If we jump in a time machine and we go five years forward and then we look back I would not be surprised upon stepping out of the time machine if someone said to me, "Yeah, all this stuff happening in 2026, that was what expanded into a broader war," which, you know, could legitimately be called the Third World War.

So you could have intensifying conflicts between Russia and the Europeans, via Ukraine. You could have a broader conflict in the Middle East where the US and Iran you know, there's a spiral toward more and more intense conflict with the Gulf countries getting sucked in, being damaged, maybe Israel being damaged.

And, you know, maybe that's enough to call it a world war. And of course, at, at the level of, like, hybrid war, there's definitely already a hybrid war between the United States and Russia and China because US is trying to contain so sanction China. I mean, sanction Russia, wanted to bring Putin's regime down through sanctions.

That didn't work actually, but that's actually what we wanted to do. And the Ukraine War was a good pretense for imposing those sanctions. Wanting to contain China's technological development, cutting them off from all advanced chips, hoping that we would win the AI race and contain them that way, hoping that we could maybe stop the oil flow to them from the Middle East.

You know, so, all of that you could lump together and call, like, World War III. I think that would not be unreasonable. Now, I think the one silver lining to what's going on right now is I think the US has bitten off a lot to chew in the Middle East. So if things do spiral out of control in the Middle East, like you could, we could do this just by counting stockpiles of major weapon systems, it will eat all of what the US has.

We've eaten half of it roughly , so we, we could eat, we could eat more if the conflict widens in the Middle East. But that means the US really doesn't have gas in the tank or weapons in the stockpiles to do anything in the Western Pacific.

Alf: Hello, and welcome to a brand new episode of Seeking Truth From Facts. Today, I'm joined by Steve Hsu to discuss the geopolitical situation from the war in Iran to Chinese frontier AI. How are you doing, Steve?

Steve Hsu: I'm great. How are you, Alf?

Alf: Not bad at all. Great, great to be back.

So to, to begin the we would beremiss of us not to begin with the ongoing war in Iran. And it appears to be increasingly rudderless in terms of an a drive towards any kind of end goal, and driven in large part by a massive sunk cost fallacy. Do you see any realistic off-ramp for the US in its ongoing war in the Gulf?

Steve Hsu: I feel like it's up to Trump's temperament how this all ends. I don't see a way that the US can open the Strait of Hormuz. I don't even see how the US can open the Red Sea now that the Houthis have closed that as well.

I don't think we're gonna mobilize, or it would take us actually quite a long time to mobilize a large ground force. I don't think we can, reduce just through air attacks the Iranian capability to close the Strait of Hormuz. So I think our options are quite limited unless we wanna really escalate in a very, very ugly way, which might even beyond, be beyond our capability in the short run. Like, we would have to, you know, pre-position a lot of material before we could actually escalate in terms of a ground invasion. So I think our options are quite limited. It's sort of the recognition among Trump and his team that this is the actual reality of the situation. You know, the thing that's gonna happen, I think, before then is a real oil crisis. I think the strategic reserves are running low. Nobody knows exactly when they'll run out, but when that happens then, you know, the pressure will just ratchet up another notch.

Alf: We'll come onto oil shortly.

You among many have commented on how the current war is severely depleting the American missile stock. Could you maybe go into more detail about this and its implications for the US-China competition?

Steve Hsu: Yeah, I was just looking into this. And so there are published reports about this, and there have been leakedconversations, you know, between the Pentagon and the White House about this. Roughly speaking, I think we've used some, depending on the specific system we've used something like 30 to 80% of our total stockpile of many important systems ranging from Tomahawk missiles to, you know, THAAD missile interceptors. That's a material chunk of our total stockpile, and used in this one regional conflict, which was a you know, a war of discretion which Trump didn't have to actually start. So it's kind of crazy that we did that. That accounting doesn't even take into account the destruction by Iran of huge radar systems and such on our bases, our base infrastructure in the Gulf Coast countries. So, I mean, the, the cost of this war has been huge. In order to replenish our stockpiles of these key systems not talking about the radar systems and such in the Middle East, that maybe we can't do at all because of some limitations on our access to some rare earth elements.

But just to replace the missile systems that we've depleted you know, it looks like in the, in, in the estimates I've seen, it would take at least a few years. So we wouldn't be back to the strength levels we were at prior to the Iran conflict until roughly 2030-ish. So that's quite bad. So that, that leads to a window of vulnerability where the Chinese could, if they, say, moved on Taiwan, we would be in a really tough situation in terms of our weapons stockpiles.

Now separate from that, there was a CSIS war game, which, you know, wasn't completely realistic in many respects, but I think it was reasonably realistic in terms of just counting up how many munitions we would expend in the first few weeks of the war. And I think they had us completely running out of our stockpile within the first few weeks of the war.

So, you know, now probably would run out in half that amount of time, so we could only sustain operations, you know, high tempo operations in the Taiwan conflict for, you know, roughly 10 days or something. So we're in a very bad situation and, and all this was discretionary. We didn't have to end up here.

