying to be sort of honest about this, is that is a part of passing these laws is can you actually enforce them? And that was a very clear line and enforceable point of a break in the chain that you could stop EUV machines from going to China.

That was a fundamentally different technology. It was hellishly difficult to develop. The idea of EUV goes back to the mid or early 2000s, maybe even late 1990s. It took 10, 12, 13 years and multiple rescues of ASML to even bring it to market. Right. And can China reinvent it? Yes, they can eventually. Like, if you know something's possible, you can do it.

But it's like, this is actually a place that you can draw the line. And I have been much more uncomfortable with the. with the, we're going to, like, oh, they made us, they used DUV, which we let them by for years. Like, what are you actually trying to accomplish here other than handicapping sort of U.S. companies sort of in the long run and increasing the prospects of competition and losing your crown jewel in the very, very long run? This is very valuable leverage. Is this the right place to sort of trade away?

And from a China perspective, I said that Huawei 7 anime was bad for China because the longer, they stay on the let's get on the leading edge what we have the worst when the best thing they can do is go back to basics and build up from ground zero like all this time

long has been that you have to start with the trailing edge and build up node by node in order to develop. The pushback is China's doing that and I just made the case before companies should dual track.

It's like I fully admit I'm arguing against myself on a lot of this stuff. I still don't know if the chip ban is a good idea. And the thing that this week has sort of brought back to me, I was arguing against it with you. Yep. And not fullheartedly, because I'm still not totally sure, I am worried from a cultural perspective about the U.S. What signal it sends to ourselves. Yeah, that we're going to fight through blocking our past innovation instead of filling the

incentive and motivation that we have to run forward even more quickly. and where do I land? I land in a pile of mud. I freely admit, I think it's sort of the ultimate takeaway here. Well, did you read the post from Dario Amade, the Anthropic CEO? Because he wrote pretty articulately about export controls and arguing for enhanced export control enforcement. What did you think of his arguments there?

I mean, it's a very valid argument. And I think you've been making the argument and you made it very well. And if people agree with you and agree with him, I get it because I might agree with you as well. I'm down you're in the mud. Yeah. Grease pig. I know I appreciate it. Well, and it's interesting to me because in a political context, I have no idea how the Trump administration might handle this.

Because obviously it's an unbelievably sensitive issue to the Chinese. Let's get the Trump angle in a second. Okay. The issue and concern I have with Dario articulating this, he's the one that wrote the GPT2 post, right? Like, and there's a, I just, I'm instinctually a little suspicious of these folks that are simultaneously saying AI is the biggest danger. It's a danger to humanity and also devoting their lives to building it. Yeah. I was going to say, and again, if you want to, if you want to, if you want to like our email earlier, throw it back.

in my face and say you're just arguing your preexisting priors. Yeah, I am. But that is my preexisting prior. It makes me concerned and suspicious. And it is deeply in their interest and opening eyes interests to block this sort of thing. But yeah, that's where I'm at. Okay. Yeah. Well, and what I was going to say about Trump is I, he's managing a much bigger relationship with Xi Jinping and the Chinese. And so I think there are factions of his security team that would want to enhance and expand the export controls on chips and chip making equipment in particular and get some of the allies on board, whether it's ASML or South Korea or Japan. And I don't know whether Trump, Mr. Executive, we've talked about the broadening power. I don't know whether he's willing to ruffle she's feathers

over the next year or two here. And maybe he shouldn't, you know, because I think it is a bulwark against continued tension heightening. So we'll see. Well, the thing that is worth keeping in mind, and this sort of takes this full circle to why it mattered that Deep Tseek was Chinese and it's actually the Chinese factor. I think your analogy to Mistral was a great one.

Like if Mistral did this, everyone would be falling over themselves with delight and glee. Yeah. Wow, this is cool. Yeah. It's an excellent, excellent point. And there's a bit where I mentioned maybe this is a good wake-up call. I'm not sure that people in the U.S. have fully internalized how precarious our position relative to China is, particularly in a warfighting scenario.

