The AI Super Bubble: Will 2029 Be Our 1929?

Every generation convinces itself its defining technology breaks the old rules of leverage and valuation. In the 1920s it was the motor car, the radio, the household appliance. Today the ai bubble comparison being drawn is with 1929 itself, and it's coming from Professor Steve Keen, the UCL economist who flagged the 2008 crash before most of Wall Street was willing to say it out loud. His case rests on three things that rarely get discussed together: the valuation gap between American and Chinese AI firms, the debt quietly financing the American build-out, and the physical electricity grid underneath all of it.

Is the AI Boom Really Heading for a 1929-Style Crash?

Steve Keen thinks so, and he isn't hedging much. Andrew Neil's interview with Professor Steve Keen, "China Has Won the AI War", lays out a prediction that most American AI companies go bust, with perhaps one in ten surviving, and China emerging as the structural winner, not because its models are better, but because its economics hold together. His timeline is roughly two years. Precise crash timing is notoriously hard to call, and economists have cried wolf before, but the underlying mechanisms he describes are concrete enough to examine on their own terms regardless of whether the reckoning lands in 2027, 2029, or later.

Keen isn't alone in reaching for this comparison. The New Yorker has run a financial-page piece titled "The A.I. Boom and the Spectre of 1929." The Economist has asked outright whether China could pop America's AI bubble. Bloomberg has reported Chinese hedge fund managers describing the rally as a "super bubble" whose collapse point "may not be far away."

The Trillion-Dollar Bet: Why American and Chinese AI Valuations Have Diverged So Sharply

The Numbers Behind the Gap

American investment in AI is expected to exceed a trillion dollars over the next couple of years alone, and Keen puts the combined valuation of hyperscalers, chipmakers and the wider AI-industrial complex at roughly $20 trillion. OpenAI is valued at around $852 billion, Anthropic at around $965 billion, figures that assume both companies will eventually charge prices matching those numbers.

The Chinese Comparison

That assumption is under direct pressure. China's Moonshot AI, valued at around $31 billion, released Kimi K3, a model that rattled Wall Street on release. DeepSeek, valued at roughly $71 billion, has models Keen says achieve 80 to 90 per cent of the performance of the best American systems at perhaps a tenth of the cost. Put plainly: American firms currently trade at 10 to 25 times the valuation of Chinese competitors offering broadly comparable capability. That's not a rounding error in anyone's model. It's the shape of a bubble waiting for something sharp.

A bubble, in plain terms, is simply a price detached from what underlying earnings can plausibly justify, sustained by the belief someone else will pay more tomorrow. Bubbles don't require dishonesty or stupidity. They require optimism, cheap credit, and a genuinely transformative technology story, exactly what railways, radio, the internet, and now AI have each supplied in turn. What ends a bubble usually isn't the story being proven false, it was largely true for railways and the internet too, but the moment financing costs catch up with cash flow that hasn't materialised yet.

The Advantage Nobody Priced In: China's Electricity Grid

The Transmission Gap in Numbers

China runs its long-distance transmission system at roughly 800 kilovolts against an American backbone still largely built around 345 to 500 kilovolt lines, some of it dating to the 1970s. China now operates 45 ultra-high-voltage transmission projects spanning more than 40,000 kilometres, including 23 lines running at ±800 kilovolts DC and the world's only ±1,100 kilovolt line, according to National Energy Administration data. State Grid alone has built 43 of these UHV corridors and plans fifteen more by 2030, giving China cross-provincial transmission capacity exceeding 380 million kilowatts. The United States added a total of 888 miles of new 345kV-and-above transmission in the whole of 2024, and its first genuinely high-capacity 765kV corridors are only now being planned in Texas.

Why Voltage Is the Whole Game

The physics is simple: doubling the voltage on a line lets you push roughly four times the power down it, with proportionally lower losses over distance. China built this ultra-high-voltage transmission China capacity for hydropower and renewables long before AI arrived, which is exactly why, as Keen puts it, Chinese entrepreneurs "take energy as something which is going to be available almost whatever you do with it." American AI developers, by contrast, have to think constantly about power conservation because the grid is already close to its limits. China's generation capacity runs at nearly double peak demand, a deliberate strategic surplus, against an American reserve margin of roughly 15 per cent above peak. China's annual increase in AI data center electricity demand alone now exceeds Germany's entire annual electricity consumption.

The Debt Nobody Can Quite Measure: Private Credit's Hidden AI Exposure

If energy is the physical constraint, financing is the invisible one. Reporting drawing on Nikkei data found five major American technology companies carrying $1.65 trillion in off-balance-sheet debt, much of it channelled through private credit AI data centers debt, arrangements that are neither conventional bank loans, publicly traded bonds, nor equity, sitting largely outside regulatory view.

Private credit lending to the American technology sector stood at roughly $450 billion in early 2025, up $100 billion on the year, and Keen believes it has likely doubled again since as AI spending accelerated. Some estimates put outstanding private credit exposure to AI firms reaching at least $600 billion by 2030, against a broader global private credit market already estimated above $3 trillion. This matters because private credit grew explicitly as a way around the banking regulation introduced after 2008, and because, as Keen bluntly puts it, "nobody even knows just how much private credit there is in the AI investment boom, just that there's a lot." Factor in associated bank lending, and true exposure could run three to five times the headline private-credit figures.