Alf: And, and you, you mentioned as well earlier that the, the depletion of America's strategic petroleum reserves, that, that I believe they're at their lowest levels in over 40 years, and they're, and they're shrinking shrinking rapidly. Do you see, do you see a risk of a kind of serious 1970s style energy crisis?

Steve Hsu: Well, I don't know if it could reach the levels of what we experienced in the '70s. I was actually a little kid when this happened. I have a vague memory of, like, really long lines of cars in the, on the road to get gas, and that was a real, you know, for, for 10, 20 years, that scarred the memories of all Americans, but people today can't have never experienced anything like it. I'm not sure it'll reach those levels of catastrophe, but definitely we could have a recession. We could have, you know several points of GDP knocked out by you know, shortages of petroleum-related products and maybe even worse. And I think that's something Trump would want to avoid.

And I think that is likely to hit before the midterm elections. So in other words, unless Trump is able to con the Iranians again into, like, stopping the conflict and letting the oil flow again, which I think is not gonna be easy for him to do this energy crisis will hit us before the midterms, which would be a disaster for the Republicans and for Trump.

Alf: How does the situation with the strategic petroleum reserves, how does it compare to China's situation? 'Cause obviously China's a, an energy importer. It has the world's largest SPR stock. So, so what's the difference in those two situations?

Steve Hsu: Yeah. So as far as the US strategic oil reserve, you know, I'm not an energy trader, but I actually know some people who are energy traders . And the typical thing you were hearing from those guys prior to the MOU was that we had kind of a month of stockpiles, you know, a month could go by before we really were in trouble.

and so since the MOU has kind of fallen through, we're kind of gradually approaching that one-month threshold. I think we're gonna get there pretty soon. Definitely this summer we'll get there if, if things are not improved. I think there are a bunch of technical details, like the way that the oil is stored.

If you drain the reserve too far down, you can have real problems with these giant salt, salt caverns, because the salt caverns are sort of stabilized by the presence of the oil in there. So if you draw it down too much, the actual structure could be unstable and all kinds of, you know, bad things could happen.

I'm not an expert on this. This is just what I've been told by these people. We don't really wanna end up there, and so there is some kind of nonlinear pressure on Trump to try to settle this thing out you know, really within the next, I would guess, 30 to 60 days, certainly prior to the midterms.

Now the Chinese situation is also quite interesting So I think, you know, the analysis I've seen says that prior to this crisis, the Chinese were buying more oil than they needed. And so, so what people thought was the demand, sort of daily demand of sort of, I don't know, what, 10 billion barrels of oil a day that China needed, something like this you know, that might have been overstated because they were use- they were buying more than they needed in order to fill their strategic reservoir reserves.

And, you know, some people estimate they have a billion barrels or something like this. Nobody, nobody actually knows. And then when this conflict started and the flow of oil from the Middle East to Asia really declined a lot people think the Chinese both opened up their strategic reserves at least slightly, were no longer buying beyond their needs as they were in the past and also maybe selectively turned off certain things like certain refining activities or petrochemical production activities.

And so there was, I think some people estimate a net 5 million barrels a day swing from just, just discretionary things that China did. And that is what, quote, "saved the world" from a big oil shock. That there, there could have already been an oil shock, but because the Chinese tuned their demand down by something like 5 million barrels a day then, then there wasn't a shock.

What's happening recently, I don't know. There's some rumor that the Chinese are coming back in the market and starting to buy again at a higher rate, which, which, you know, which could cause problems. You know, like, one of the theories of all this was, oh, it's good for the US to cause trouble in the Hormuz Strait because the Chinese really need this, the energy resources that are passing through there, and they'll be the ones really sweating if oil flow is, is cut off.

Now, oil flow for them wasn't really completely cut off because their ships were getting through, or at least some some fraction of their the oil destined for them was actually getting through. So that's another thing I've not seen, like, properly accounted for, like how much were they actually getting throughout this whole crisis.

But This sort of China hawk theory of the case is like, "Oh, forget about Iran. The real issue is just to like, you know threaten China and cut off their energy supplies, and we still have that sort of chokehold over them." But I think it, you know, it didn't actually materialize that way. They didn't they, they seemed to be, Xi Jinping and his team seemed to be quite calm through this whole thing. Now, maybe if the thing continues long enough, they will draw down their strategic reserves and, you know, they'll start to get desperate, for more energy. But I think before that happens, we'll have an oil and energy shock for the rest of the world.

Alf: I mean, you, you mentioned that kind of China hawk idea of, you know, cutting off the straits to, to disrupt the flow to China. Do you think another possible way that, that might backfire for them could be in making China pivot more towards energy imports from Russia, and therefore bringing Russia and China even closer together?

Steve Hsu: Yeah, I think that whole thing, which, you know, you and I have discussed before how the Ukraine war really pushed Russia into the arms of China, and so their energy cooperation is very tight now. But still I hear these rumors that the Chinese in negotiating with the Russians over these big pipelines and, and long-term contracts for the price of, like gas and oil through these pipelines the Chinese are driving a hard bargain.