And so this is just in general, their industrial capacity, the number of ships they can build, the number of – we can't – build stuff. We can't build ships. We can't do X, Y, Z. Like, could we spin that back up? Eventually, uh, we didn't spin up artillery ammo very well in the, you, we're still producing way less than Russia is. Like the, the, there's, I'm not sure it's fully internalized how bad that is. And also how much worse it's going to get in a new type of warfare defined by things like drones and robotics. All of which. the components is completely housed in China. Like all the little actuators and

motors and batteries and all these sorts of things are dominated by China. China, if this is a world we end up in where that's what defines a war, we have a big problem. And there is probably

an infusion of humility that's necessary. And if it comes in the maybe overstating what Deep Seek did, but puncture the attitude of presumed technological superiority. When it comes to fighting a war, fighting a war is not aggregating users

in capturing demand. Fighting a war is actually expending physical goods and actual people in a war of attrition. Again, my aggregation theory is not worth very much in that scenario. In the U.S., everything about U.S. tech in our economy as a whole,

is all about consumption. Like, like, we rule the world economically through leveraging our willingness to buy lots of crap. Like that, like,

at a very sort of fundamental level. And in that context, I don't want to give Trump too much credit because who knows what's like, he likes tariffs, he likes thrown out XYZ. What I recognize, though, is there's a real arrogance

in the way we deal with this sort of stuff. And an assumption that the world, as it was 30 years ago, is the world that it is today. And it's just not true. And is it actually wise to keep operating with a level of arrogance and that we can dictate to the world what it is or what it isn't if we increasingly don't have the goods to back up our talk?

And so when you think about something like Taiwan, like people are flipping out over this potentially applying tariffs to Taiwanese chips, I already wrote about this as a trajectory. I mean, I'll get to it again. but I kind of made my point that Trump's critique of the Chips Act that just subsidized supply was fundamentally wrongheaded. You needed to subsidize demand. Tariffs are a way to do that, but I think a less effective way than just straight up making buying guarantees for Intel and buying chips.

But it's not insane. When you think about what is necessary in terms of driving demand, there's also a broader point that our total dependence on Taiwan is a big problem. like the and if the Taiwanese government you know they completely rationally don't want to allow leading edge TSM capabilities to go off island because it is their Trump card no pun intended to make sure the U.S. comes to their defense but how long is that sustainable if there's very real questions about our capability to fight a war yeah no exactly again to the analogy for Trump has always been a bowl in a China shop and there's real questions to raise about his ability to put

stuff back together. But sometimes there is value in stuff being broken. And one of those taboos that might be worth being broken is the total shelf assurance amongst most of the U.S. that we can win a war. Right. And in that world, is there, is this sort of screwing Taiwan? Maybe.

Does TSMC benefit from a manipulated currency or, or, uh, or, uh, or, you know, or, uh, is this sort of screwing Taiwan? or a currency that doesn't seem to quite reach a right level that gives them a fundamental cost advantage making chips in Taiwan? Maybe. Like there's like the hands are not clean on all sides here. There's a, there's a tendency to put like Taiwan is an ally.

They're not 100% in line with US interests. Yeah. And, and I just think there's a, again, I'm not saying I agree or disagree yet. I'm still really thinking about it. But I do think there's a real paucity of, humility and awareness of the situation that we're in right now. Yeah.

And there might be an argument, by the way, that we need to lean into it, that we should come to some sort of agreement with China that locks them in as they're going to make stuff and we're going to buy stuff. And what does that mean for Taiwan? Well, if we better have our own chip making possibility because are we going to like, like the worry about Taiwan being a part of China is then China now has concluded. cut off our chips. Right.

Well, then we should figure out how to make our own chips domestically. Because that's the exact same worry of a war scenario, which is that China blows up TSM. Then we still need to make our chips domestically. Like, it's very rational. Again, I love Taiwan. I've lived here for 21 years. Taiwan's democracy is amazing.