Why 1929 Is the Better Analogy Than the Dotcom Crash

The dotcom bust is the more commonly cited precedent for tech overreach, but the mechanics Keen describes look more like 1929. By September 1929, margin debt, money borrowed to buy stock with only 10 per cent down, had reached about $8.5 billion, roughly 10 to 12 per cent of America's entire stock market capitalisation and close to the size of the federal budget. That leverage worked beautifully while prices rose. When real-economy cash flow began softening as early as 1927, two full years before the crash, margin calls forced sales, sales depressed prices further, and the whole structure cascaded within weeks in October 1929.

The AI financing structure carries a near-identical shape in modern clothing. Instead of margin loans on stock, it's private credit funding data-centre construction. Instead of a broker demanding more collateral, it's a lender demanding debt service from a company whose revenue hasn't caught up with its capital outlay. Keen's own framing of the trigger mechanism is worth stating directly: when you're building data centres and laying out the systems, you don't have the cash flow yet, and the real question is whether you can get cash flow to a profitable level before losses accumulate too far in the meantime. That's precisely the bet 1920s utility and investment-trust shareholders made on electrification, another capital-intensive, transformative technology that produced spectacular bankruptcies even as the underlying technology went on to remake the economy.

The Triple Whammy: Chinese Competition, Hidden Debt and Physical Limits

Competition Undercutting Pricing Power

Chinese competition is compressing the pricing power American valuations depend on. If commoditised, cheaper Chinese models capture enough market share, the premium pricing baked into American valuations becomes unsustainable, and open-source Chinese models accelerate that commoditisation by letting anyone run a capable model without renting a hyperscaler's data centre.

Debt Cascading Through an Opaque System

Private credit exposure means that when revenue disappoints, the pain won't stay confined to public shareholders. It cascades through an opaque, interconnected, largely unregulated part of the financial system in ways regulators, by their own admission, don't fully understand yet.

Physical Build-Out Unable to Match the Timeline

The physical build-out simply can't happen on the timeline current valuations assume, because the grid, the transformers, the turbines and the skilled labour to install them don't exist in the required volume, and can't be conjured in under seven years even with unlimited capital. Layer geopolitical strain, Keen cites the wars touching Ukraine and the Middle East, plus China rebuilding its strategic oil reserve, on top of that, and you get what he calls "a triple whammy of bleakness."

What This Means for Business Leaders, Not Just Investors

None of this means artificial intelligence is a mirage. Every historical analogue Keen invokes, railways, electrification, aviation, the internet, produced a technology that genuinely transformed the economy, even as the specific companies financing its early build-out mostly went to the wall. The lesson isn't avoid AI. It's distinguish the technology from the balance sheets financing it. A collapse in hyperscaler and private-credit valuations would be a serious financial event, but it wouldn't erase large language models, chip design advances, or the productivity gains already banked in sectors from logistics to manufacturing to, closer to home, the same AI adoption curve is already reshaping European auto finance.

For anyone in industries built around long capital cycles, automotive manufacturing, leasing, energy transition, the more durable lesson is about physical planning horizons rather than the crash itself. China's grid advantage wasn't built for AI. It was built over two decades for industrial policy and manufacturing competitiveness, and AI simply arrived to find the plumbing already in place. The West's failure isn't a lack of ambition on AI. It's decades of underinvestment in the unglamorous infrastructure, transformers, substations, high-voltage corridors, that any capital-intensive transition eventually depends on. It's also worth returning to a question we've raised before about AI's effect on how we think and decide, because a genuine correction would test how much independent judgement business leaders retained through a boom that rewarded following consensus rather than questioning it.

Frequently Asked Questions

1.Why does China have an electricity grid advantage in AI?

China runs transmission at roughly 800 kilovolts against America's 345 to 500 kilovolt backbone, operates 45 ultra-high-voltage projects spanning over 40,000 kilometres, and maintains generation capacity at nearly double peak demand, versus a US reserve margin of only around 15 per cent.

2.What is private credit's exposure to AI data centers? 

An estimated $1.65 trillion in off-balance-sheet debt sits across five major US tech companies, financed largely through private credit arrangements outside conventional regulatory view, with total exposure potentially three to five times higher once associated bank lending is included.

3.Is the AI boom similar to the 1929 stock market crash? 

Steve Keen argues yes, more so than the dotcom crash, because 1929 involved concentrated margin leverage cascading through forced selling, a structure closely mirrored by today's private-credit-financed AI data centre debt rather than a purely equity-driven bubble.

4.How much cheaper are Chinese AI models compared to American ones? 

DeepSeek's models reportedly achieve 80 to 90 per cent of the performance of leading American systems at roughly a tenth of the cost, while American AI firms are valued at 10 to 25 times their Chinese counterparts offering broadly comparable capability.

5.What is Steve Keen's prediction for the AI industry? 

Keen predicts most American AI companies will fail, with perhaps only one in ten surviving, and that China emerges as the structural winner within roughly two years, driven by sounder economics rather than superior technology alone.