And last I heard, they hadn't actually signed a long-term agreement because the Chinese were still pushing for lower prices and sort of felt like had... they had the Russians over a barrel. and I think as far as I know, that's still the case. Now, you know, if they really were stressed out overwhat's happening in the Middle East, that you, you might think they would settle those long-term contracts sooner rather than later, but I haven't heard that they, they have.

So I don't know that the Russian- Russia-China integration can get any tighter at this stage. It's pretty tight. But yes, I mean, there's a negative aspect to the Middle Eastern war in that it drives the two, those two parties still a little bit closer together.

Alf: There's something else I've, I mean, to pivot slightly, something else I've noticed in recent weeks regarding the, the, the, the, the war is a seeming kind of disengagement or at least apparent disengagement by Israel. So the foreign minister who's a hardliner Smotrich, he said that it was to Israel's advantage to remain out of the, the war from kind of this point onwards.

Have you noticed this trend, and if so, what, what do you think its consequences will be?

Steve Hsu: So my theory of the case here is that Israel's best case scenario is for the US to pulverize Iran, and they, because they're just the winners from that, right? And if Iran responds by attacking US bases which are in Gulf Arab states, that's actually good for Israel too in the long run because they would like to weaken those states too.

Ideally for Israel it is by it would emerge as by far the strongest military power, military and economic power in the Middle East. And so if Iran is beaten up mainly by US airstrikes, Iran retaliates against Gulf countries which are hosting US bases from which those attacks could be launched, so that then that ruins the economies and militaries of, you know Arab states, GCC countries.

This is all good for Israel, right? Now, of course, if Israel gets sucked into the conflict and the Iranians do things like knock out Israeli power plants, knock out Dimona, the nuclear react reactor in Israel knock out Israel's desalination plants, that's a disaster for Israel, and they're quite vulnerable.

They have a very finite set of very, very crucial bits of infrastructure which, if knocked out, would really cause a complete disaster there. So I think their ideal situation is they don't have to get involved in the fight. The US beats up Iran, Iran beats up the Arab states, and you know, finally, eventually it ends, and Israel's a net winner from all this.

So I don't think Smotrich is I don't think his comments I didn't find his comments actually surprising at all. I completely understand. His comments are completely consistent with my theory of the case.

Alf: And, and in terms of that, that outcome you mentioned of Iran getting getting pulverized, that, that trumped a whole civilization will die tonight tweet.

How, obviously it's hard to say, but how, how likely do you think that situation is to materialize?

Steve Hsu: I don't think we can do it,. You know, like even if we used all of our munitions, right? Say we've used half right now, or we've already used half, and we just use another half. Sure, it's really bad for the Iranian economy.

The Iranian people are gonna suffer. They're gonna have lots of bridges and power plants and things that need to be rebuilt. But it's, it's not easy to drive a country to submission if they have a unified political culture, a unified view of what's happening. If they think of themselves as a people and a nation state that should hang together, then, you know, they can countries can endure tremendous, tremendous amounts of you know, pain. Look at Japan at the end of World War II or Germany at the end of World War II. The strategic bombing, or Korea, I think actually there was some crazy statistic, like by the end of the war like every building in North Korea above a certain size had been destroyed.

And so countries can survive that. They don't like it. It's not good for them. It's terrible for their people, but they can survive it if they do have a unified political culture and sense of nationhood. And my, my bet is that Iran is in that category. So You know, I don't think we can just I don't think we can knock them out.

We can weaken them, which for Israel, from Israel's perspective is perfectly fine. but can we actually fully knock them out? I don't think so. Can we stop them from threatening oil tankers that are trying to transit the Strait of Hormuz? Like, you know, you have to look at the base sort of physics or military reality of that.

So you have ships traveling through, you know, a channel, which at its narrowest is only tens of kilometers wide. And, you know, those ships can be heavily damaged by little tiny drones. They can be damaged by artillery. They can be damaged by ballistic missiles, which are launched from a thousand kilometers away in the mountains of Iran.

So, so, like, for us to knock out enough of their stuff in all of those categories that I just described, let alone mini submarines and speedboats and, you know, all the different things that they could throw at oil tankers, that we have to neutralize all of that. Otherwise, insurers, people who own the ships people who actually are on the hook for the oil in those ships, nobody wants to risk it going through that kind of gauntlet.

This was known for a long time like, there were very famous war games done where we war gamed out an invasion of Iran or at least a, a crisis in the Hormuz Strait with Iran. You know, why did we war game this stuff out? Because the US has been under constant pressure by Israel and Israeli interests to start a war against Iran is, you know, for the last thirty plus years.

And so of course, the Pentagon, you know, is aware that yes, there are these political pressures for the US to do this stuff, so they need to study, "Well, maybe we're gonna have to do it," or, "Maybe our civilian leadership, you know, is gonna order us to do it." So they have to run war games and, and test out how these things would play out.

It's just obviously hard, right? 'Cause all the things that I mentioned, you can hide up in the mountains with your drones and you know, still attack the ships from a long way away. And how can we possibly neutralize, you know, hundreds of kilometers of shoreline and then going inward hundreds of kilometers which could be potential launch sites for missiles.

