中国已戳破美国AI泡沫！英国教授犀利点评：美国拿什么和中国竞争？

Uh the US will invest well over a trillion dollars in AI, artificial intelligence in the next couple of years alone. It's the transformative

alone. It's the transformative technology of our age. But has the US already fallen behind China in the global AI race? And if it has, did that

lead to yet another great financial crash? This is the Andrew Neil report.

crash? This is the Andrew Neil report.

Well, I expect most of the companies, the American ones, to go bust.

Oh. And we will see I think maybe you know one in 10 of the companies surviving.

The Chinese are the winners.

Yeah. And that's they're not necessarily behind the Americans either. I I

wouldn't like to be an investor in AI right now, let alone somebody who started one of these.

[music] forch.

Fore! Foreign! Foreign!

for The scale of US investment in AI is eyewatering. It propels economic growth

eyewatering. It propels economic growth in America. It leaves Europe in the

in America. It leaves Europe in the dust, but not China. Indeed, US AI stocks are starting to shake a little as the Chinese threat starts to

materialize. This month saw the release

materialize. This month saw the release of KK3 from a Chinese company called Moonshot. It's a powerful AI model

Moonshot. It's a powerful AI model that's cheaper than many of its US rivals. But China's not all that's

rivals. But China's not all that's worrying AI in America. So this week, the Andrew Neil report is going to take a deep dive into the challenges facing

America's AI revolution, China, debt, and energy. Then speak to Professor

and energy. Then speak to Professor Steve Keane of University College London and a global authority on such matters to get the measure of the problems.

Let's start with China. Perhaps the

biggest threat to American AI companies is that they're undercut by cheaper, what you could call commoditized AI from the likes of Deep Sea in China. That

would really threaten the implicit pricing models behind today's equity valuations. They wouldn't be able to

valuations. They wouldn't be able to charge the prices they hope for their services which are propelling the size of the valuations. Now if it's true that for

valuations. Now if it's true that for example deepseek v4 uh can achieve 80 to 90% of the performance of uh anthropics

claust you can see American AI has a problem and let's just look at the current valuations let's chart start with

China's two biggest moonshot AI that's valued around $ 31 billion deepseat that's valued around 71 billion a lot of

money but look at the two American American giants. Open AI valued at 852

American giants. Open AI valued at 852 billion. Anthropic valued at 965 billion

billion. Anthropic valued at 965 billion almost a trillion. So this means that US AI firms are currently valued 10 to 25

times more than their Chinese competitors despite similar capabilities. I think that's what's

capabilities. I think that's what's known as systemic risk.

So much for China. Then there's the financing problem. Now, a recent study

financing problem. Now, a recent study by Nikai, the Japanese stock exchange, revealed that five US tech giants have

hidden offbalance sheet debt of $1.65 trillion, much of it accumulated via what's called private credit. Now, the opaque

private credit. Now, the opaque financing of AI investment by private credit sources. They're not banks,

credit sources. They're not banks, they're not bonds, it's not equity, but various sources of largely unregulated credit. That's a potential threat not

credit. That's a potential threat not just to the expansion of AI, but to the wider financial system. Now, some fear that the next financial crash will be

caused by a meltdown in private credit, bringing AI investment down with it.

Now, nobody even knows just how much private credit there is in the AI investment boom, just that there's a lot. Private credit as a source of

lot. Private credit as a source of investment finance has grown since regulations following the great financial crash of 2008. They put

greater restrictions on bank lending and other forms of finance. It's essentially

private credit a way around these regulations. Even the banks are now

regulations. Even the banks are now lending via private credit vehicles.

Private equity and private credit are intimately intertwined. Now, by early

intimately intertwined. Now, by early 2025, it was reckoned that private credit lending to the technology sector in America stood about 450 billion. That

was up 100 billion on the year. It's

probably doubled since then as the AI spending boom has gathered speed. Now

one of the risk is that private cit is so deeply embedded in loans to software companies perhaps as much as 500 billion year there and yet these are the very

companies thought most at risk from AI which can develop its own software. So

the more private credit invests in AI the more it might be undermining its previous investments. The global private

previous investments. The global private credit market is now estimated to be over three trillion. We don't actually know how much. That's because of the

lack of transparency, but it's a lot.

