Sounds conspiratorial? Maybe. But the mounting concerns are real.
The mooted OpenAI and Anthropic initial public offerings are getting pushed back, even as the record-breaking US capital expenditure spree continues to accelerate, and debt piles rise.
The US hyperscalers, along with Nvidia and Broadcom, have more than $3 trillion in off-balance-sheet commitments and guarantees, with more "financing structures under development," according to Morgan Stanley.
On top of this is the intricate web of circular financing that raises concerns even among some AI optimists - AI could revolutionize society, but an unwind of these convoluted financing schemes could create a lot of pain for investors along the way.
The real risk keeping US AI leaders up at night may be less Arnold Schwarzenegger and more Adam Smith.
Cheap large language models from China are proliferating rapidly, putting significant competitive pressure on costlier US offerings. For example, a version of Chinese startup DeepSeek's flagship AI model could be more than 100 times cheaper to run than Anthropic's Claude Fable 5 model. Products from other Chinese firms are up to 50 times cheaper than leading US versions, and Chinese LLMs are rapidly closing the performance gap with the more expensive US frontier models, analysts say.
The US AI industry is facing this rising competition by spending and investing on a massive scale. Hyperscalers are expected to invest up to $1.5 trillion in AI next year.
“You have to spend money to make money, or at least the hyperscalers are testing the motto. The AI buildout is expensive, but likely worth it,” says Ryan Sweet at Oxford Economics.
But for that to be the case, the big outlays must be followed by big revenues. Apollo Global Management's Torsten Slok estimates that hyperscalers' operating cash flow needs to more than triple from last year's $600 billion to $2 trillion in 2030 to justify these vast expenditures. That’s a high bar. And it’s one that could become even more difficult to reach if China’s cheaper LLMs see their market share continue to grow.
If these gains don’t materialize, hyperscalers’ capex may be slashed, slowing chipmakers’ revenue growth and Wall Street’s AI-driven rally. This, in turn, could ultimately slow US economic growth.
Few expect any firm understanding on AI to be hashed out at Trump and Xi’s meeting this week, but the statements from both on the technological front could speak volumes about whether US tech firms have more to fear from rogue agents or simple economics.
QQQ•