You know, that's, that's a huge area. It's like the you know, the area of like half the area of California or something . You know, it's like, it's like you can't really neutralize all the different ways that, the Iranians can use to attack ships. So I just don't see how we're gonna get to that point

Alf: Aside from a kind of aerial flattening which is, as you say, I mean, and can only go so far i- in a country with so much a kind of asabiyyah, like, like group solidarity and the, a- and the naval limitations, I suppose the on- the only other option is a ground invasion.

How likely or unlikely do you see that being from what-

Steve Hsu: Well, I mean, the amount of territory you'd have to occupy is huge, and you'd have to subdue every, you know, square kilometer of that area. Because if you have people hiding in the mountains, but they still have some drones or some rocket launchers and things like this.

You know, a tanker, a civilian tanker, if you think about it, is this gigantic tube filled with oil. It's not moving very fast. It's very, very visible from satellite. You know, everybody knows where they are at all times, and they're just this big fat target, and a little drone with you know, a few kilograms of high explosive can really cause a problem for that kind of it's not armored, right? So, so you know very, very tough even with a land invasion. Now, could we actually conduct a land invasion? Like, how big is our army? You know, like, how many people could we mobilize and position in a place where they could you know, make a marine landing on the coast of Iran, invade the country?

So that, that took a six to six months to a year to pre-position enough material to do that to Iraq, which is a smaller country, weaker country, doesn't have the missile capability that, didn't have the missile capability that the Iranians have. So I just think this would be a year-long logistical exercise to prepare for that kind of operation, and then that operation itself could still fail.

In other words, we, we could be occupying huge chunks of Iran, but nevertheless, ships are still not able to transit the Strait of Hormuz. So I just think the degree of difficulty is way higher than what happened in Iraq, and we're not really... Our military is not constituted to do that kind of operation anymore that we did in Iraq

Alf: Yeah, and I mean and some of the logistical aspects would be terrible as well. I mean, the IRGC from kind of available evidence seem pretty well-trained for a kind of long guerrilla war against US forces. Do you see any potential for a limited ground invasion on Kharg Island or Qeshm Island or anything?

Steve Hsu: Well, I think those things, most people who analyze that would just say this, this would be a kind of killing field because those islands are within artillery range, and there are huge logistical problems of supplying the troops even with water once they land there.

So just seems like a crazy thing to do. Like, I don't get why we, you know, why you'd wanna do that. Now, let me just make another point, which is that the reason we one of the reasons we have not been able to subdue Iran is because they've built a lot of their drone and rocket, both manufacturing and launch facilities under mountains.

So imagine you have a, it's a very mountainous country, Iran, right? So imagine you have some mountain and you just pick the end point of some road, and you say, "I'm kinda gonna drill into this mountain. I'm gonna tunnel into this mountain," and I have enough civil engineers that I can, I can do that. I mean, you know, the Chinese and lots of other, you know, countries have managed to build, like, huge underground complexes.

It's pretty low tech. It, it's, you know, somewhat expensive, but governments can certainly do this.

Alf: Korea did it as well. That's where they did a lot of their early tests.

Steve Hsu: Yeah, absolutely. So you just drill into you're going sideways into the mountain. You just basically tunnel directly into the mountain, but then the, the key facilities that you put under there are, you know, a kilometer of rock.

You know, they're under the mountain, a kilometer under that. And so no amount of US conventional weaponry, no amount of precision strike or you know, bunker busting bombs can really damage the core facilities that you build that way if they're really, quote, "under a mountain." Now, you can attack the entrances, but they're not stupid.

These Iranians are very smart. So, like, you know, they can easily build, oh, we'll have 10 entrances to this complex. As you figure out with satellites and stuff where the entrances are, and you bomb them, they're creating new ones, right or digging out the old ones. And so long range air power against underground facilities is not really a winner, okay?

Now, we could use nuclear weapons. That would be a big step, and maybe that could be a winner. But conventional weapons against stuff deep underground is not a winner. Even nuclear weapons against stuff deep underground, it's not clear whether it's a winner. and this is what's actually happened. So, so th- they probably routined most of their ability to make and launch both drones and missiles from these underground complexes in Iran.

Now, a priori, before the war started, you could have said, "Oh, this is some shitty, crappy theocracy. They're incompetent." You know, like, "I don't know where they got these missiles. but surely they could not have competently had the civil engineering to build, you know, these underground facilities that I just described."

Well, I think the answer is, yeah, no, they actually did build all this stuff, and we have not been able to take it out. And it's understandable why we haven't been able to take it out, and there's no obvious way that we are going to take it out in the near term. So what that means is, like, if you land on Kharg Island, as your troops are trying to get to Kharg Island from 1,000 kilometers away under some mountain, the Iranians can launch missiles at that ship or drones at that ship.

And, the troop carrier, all those things, the supply logistic ships, all of these things are gonna be subject to missile and drone attack, you know, unless we eliminate these underground facilities, which we can't. We don't know how to. Similarly, if Israel got in the from these underground facilities, they could launch ballistic missiles and drones at Israel from these sites.