And some estimate that outstanding private credit to AI firms will be at least 600 billion by 2030. Now, of

course, private credit is only one source of finance fueling the AI boom.

Corporate bonds, banks, equity, other sources have put in even more. But the

fear is that private credit is the most vulnerable. if AI doesn't start

vulnerable. if AI doesn't start generating the revenues that will be needed to service this debt and pay it back. As Nicki discovered, private

back. As Nicki discovered, private credit is most often involved in offbalance sheet joint ventures, special purpose vehicles and the like. somewhat

shadowy financing mechanisms often first to combust when things go wrong, bringing down not just AI but companies throughout the whole financial system.

And what could go wrong other than China? Well, there might not be enough

China? Well, there might not be enough energy to power all these AI data centers springing up even in energy rich

and energy cheap North America.

hyperscalers, chipmakers, large AI complexes, they now have a valuation at a jaw jaw-dropping 20 trillion dollars.

And that is predicated on a quick and massive buildout of AI infrastructure to generate the revenues, to justify the investment, to pay back the debt,

service the debt, above all in data center hubs. that the electrical

center hubs. that the electrical infrastructure needed to sustain these hubs, well, it just isn't there and it won't be anytime soon. Start with the

fact that the US power grid hasn't been upgraded since the 1970s and it's an immense job to begin now. And the lack of energy isn't actually the primary

problem. It's the global bottlenecks of

problem. It's the global bottlenecks of transformers, substations, switch gear, transmission lines needed for that upgrade, plus the acute shortage of

skilled labor in the United States to install and run these things. Now, the

data center appetite for power and water is insatiable. People don't realize just

is insatiable. People don't realize just how much. Let's just take the hub, the

how much. Let's just take the hub, the data hub at Hayes County in Texas. It's

a good example. Texas, of course, you don't get more energy, Rachel, than that. But that hub can use up to 10

that. But that hub can use up to 10 million gallons of water a day for cooling and power generation. And in the process, it's depleting what's known as

the Edwards Aquifer, which is essential to the Austin San Antonio corridor. You

can't do both. You can't keep the water running to the people and give AI all the water it needs. Now, this

electricity crunch may well be the biggest danger to AI of all. The data

companies, let's stick with Texas because, as I say, you don't get more energy uh uh so energy intensive than that. Data companies have asked the

that. Data companies have asked the Texas power system operator, it's called Aircott, for grid connections to 200 projects in the years ahead. Now, that's

a total request amounting to 446 gawatt by 2023. Let me just put that in

by 2023. Let me just put that in context. It's five times the current

context. It's five times the current peak power for the entire state of Texas. It's 10 It's 13 times the UK's

Texas. It's 10 It's 13 times the UK's average use of power for the whole country. Now, expansion on this scale

country. Now, expansion on this scale and on this timeline is simply impossible. So massive investment is

impossible. So massive investment is taking place in AI infrastructure that won't come on stream for years because of power shortages. Let's speak now to Professor Steve Keen. He keeps an eye on

all these matters. He's at University College uh London. Professor, thank you for joining me. If I can, let me start with China. What is your overview of

with China. What is your overview of where we are with this Chinese American competition in AI?

Well, it's interesting in that what China has done is prepare the ground for something like AI before I actually came along. So, China has built rebuilt an

along. So, China has built rebuilt an enormous energy transmission system throughout the country. Uh we're using 800 volts rather than 400 volts which has a in technologically means you have

a higher amount of capacity to push energy from one part of the country to another. Uh and what it means is the

another. Uh and what it means is the people in China basically take energy as something which is going to be available almost whatever you do with it. America

is still stuck with the old 400 watt system availing.

So in some ways China set the ground for the demands of AI before AI came along.