You know, I would love to stay in this gray area where it's functionally and independent country. I do feel truly free here. Well, COVID was a little, little iffy. Yeah, I tested your patience. But there's like that the job of if you're concerned about national security, if you're concerned about these long term things, you have to think about these issues. Yeah. That's why I wrote the geopolitical chips article years ago. Like, like this is a real problem. Right. Well, and I don't say it trivially when I reference Trump's executive discretion. I sort of made it sound like a joke about five minutes ago. But I mean it if we don't go the direction that I imagine a significant portion of his administration wants to in terms of expanding export controls,

there may be a reason. And that actually may be a rational decision because it will absolutely infuriate the PRC side. The relationship will devolve further. And it's not lost on me that the PRC has been in the midst of a manufacturing buildout that is not paralleled in the last hundred years beyond like pre-World War II Germany and early 20th century the United States. Like they're in a much better position to fight a war than the U.S. is, at least from a manufacturing standpoint and shipbuilding and all of it. From the U.S. perspective, the optimistic take is, well, our AI is going to become so good that we can technologically win.

And that's the argument for the chip controls, is that we will overcome the manufacturing deficiency by virtue of having superior technology and AI. And that's why, yes, we're throwing away, we're playing our one-time play Trump card now. Again, no pun intended. But this is the time to play it. And maintaining advantage now will then compound in the years to come. That's right. That was part of the logic.

Because the AI is going to make the AI better. Right. So this viewpoint, the funny thing is, is that the chip ban, if you dig down, if you dig down, to the fundamental assumptions undergirding it is actually the same as the Sam Altman view or the Dari Amadee view, which is we're not going to have commoditized models. We're actually going to have takeoff and we're going to have a sustainable superior advantage that's going to maintain over time. That's why the deep seek thing's a big deal.

If that's not the case, how many decisions were made with that assumption that might not be true. Yeah, well, and would Deep Seek exist if not for nerved Nvidia chips that had been sold into China is an open question too. Well, I mean, to be clear, bite dance and Alibaba, they all have pretty decent models. So like, you can. But are they working with Nvidia chips or are they working with what Huawei has developed? And Huawei has been buying up chip making equipment for years, much to the dismay of people in the security community who think that you should be much tougher on Huawei. I don't know. Well, I think most of the trainings probably happen in Nvidia chips.

I think that's where, frankly, a lot of the smuggling is going to. And there's cloud options too. I think Huawei, yeah, I think for inference, it's more plausible. Inference, you need less of that general bandwidth. You're more constrained by memory anyway. But I don't know. It's hard to say for sure. But the other thing is you do have this constraint of power and data centers,

which entails the ability to build stuff. and China is better at that than we are. And so there's a bit you can just sort of brute force scale that operates inefficiently, relatively speaking, and it's okay because particularly when you think about it, at the end of the day, the military is going to get all the best chips. So even if you constrain private industry,

it's not like, you know, so. Don't figure it out. Again, I can make the argument for the chip band as well. Can I read one more email on the chip? in. Do you have time here? Yep. All right. Michael says part one of Michael's email is first, Andrew could not be more wrong with his comment. Oh, finally. Someone that takes my side. That somehow the chip export control. We have to raise questions about you, you choosing these emails. I was wondering. Oh my God. Listen, I just wanted to give you a chance to respond to a fiery rant.

Michael says, Andrew. No, that's a great rant. That was actually, that was the best emails we've gotten a long time. I will give credit where credit is due. But you know, Michael's turn. Terrific energy. Andrew could not be more wrong with his comment that somehow the chip export control regulations were watered down by industry and that's why they don't work. You spent a lot of time talking about cope on the last episode

and that's some hardcore cope. The regulations... I just want to... The cope is such a great word. It's been a great, great addition to the discourse. I'm glad we're helping... Oh, my God. My only request,

honestly, the only reason I read this email is because I want to make a formal request to all our listeners, to tech Twitter, to you. Can we all just take a two-week break from saying the word cope? It's everything. No, we cannot. We could not. I love it.