And again, it's very tough to knock these things out. So you know, if you remember the beginning stages of this war, we there were all these bullshit statistics about, "Oh, we've locked knocked out 90% of their launchers," and all this stuff. I think that's mostly untrue. I think they still retain most of their capabilities in this area, and those capabilities could be brought to bear against ships that are trying to get troops to these islands, or they could be brought to bear against critical infrastructure in Israel. So I think it's just not a good situation.

Alf: And it, it interesting as well, there seem to be, at least to some degree, some parallels with Afghanistan because I mean, the US learnt the hard way during that 20-year war you know, the perils of mountain warfare. Except this timetaking out a mountainous IRGC base would be a much harder task even than taking out a mountainous Taliban base.

Steve Hsu: Yeah. In the case of Afghanistan, you had very, very determined but weak yeah in terms of capabilities and technology, mujahideen, people like these guys Taliban guys. You know, and we still couldn't win.

Alf: Yeah.

Steve Hsu: Right. But with Iran, you have they're not, like, at the absolute cutting edge of their missile technology or drone technology, but they're close enough that they, they do have some hypersonic missiles. They have some accurate ordinary ballistic missiles. They have pretty good drones, long-range drones, Shaheds. So just a totally different situation. And they did I think they did the civil engineering necessary, which was is not, definitely not rocket science.

Like I just described that they are actually not bad at rocket science. But civil engineering is way, way easier than , than rocket science, and if you do have mountains that you can tunnel into, then the there's a real problem for your enemy who's only gonna use air power against you.

Alf: And I mean, moving to the, the American home front you mentioned that save a miracle things are, things are gonna get pretty, pretty difficult for Trump ahead of the midterms. What are your predictions for the, the midterms insofar as you have any, and potentially. -

Steve Hsu: well, I don't know. I don't have any alpha over Polymarket, right? So I think Polymarket is saying it's kind of like 50/50 chance that the Republicans loses Senate. You know, probably very quite likely they're gonna lose the House.

But even a decent chance they could lose the Senate as well. So, so like I don't have any alpha over those predictions. I think those are reasonable predictions. But I think the one thing I would say is that I watched them tie up Trump into knots in during his first term, like every day you know, like, "Oh, let's impeach him.

Oh, let's impeach him over this. Let's impeach him over that. Let's start these special investigations of, you know, his corrupt finances," and you know. So there's plenty of soft targets for them to go after once they control one or both House one or both, components of the legislature . And so I just predict Trump is a lame duck now. I actually think his final two years he'll just be playing defensive ball the whole time

Alf:

And kind of changing gears to, to China now. Obviously, we've, we've touched on some recent developments in, in the US-China competition as they relate to the war. I mean, the other, the other big the, the other big news story recently has been the release of Kimi, the frontier AI model, which is competitive with the likes of Anthropic's Fable. What do you make of the release generally?

Steve Hsu: Yeah. It's super interesting because I think even people like me that were watching it carefully were not expecting them to release a model that good at this point in time. Like, I thought they would've remained, you know, a good six months, solid six months behind for a while.

You could argue now they're only three to six months behind. You know, even that gap, like if they're three months behind. So say they have a model which is just a little bit worse than GPT-5.6 and Fable. but it's, it's quite inexpensive. It's open source. This puts tremendous pressure on Anthropic and OpenAI as they approach their planned IPOs.

Because if you're gonna value those companies, each of them at a trillion dollar market cap, you're projecting their ability to continue to grow their revenue and margins in a very aggressive way in order to say that they should be worth a trillion dollars. 'Cause they're, they're losing money right now, right?

Most companies that are worth a trillion dollars make money. They lose money, right? So, so how are they gonna make money if there are Chinese competitors shipping a product which is competitive with their best products, but less expensive? It just, you know, it doesn't mean that OpenAI and Anthropic won't, can't be successful companies or that they can't still lead the space, but their margins could be compressed by quite a bit, or their market share could be compressed.

And so then, you know, you have to reevaluate all of your sort of net present value calculations of future cash flows of, like, what's really gonna happen to them. And then you might say, "Oh, this, this company's actually only worth $300 billion, not $1 trillion," right? And, you know, a company that didn't exist a few years ago being worth $300 billion, in ordinary times, that would still be one of the greatest value creation tech innovation stories of all time, right?

So Anthropic didn't exist five years ago value valuing at 300 billion That would still be an amazing story, one of the greatest stories in the history of technology. But the investors who buy it at the IPO price could lose most of their money. So , so it's a very, very complicated situation. Lots is evolving on the, in the terms of the frontier AI race.

Number one, it seems to be clear now that there are many pockets of human capital in China, so many different separate companies, whether it's Z.ai Moonshot, which makes Kimi, DeepSeek. Even Xiaomi launched a kind of quasi-frontier model. The latest Qwen model from Alibaba is supposed to be kind of at a fable level or just below a fable level.

So there, there are many, many entities in China that can make frontier, you know, maybe, maybe six months, you know, with a lag of six months frontier-level models. And that just seems to be very clear now, right? So it's not that hard in China if you have the resources to hire 2 or 300 people, you can ship an almost frontier model.