And then a major part of the the entrepreneurial side of Chinese industry which is massively underestimated by America has focused on reducing the costs of doing the processing involved

in running these AIs. So uh the basic story is there's enormous amounts of multiplication matrix mathematics going on to enable these machines or data

systems to communicate with you as if you're talking to a human. and and the calculations themselves are immensely expensive in terms of energy. You've got

some of the models involving let's say they say two trillion parameters. That's

two trillion numbers that need to be multiplied against each other multiple times to converge on some advice to you on whatever you're looking at whether that's how to send up your father-in-law

in a birthday card or how to solve a an unsolved conundrum in mathematics. So

the energy needs are enormous and what the Chinese have done as well is say that this is something which should be open and available to everybody rather than restricted and only available to people who purchase a particular

proprietary system. Uh so they're going

proprietary system. Uh so they're going open source and that's one of the major challenges of Kimmy. You can actually see what the parameters are. You can

potentially load it on your own hardware. You don't need a data center

hardware. You don't need a data center necessarily whereas you do for the American systems. So I think at a global level it's the Chinese who were set up to win this contest. And of course that's not at all what the American

entrepreneurs who dived into AI thought was going to happen.

Indeed. And it's not what's behind the valuations of AI companies in America either. Just before I move on on on

either. Just before I move on on on other Chinese issues, just to be clear, it's interesting what you said there that the the kind of energy constraints I was talking about in my monologue that

AI in America is about to hit. By and

large, you're saying they don't exist in China.

No, that's right. They the higher vault.

This is again one of the advantages of China being run by engineers rather than economists and lawyers. And I think that's the reason why everybody in the polar bureau, not everybody, but the vast majority of engineering backgrounds, they're aware of the

advantages of different uh transmission systems in a way that somebody doing an who's done an economics degree, you know, PPE course has no idea. And they

say, well, we need energy for an advanced economy. How do we distribute

advanced economy. How do we distribute the energy more easily? We need to go to higher voltage. And that's what they've

higher voltage. And that's what they've been doing. So the energy has been built

been doing. So the energy has been built there, I suppose, almost a matter of course. And then when you talk to

course. And then when you talk to Chinese entrepreneurs, they are fairly they don't worry about the availability of energy. Water is a different story,

of energy. Water is a different story, but they don't worry about the availability of energy. So that

combination having available energy to start with and then producing systems which don't need the same amount of energy in terms of data centers gives the Chinese an enormous advantage and

the Americans I think that you're finding out what it happens when you have a centrally planned versus a non-entrally planned system. And in this case, the central planning means the energy is already available or it's not there for the disagregated American

system. [snorts]

system. [snorts] In general, is it true to say that although the American AI

systems may be more sophisticated that the the Chinese AI models are much

cheaper and have 85 90% of the capacity of the most sophisticated American models. So if you

if you combine cost with capacity and capability, the Chinese are the winners.

Yeah. And that's they're not necessarily behind the Americans either. There's a

lot of extremely clever engineers. Uh

you often did their PhDs in America and went back to China. Uh they're

outstanding intellects and a lot of it is how do you manage to do these matrix mathematics uh more efficiently? How can

you cut down the the load? where where

do you get points where there's overkill versus where you can stop and get most of the answers you need and therefore if you do that you get more time to actually train the results and you get a higher quality results so I think the

Chinese have been out innovated the Americans in some ways the Americans have done what they seem to do as a species that is they use brute force to overwhelm the rest of the world the Chinese have used intelligence rather

than brute force this time round so what is the likely American response to B I know that the Trump administration has tried to limit the

export of some of the most sophisticated chips uh to China uh which chips which China at the most sophisticated level does not produce itself. It does need

the these imports. But am I right in thinking that nothing America has really done to try and dampen the Chinese AI uh

projects has had much of an impact?

No. In fact, in fact, in some ways, it spurred the Chinese to innovate more.

