AI valuations are reaching stratospheric levels, but this is also intensifying concerns if they might now be in bubble territory.
Sridhar Vembu, who stepped down as Zoho CEO earlier this year to focus on deep R&D as Chief Scientist, has never been shy about calling out hype. “AI is clearly an investment bubble,” he posted on X. “The justification is that all massive tech waves spark financial bubbles so saying it is a bubble doesn’t negate the tech itself. And this one is the biggest bubble yet. How to navigate this without losing one’s shirt is the key,” he added while quote-posting a viral post.
The Round-Trip Revenue Trick
The post Vembu amplified lays out a specific and damaging structural argument. The core claim is that much of what passes for AI revenue is actually a circular accounting loop — not real demand.
Here is how it allegedly works. A tech giant — Microsoft, Google, Amazon, or Oracle — makes a headline investment in an AI startup. But rather than cash, the investment is partly or wholly delivered as cloud credits. The startup then uses those credits to rent the very servers owned by its investor. The cloud giant records that usage as fresh revenue from a paying customer. In effect, the same money travels in a circle, and gets booked as a sale at the end of it.
The documented example cited is Microsoft and OpenAI. When Microsoft committed $13 billion to OpenAI, much of it came in the form of Azure cloud credits. OpenAI used those credits to train its models. Microsoft then counted that compute usage as cloud revenue. The result: OpenAI’s annual cloud bill has reportedly ballooned to over $60 billion — more than double its actual revenue of $25 billion — sustained in part by this recycled funding loop rather than real external demand.
Anthropic runs a structurally similar arrangement. Over nine months, it reportedly spent $2.66 billion on Amazon Web Services — a figure that at the time roughly equalled its total revenue. Amazon is both Anthropic’s largest investor and its primary cloud provider. Google holds a significant stake in Anthropic too. The lines between investor, customer, and supplier are completely blurred.
This is not a small footnote. Corporate filings show that OpenAI and Anthropic together account for over half of the combined $2 trillion future cloud backlog held by Microsoft, Oracle, Google, and Amazon. Microsoft reportedly has 49% of its $627 billion future backlog tied to OpenAI. Oracle has an extraordinary 54% of its $553 billion pipeline resting on OpenAI alone. These are not diversified books of business — they are concentrated bets on startups that are themselves funded by the same companies holding the backlog.
Paper Profits Built on Paper Gains
The second mechanism in the argument concerns how tech giants are reporting profits. Every time an AI startup raises a new funding round at a higher valuation, the tech giant that invested in it marks up the value of its stake on its balance sheet — and counts that unrealised, paper gain as profit.
The numbers are striking. In Q1 2026, Alphabet reported a record $62.6 billion in profit. But nearly $28.7 billion of that — almost half — was a paper markup on its Anthropic investment. In the same quarter, Amazon reported $30.3 billion in profit, with $16.8 billion of it attributable to an Anthropic paper gain. Meanwhile, Amazon’s actual free cash flow reportedly collapsed 95% to just $1.2 billion, as it spent $44.2 billion in real money building physical data centers. The profits are on paper. The capex is very real.
This creates dangerous concentration risk. Anthropic’s valuation has soared to $380 billion, and there are questions about how its revenue is counted. OpenAI’s own executives have claimed Anthropic overstates its revenue run-rate by $8 billion due to its practice of booking cloud reseller revenue on a gross basis — inflating headline numbers while margins quietly suffer. Even if both companies are genuinely growing fast, the accounting fog makes it very difficult to know what is real.
The Dot-Com Echo
The post draws a direct parallel to the 2001 dot-com crash. During that era, Global Crossing and Qwest Communications swapped identical fibre-optic network capacity with each other simply to manufacture the appearance of sales. Qwest was eventually forced to erase $1.4 billion in fictitious income. Global Crossing went bankrupt. Those swaps were illegal.
The AI loop, the post argues, is functionally similar — but entirely legal under current accounting rules. That distinction matters enormously for how quickly a reckoning might arrive. There is no regulator currently poised to force restatements. The architecture is designed to be compliant, even if it obscures economic reality.
The self-reinforcing nature of the loop is what makes it particularly volatile. Inflated revenues lead to inflated valuations. Higher valuations push AI stocks into index funds and retirement accounts automatically. That inflow drives prices higher still, attracting more capital and more investment — all without the underlying AI technology generating meaningful free cash flow. It is a machine that runs on sentiment and accounting, not profit.
The Bubble Doesn’t Invalidate the Technology
Vembu’s framing is important here. He is not saying AI is worthless — far from it. He has moved away from day-to-day company operations precisely to focus on AI R&D, and has previously warned that India’s software industry faces an existential reckoning as AI restructures what software work actually costs. He takes the technology seriously.
His point is that bubbles and transformative technologies are not mutually exclusive. The railways were real. The internet was real. The bubble around each was also real — and when it burst, it caused enormous collateral damage to investors who confused the value of the technology with the value of the stocks built around it.
The question Vembu implicitly raises is the one that matters most for investors: how do you benefit from a genuine technological shift without being caught holding paper profits when the accounting chickens come home to roost? That is a harder question than either the bulls or the bears typically acknowledge — and it may be the most important financial question of this decade.