A model which is actually better maybe than the best Google model right now, better than the best Meta model , maybe better than the best SpaceX Grok model, right? So, so that's like shocking, right? That there's like, oh, there's five entities in China that can do this, right? And each of them only has a head count of 200 or 300 people versus like several thousand people for OpenAI and Anthropic.

So that already like you have to take, you have to think hard about what, what is, what is this like how is this possible and what, what does this actually mean? That's one set of things. Th- aside from that, there's the whole hardware semiconductor side of this, which is that during the last year or two, a lot of the innovation has been in Huawei and other Chinese companies shipping their own AI accelerator chips, and these open source models that are produced by the Chinese companies now can run on that hardware, so they don't even need NVIDIA hardware to run.

And what does that mean for NVIDIA's future cash flows? What does it mean for the competition between the Western technology stack and what is now a, you know, nearly indigenized Chinese semiconductor stack that can make chips all the way up to chips that run frontier-level AI models? So all of that is happening in the background, and it's a subject of intense interest if, if you're like kind of an expert on AI-related stuff or semiconductor stuff.

But I would guess people in Washington are not tracking this very well at all. And, you know, so there'll be some moment where like strategic planners in Washington realize like, "Oh, we don't have a huge AI advantage over China now."

Alf: Something interesting about the, the US-China AI competition is, is the difference between China's kind of open source approach or the, the open source approach of some of the Chinese companies. And what we're seeing on the US side is kind of increasing actually government limitations on AI we saw it with Fable, for example. What do you make of that contrast, and what do you see that that contrast's implications being?

Steve Hsu: Yeah. So in a way it's kind of funny because I find Anthropic's style of communication very off-putting. So, so they're, to me, they seem to always be talking up, their stuff. And so if you remember, there was all this hype about Mythos, the model that they never really released, but then they were, they, they showed it to the US government. And so there's this idea that these models are dangerous, they shouldn't be released, and the US government should look carefully at Mythos and Fable and GPT 5.6 and higher before they're released because they have, you know, very strong cyber hacking abilities, et cetera, et cetera.

That's all true, actually. I don't actually have a problem with that. They sort of, I think maybe inadvertently got the government into this position where the government's like, "Yeah, yeah, you're right. We should probably, like, regulate you guys and, and, you know, we we need to look at your latest model before you release it to make sure it's not really dangerous," right?

In the meantime, without their knowing it, because at the time when Mythos and Fable were first announced, nobody had the idea that the Chinese were that close. You have models like Kimi, which are pretty, K3, which are pretty close and also have very strong, maybe not as strong, but also have very strong hacking capabilities and stuff, and those are just going out open source.

In fact, the, the weights for K3 were just released today, July 27th. And so, like, if you want to use K3 for hacking, you can download the weights and run it yourself and, and use it to hack. And, so we're in this strange situation now where the US labs, there's this extra layer of risk now, which I, you know, I heard investors are really concerned about this, that the, the best US lab models now could get caught up for some period of time by the government having to review them before they came out, whereas the Chinese could just release their open weights models whenever they want.

So we're in that weird situation. Also because of the government review process, you know Anthropic and GPT are sort of crippling their state-of-the-art models, their frontier models in certain capabilities. Like if somebody is asking, you know, "How do I make this modification to this virus in a wet lab?"

You know, the, the model won't tell them. If I ask the model how to build, like, a better AI model, the model won't tell them. If I ask them how to what the vulnerability, how to write a you know, a, for a zero day exploit using this vulnerability for this particular operating system or something, the model won't tell them, right?

So there are all these, like, guardrails that are now embedded in the Western frontier models, but those guardrails aren't necessarily there in the Chinese model. So it, it's a, it's a very strange situation.

Alf: And it, it's kind of an ironic kind of flipping of things 'cause with the internet, obviously China is known for the great firewall and accused of censorship there, and then the West being a, a freer environment.

It sounds as though the, the Chinese models are in a way freer to speak their minds yeah than some of the Western ones.

Steve Hsu: There was a very funny thing that happened where they were testing OpenAI itself was testing 5.6 and then a future model that has yet to be released, which is even stronger.

And they were testing those models in some kind of sandbox environment, but the models hacked the their way out of the they, they, they used some vulnerability in I think a proxy server to get internet access outside the sandbox, and basically then attacked another company called Hugging Face and hacked into Hugging Face's web servers.

And so all this stuff went on for a while and it was actually discovered by Hugging Face, and I guess OpenAI maybe wasn't even aware that their own models had been doing this that they had broken out of the sandbox and were doing this. And Hugging Face, when it was, when it detected the intrusion, wanted to analyze some very large number of logs, like thousands and thousands of you know modifications and little bits of code that the hack- hacking program had installed on their network.

And, you know, they had many, many things to look at, and so they were using, trying to use an AI, an LLM, to like track what had been done during the intrusion. And they couldn't get any US frontier models to help them with the analysis because the frontier models had all these guardrails, and the guardrail forced the model to say, "Hey, it looks like this is hacking activity.

I can't help you with this." Right? And so like, in order for Hugging Face, in order for Hugging Face to figure out what, how the intrusion had gone on their own network and systems, they had to use GLM 5.2, which was a Chinese model. So they had a, they had a locally installed version of 5.2 that, and they could run that and it would analyze the logs and all the software malicious code that had been left on their so they had to they had to basically use the Chinese technology in order to figure out what had been done to their system. So very, very ironic, situation that we're in right now.