So, for example, the bans on chips meant some and and and not being able to get, you know, machinery from AMSL to build uh chips in in their own country, whereas they sell those to Taiwan and

then they onsold to America. This has

forced the Chinese to innovate different ways to produce uh very you know low low micron chips to get enormous amounts of processing power and small small

bundles. So the the bans have actually

bundles. So the the bans have actually encouraged innovation by the Chinese.

And I think in some ways I think this has got a historical parallel because every Chinese chi school student learns that the opium wars learns that they were effectively raped by the west that

they can't rely upon. They can't trust the west. And so anything like putting

the west. And so anything like putting up a uh a barrier saying you can't have this technology actually inspires the Chinese to say we'll replace it ourselves and become self-sufficient because our mistake in the previous

times was to lose our self-sufficiency against the west.

Reuters has been reporting this week that talks are to begin between China and America on AI the American side to be left by the Treasury Secretary Scott

Bassant. Um,

Bassant. Um, is that likely to lead to anything or is continued competition between the two superpowers of AI just likely to continue?

I I doubt it because what the Chinese and the EC even made a speech on this front just recently. They want it to be open source. They want the stuff to be

open source. They want the stuff to be publicly available. Treat effectively

publicly available. Treat effectively treat AI as a public utility, not as a source of private profit. Now, when you do private profit, you make everything proprietary. You try to restrict what's

proprietary. You try to restrict what's available. And that's what the Americans

available. And that's what the Americans have done. And in a typical, you

have done. And in a typical, you mentioned the railways boom beforehand.

A typical boom and bust cycle in capitalism which shumped explained extremely well. All the firms go in

extremely well. All the firms go in thinking are going to be the dominant party. So you get massive

party. So you get massive overinvestment. The overinvestment

overinvestment. The overinvestment extends the technology through a whole of society. When the technology comes

of society. When the technology comes out, it starts undercutting and destroying other uh previous sectors. So

of course railways destroyed carriages in that sense. And then the the technology permeates society when the bust occurs. Now I think we're close to

bust occurs. Now I think we're close to that point. It will permeate through

that point. It will permeate through society after a bust occurs. But the

scale of that financial bubble, as you were saying, in America is absolutely enormous. So these Americans are in a

enormous. So these Americans are in a desperate situation. They've their their

desperate situation. They've their their costs are sometimes 10 times their revenue. They're running up massive

revenue. They're running up massive losses. Uh they think we've got to make

losses. Uh they think we've got to make sure that we get uh the revenue that's going to come out of this when the crash occurs. I think they're seeing the

occurs. I think they're seeing the inevitability of a drop in the price of tokens uh coming up at some point. So

they're trying to make it proprietary and in that sense what you've got is the ch the Chinese are voting for making this open and available to everybody.

The Americans are trying to restrict it for their own benefit. The Chinese can just keep on going. They don't have to cooperate. So I think we ultimately the

cooperate. So I think we ultimately the only only resolution for this is the one the Chinese want where AI becomes a public utility provided to everybody and it's open access rather than restricted

and privatized. But even taking that

and privatized. But even taking that into account, how where do you see things going in the next couple of years

uh with AI in into the next decade, let's say, given the China American dynamic that you've just outlined for us?

Yeah. Well, I expect most of the AI companies, the American ones, to go bust. Uh because the difference between

bust. Uh because the difference between their revenue and their cost is huge.

Five to 10 times cost being five to 10 times revenue. remember remember the the

times revenue. remember remember the the internet bubble you and I obviously live through that one and you remember pets.com uh uh and I think it was valued at more than the the the entire pet

industry in America uh there were there were Amazon was valued at more than the transportation you know the the airline systems valuations get crazy and then

they crash uh after the the putting in the technology transforms society but doesn't bleed the profit that the initial people who rush in expect so

we're getting that in on steroids this time round and we will see I think maybe you know one in 10 of the companies surviving but the wild card as it was discussing now is that there's another

country that's in this game in a way that America hasn't experienced before so when the telecommunications bubble or the internet bubble occurred it was exclusively American technology doing it this time round they've done it they've

overinvested they've over borrowed they're lever they're going to crash and on top of that they might lose all the competition to China or any forchech.