🚨BREAKING: TWO CHINESE HEDGE FUNDS JUST CALLED THE AI TRADE A SUPER BUBBLE.. THE MAN WHO CALLED CHINA'S 2007 MARKET TOP SAYS SOME AI NAMES COULD FALL 80%..

That is the part worth sitting with before you decide how seriously to take the call.
Shanghai Banxia told its investors that the trigger for the AI bubble bursting has already appeared, and it named a specific mechanism: Anthropic's annualized revenue run-rate is going to fall short of what the market has priced in. Large tech companies will recoil at token costs. Competitors will erode its developer base. The application layer revenue will not justify the infrastructure spend underneath it.
It is a coherent thesis. The problem is that Anthropic is a private company, and private companies do not publish revenue run-rates.
So where did Banxia get the number it is stress-testing? The honest answer is that we do not know. It could be a leaked figure from an investor update, which Anthropic sends to its backers at Google, Spark Capital, and others. It could be an inference built from API pricing, token volume estimates, and enterprise contract sizing, the kind of bottoms-up modelling sophisticated funds do when public data is thin. It could be something a former employee said in a room somewhere. The Bloomberg piece does not say, and Banxia declined to comment.
This matters more than it sounds.
If Banxia's number comes from a disclosed investor update, it is probably accurate but also already months old by the time it circulates to a fund in Shanghai. Anthropic's revenue has been compounding fast enough that a six-month-old figure could be significantly below where the run-rate sits today. The thesis might be built on a lag rather than a real slowdown.
If the number is modelled rather than sourced, then Banxia is essentially predicting that their estimate of Anthropic's revenue is lower than the market's estimate of Anthropic's revenue, neither of which is publicly auditable. That is a bet on relative optimism gaps between private figures, which is a very different thing from a verifiable data point.
The only way this thesis gets confirmed or killed publicly is if Anthropic discloses a revenue figure that disappoints, which would likely happen through an IPO filing, a funding round announcement, or a leak. Until one of those happens, Banxia's call sits in an epistemically awkward place. It is either a well-sourced insider read on private company performance, which would be the most important contrarian signal in markets right now, or it is a sophisticated-sounding inference that cannot be checked against reality until it is too late to act on.
What you can track as a proxy: Anthropic's API pricing changes, the frequency of enterprise case studies it publishes, the volume of developer integrations being announced, and whether any of its large anchor customers, Amazon, Google, Salesforce, quietly reduce their disclosed usage commitments. None of those are the revenue number. But they are the surface area of a private company's growth trajectory, and right now they are all you have.
The most credible version of Banxia's call is not that Anthropic is failing. It is that Anthropic's growth, even if impressive in absolute terms, will not be impressive enough relative to the valuation the entire AI infrastructure trade has been priced against. That is a much softer and more plausible claim. And it does not require a leaked number to be true.
It just requires the market to eventually ask for receipts.