Alf: And you, you mentioned earlier This kind of sudden realization, within the US that may or may not come in the future that, that, that there isn't actually a huge gap in terms of US and Chinese AI capabilities. Do you see at all, and obviously this links to a much bigger issue, but do you see this as in any way potentially contributing towards a popping of what's been termed the AI bubble?

Steve Hsu: I think so. So there, there's different parts of the AI bubble. So here's a scenario which I think is very plausible. OpenAI, Anthropic don't have a moat. They, they are able to remain ahead of their Chinese competitors, but not by a huge margin, and therefore their future cash flows aren't as as high as what some investors would estimate.

And so the trillion-dollar market caps that, you know, they've already in their private rounds preparing for the IPO, they've already been valued at trillion dollars. So there's a bunch of investors that pumped in roughly $100 billion into each of those companies at about a trillion dollar valuation. They would probably hope to IPO at a higher valuation than that.

But it could be that because of the competitive situation, they're really not worth that much. And so those investors that invested in those frontier model companies will be very disappointed. The shares that they bought at the trillion dollar valuation will not be, turn out to be worth that much.

However, companies like Nvidia or the memory companies we haven't discussed this, but there's a huge, I don't know whether to use bubble or boom, but there's a huge boom in the prices of companies that make semiconductors like TSMC, the memory makers like Samsung and Hynix in Korea, and the memory makers in China.

I think the most valuable company in China right now is the recently IPO'd CXMT, which makes memory. So that semiconductor boom or bubble, depending on how you look at it, might be sustained. So it might be that yes, you've got these small teams of two or 300 Chinese guys who can build models which are almost as good as OpenAI and Anthropic, and so that increases demand for artificial intelligence.

So there are lots of applications for it, software development, people analyzing stocks, whatever, whatever they're doing with it. And they can do it cheaply because there are all these Chinese models available, open weights. But the demand for compute to do all of this which requires memory, which requires GPUs, which are fabbed at TSMC or, you know, SMIC in China, you know, that boom might be real.

Whereas OpenAI and Anthropic turn out not to be as valuable and, and a host of other AI applications companies turn out not to be as valuable as investors thought. So, so, you know the boom or the bubble has many components, some of which might turn out to be properly valued at this moment in time, and some of which might turn out to be of inflated value. And of course it, it doesn't have to turn out the way that I just described it, but I think that's a plausible possibility.

Alf:

I had Philip Pilkington on the, on, on the show recently and who I know is also, you've also collaborated with before. I urge anyone listening to, to check out his work and his podcast, Multipolarity if they haven't already.

In that discussion that I had with him he argued that given the Iran war and the Ukraine war are so kind of embedded now and interconnected that the world was effectively in a third World War. I don't know. What do you make of that assessment?

Steve Hsu: Yeah. I think that it's plausible. If we jump in a time machine and we go five years forward and then we look back I would not be surprised upon stepping out of the time machine if someone said to me, "Yeah, all this stuff happening in 2026, that was what expanded into a broader war," which, you know, could legitimately be called the Third World War.

So you could have intensifying conflicts between Russia and the Europeans, via Ukraine. You could have a broader conflict in the Middle East where the US and Iran you know, there's a spiral toward more and more intense conflict with the Gulf countries getting sucked in, being damaged, maybe Israel being damaged.

And, you know, maybe that's enough to call it a world war. And of course, at, at the level of, like, hybrid war, there's definitely already a hybrid war between the United States and Russia and China because US is trying to contain so sanction China. I mean, sanction Russia, wanted to bring Putin's regime down through sanctions.

That didn't work actually, but that's actually what we wanted to do. And the Ukraine War was a good pretense for imposing those sanctions. Wanting to contain China's technological development, cutting them off from all advanced chips, hoping that we would win the AI race and contain them that way, hoping that we could maybe stop the oil flow to them from the Middle East.

You know, so, all of that you could lump together and call, like, World War III. I think that would not be unreasonable. Now, I think the one silver lining to what's going on right now is I think the US has bitten off a lot to chew in the Middle East. So if, things do spiral out of control in the Middle East, like you could, we could do this just by counting stockpiles of major weapon systems, it will eat all of what the US has.

We've eaten half of it roughly , so we, we could eat, we could eat more if, the conflict widens in the Middle East. But that means the US really doesn't have gas in the tank or weapons in the stockpiles to do anything in the Western Pacific. If you ask me, "Oh, will the World War extend to the Western Pacific and engulf South Korea and Taiwan and Japan and China and the US?"

Maybe the odds of that went down because the US military certainly, like, you know, aside from political generals who just tell Trump what he wants to hear, the more serious colonels and generals who actually analyze things are probably scared out of their wits because we're not in a good position to do anything in the Western Pacific given our current stockpile levels, and definitely not given the stockpile levels that we'll have if we get sucked into a wider Middle Eastern conflict.

