Big Tech booked more than $160B in Q2 'other income' from AI stakes

So the largest single number attached to AI in Alphabet's quarter is not what customers paid for AI services. It is the revaluation of equity stakes in AI companies, and the sums differ by a factor of four.

The mechanism is straightforward and worth stating plainly. A hyperscaler invests in an AI company. The AI company spends much of that money on cloud capacity, some of it back with the investor. Its valuation rises in a subsequent round. The investor marks up its stake and books the gain as income.

Every step is lawful and normal. The aggregate is a system in which the reported financial benefit of AI to the largest technology companies is substantially a function of private valuations that those same companies help set.

The circularity is now the market structure

The same pattern is visible everywhere this month, and Nvidia is its clearest expression.

Within one fortnight Nvidia agreed to buy Hugging Face for $12.9 billion, committed about $2 billion to Nscale's pre-IPO round, discussed as much as $2.5 billion into Thinking Machines Lab at a $40 billion valuation, backed Lambda as it signed a $35 billion contract with Anthropic, and took the lease on a Texas data centre from Hut 8. It then forecast roughly 70 percent revenue growth for fiscal 2028.

SB Energy, meanwhile, granted OpenAI $5.5 billion in warrants tied to a 20-year lease, ahead of the developer's own IPO.

Chips, capital, buildings, power and demand are held by parties with equity in one another. Each transaction generates revenue for someone and a markup for someone else, and the outside observer cannot separate demand that exists from demand that has been financed into existence.

What would reverse it

Unrealised gains reverse. If AI valuations fall, the same line that produced $97.983 billion of income produces losses of comparable magnitude, and it does so without any change in the underlying operating business.

That is a volatility problem rather than a solvency one for companies of this size. It becomes something more when the marked-up stakes are in companies whose own revenue depends on continued capital expenditure by the entities holding the stakes.

The test is arriving. Anthropic's S-1 is expected in late September, on a valuation case bankers have discussed at up to $2 trillion, resting on projected 2028 revenue of $190 billion to $200 billion. Nscale is raising pre-IPO. A public market price on any of these establishes a mark that the private round-setting process does not control.

Until then, $160 billion of quarterly income across four companies rests on valuations that no arm's-length buyer has tested.