It doesn't have to be a permanent break. Just give me two weeks. Take a breath, everyone. All right. Michael says, that is some hardcore cope. The regulations were completely dumped on industry without any engagement or opportunity for feedback. That's especially true of the recent AI diffusion and foundry rules.

I don't know if they'd be better, worse, or whatever if industry were more involved, but industry was absolutely not involved in any meaningful way. This email made me smile because it reminded me how different the audiences are across the various podcasts we host in the bundle. The Sharp China audience, many of whom are in D.C. and in government would have a very different take on what happened to the chip controls. But I'm sure there are lots of people in the Sharp Tech audience who feel like the government was dumping all this on them without having any idea what they're trying to regulate.

I'll just say what I recall is that there were efforts to get Greg Allen removed from his post at CSIS because he was too competent at explaining the areas in which the chip bands were failing to achieve their intended purposes. And in general, the lobbying behind the scenes has been pretty well chronicled. Semi analysis actually did a great job in October, pointing out all. the different ways the chip controls had failed and why they were strategically important. But to Michael's point, I think the resulting incompetence is ultimately the responsibility of the Biden administration. And I also think he's referring to the AI diffusion and foundry rules.

And I know less about that process. That probably was dumped on industry without any meaningful involvement and engagement. And I don't think those rules were the right way to approach any of this. So on that point, I can agree with Michael. Do you have any thoughts? Yeah, I mean, I think the, you go back to even also to the Biden executive order on AI. I just want to double down on that bit I made where they became like true believers in this AI takeoff scenario. And it's like, we have to do everything possible to stop this. And that's why this is an important point. And the question of model diffusion is a really important point to talk through. It's meaningful. it's obviously super meaningful from a business analyst perspective, right?

Where does value accrue in the value chain? But it's actually undergirds these very deep fundamental questions. Like, are we going to sacrifice our sub-acductor industry in the very long run because this is the one time that we have to get it right? Yeah. Right? So that's one point. The other thing, it's not that you and Greg Allen and semi-analysis are wrong about these chip controls.

Yeah. It is worth considering the extent to which. the chip controls being leaky is a pressure valve, right? That prevents us from actually playing these scenarios through to the end. And we do need to think deeply about what happens if we had perfect chip controls. Right. If China actually did not get any chips. What is the rational response of China in that scenario?

Yeah. I mean. And it's not a pleasant one. Yeah. And that's what I mean. I think I wonder whether Trump is considering that as he weighs all this, as things are for. Who knows?

I'm not going to try to jump into that mind. But Michael, part two. Ben is correct on ice engines. I had a fun opportunity earlier in my career to get the red carpet rolled out for me at Daimler and BMW. If you go into the Mercedes-Benz Museum, you go on this long elevator up to the top floor,

and when the door opens, you're in a dark room with a spotlight shining on the first ice engine, the company produced. The key technology that they had was the engine. The same was true of BMW. The engine is the key technology. Everything else flowed from that. It's also true of Honda, based on my understanding of the company's history. And I also think it's true of Ford. He developed an engine first that none of them have adapted particularly well to the EV environment isn't surprising, given that their key differentiating technology was fundamentally disrupted. Ben, I include that

part, only to say that I would love to go on a tour of Daimler and BMW, get the red carpet pulled out for us. I feel like we need a SarpTech field trip here. I'm very jealous. Let's go to Frankfurt sometime in the next year or two. But thank you for the note, Michael, and thank you to everybody who wrote in. We did, we got like an avalanche of email. So this is a bit of a longer episode, but we got through as many as we could. You were efficient, you know, taking your cues from Deep Seek on this episode. And here we are. We will be back next week. And it's not going to be wall-to-wall deep seek next week.

But Ben, don't promise things you can't guarantee. That's true. Enjoy the weekend. And I will talk to you soon. Talk to you later.