I agree with Phil that, like, I think historians will look back on this and say there was a worldwide conflict over, you know, who would replace the US as the hyperpower, the hyperpower hegemon, right? So the US is no longer that, and the transition into this more multipolar world is ongoing, and it involves some hot conflicts in Ukraine and Iran but also lots of hybrid warfare economic, technological competition, which is sort of borderline warfare, but more like a Cold War.

And so I think future historians who wanna conceptualize everything that's happening now may look back and call it the Third World War. Hopefully it doesn't become a real maximalist Third World War which would definitely involve nuclear weapons or at least would involve really all-out warfare in the Western Pacific which would really destroy, I think, you know, the US Pacific military presence, lots of bases in Japan and Korea, maybe inflict a lot of damage on Chinese infrastructure.

You know, hopefully we don't get there. But in a general sense, yes, historians might look back and say, "Yeah, this was a kind of Third World War that's already ongoing by, you know, the, by 2026 or 2027." If you could just add one more thing which we didn't touch on.

Alf: For sure, yeah ...

Steve Hsu: Which is the nuclear situation in Iran. So, you know, what, part of the time when Trump is talking he's saying, "Oh, the main thing here is that we just stopped Iran from getting a nuke," right? What, when people, when people criticize Trump and say, "Why did you do this crazy thing? Like, why did you

Alf: Yeah

Steve Hsu: attack Iran?" You know, this is, it, I know the Israeli the Israelis really wanted this to happen.

Netanyahu really wanted this to happen, but it's not in US interests, right? So a lot of people would sort of say that to Trump, and one of his rhetorical rejoinders is, "Well, do you want Iran to have a nuclear weapon?" Right? So that's

Alf: I think, Rubi- Rubio has a similar kind, he says a similar thing.

Steve Hsu: Yeah. So, so that's a, that's a way, you know, that's in our word salad American culture where, like, shitty, dumb arguments are, like, okay, 'cause people are not used to rigorous thinking and they're not familiar with facts and stuff like, you know, or, or, like, base reality and physics and stuff. So, so in our current, in the way our system works, like, that's okay.

Like, you, you, okay, if that's your rejoinder, you're still, like, a, quote, "respectable politician and leader," right? So that's, that's where they are. And, but then when even within our crazy word salad system, you're allowed to at least ask the next question which is, like, "Okay. Well, did we actually prevent the Iranians from getting nuclear weapons," right?

Like, like, okay, that's your justification. Fine, let's, let's just pursue that one step further. did we stop the Iranians from getting nuclear weapons? And I would guess the answer is no, so I think they still have their nuclear dust. I think they still have many, many centrifuges probably buried deep underground.

I think that they're pretty traumatized. Like, if you look at the the funeral for Khamenei, which, you know, I guess definitely tens of millions of Iranians, not to mention Shias in Iraq and other places turned out to mourn what, you know, assassination of a leader during a negotiation. I mean, that's pretty bad, right?

So people get pissed off over this stuff, right? So all the moderates in Iran are either quiet now or they've become hardliners, right? So they either explicitly now are hardliners, realists who no longer believe in a liberal Western international order or that the US is trustworthy, can be negotiated with.

They either like explicitly now have switched their worldview openly or they're just quiet, okay? But either way, the hardliners are in power in Iran and so I would not be surprised if they go for nukes, right? So, so this whole conflict, like you say, what's the off-ramp for Trump? How is this gonna end, right?

Because Trump can't really accept the terms in the MOU politically. It's pretty tough for him to accept those terms. Have seemed to have no intention of backing off significantly from what was in the MOU, right? So how does this end? Well, one way it could end is the Iranians just say like, "Yeah, we just did a test and we have 10 nuclear devices now." And then we're in a completely different regime.

Alf: And so that's something that would be similar to like what, what North Korea did when they announced that they had weapons.

Steve Hsu: Yeah, absolutely. So anybody who studies the history of that is like US is was constantly rattling sabers you know, trying to threaten North Korea until they developed their bomb. Now nobody talks about North Korea. Trump goes there and shakes hands with, you know, the, the leader and, you know. so, so what could happen is like Iran could just say like one day like, "Hey, guys. Yeah, we developed a bomb. That seismic event that you detected, that was our 10 kilogram test, and we have 10 more bombs like that.

And you know already we have the delivery systems, the missiles to deliver them. So let's just, can we just, let's just, call things off." Mm-hmm. "Like, we don't wanna, we don't wanna have a nuclear exchange with you. You don't wanna have a nuclear exchange with us. Let's just call, call a ceasefire in, in the greater Middle East." And, and maybe that'll, that's how this whole thing is gonna end. And if that happens sometime in the next year, I wouldn't be that shocked.

Alf: And on, on that note thanks so much for coming on, Steve. It's been a great chat as ever. Covering a very broad array of topics from the ongoing war to technology and, various things in between and how they all intersect. So yeah, thanks so much for coming on the show, and I look forward to the next one.

Steve Hsu: Yeah, I look forward to it as well. These are, you know, in historical and geopolitical times, these are among the most interesting times that have transpired in my lifetime. So, look forward to further discussions.

Alf: Most definitely true. Thank you.