Did Big Tech Just Fail the Capex Test, or Pass It?


Alphabet reported cloud revenue up 82% year on year. That is not a good number; that is a spectacular number, well past what the street was looking for.

The stock dropped more than 7% the same day. Worst day in over a year.

I have been sitting with that for two weeks, and I think it is the most important thing that happened in the markets this quarter, more than any of the beats that came after it.

The number nobody was looking at

Here is what investors actually reacted to. Alphabet raised its 2026 capital expenditure guidance to as much as $205 billion and posted its first negative free cash flow quarter since it went public in 2004. Twenty-two years of positive free cash flow, gone, and the reason was AI infrastructure.

Meta reported a week later. Revenue up 28% to $60.8 billion. Fine. Good, even. But free cash flow fell from $8.5 billion to $784 million because $31.1 billion of capex ate almost all of the $31.9 billion the business generated in operating cash. Shares fell as much as 11%.

Two companies. Both grew revenue. Both got hit.

Somebody at Bold Wealth Partners told Fortune that capex used to be a case of the more the better, and now it's the less the better. That's the entire shift in one line. For three years, spending on AI was read as a signal that a company saw demand coming. Now it's read as a cost.

Source: Fortune




So why didn't the market crash?

Because Microsoft and Amazon told a better story, and I mean that fairly literally.

Microsoft went up about 8% on the print. Amazon went up around 10%. Both of them are spending enormous amounts of money too; Microsoft guided FY2027 capex at $255 to $260 billion, well above what anyone expected. That's more than Alphabet. It went up anyway.

The difference was the sentence attached to the spending. Microsoft's message was that Azure demand exceeds supply. Amazon said something similar and pointed at bookings. Alphabet and Meta spent the money and asked the market to trust them on the payoff.

Same behavior. Different framing. Opposite share price reaction.

Source: TradingKey earnings scorecard

I find that genuinely unnerving, and it is why I do not think we have dodged anything. If the difference between an 8% gain and an 11% drop is whether management can convincingly say the buildout is demand-constrained rather than speculative, then what's holding the market up is a narrative. Narratives are the thing that goes first.


What I actually think the bubble is

I do not think the AI bubble is a question of if. I think it has always been there. The technology is real, the revenue is real, and the bubble is still there, because a bubble is not about whether something is useful. It's about the gap between what has been paid for and what has been proven.

But "it will pop eventually" is a useless thing to say. Everything pops eventually. So here is the version I will actually defend:

The trigger is not valuation. It is a ratio. When infrastructure cost is growing faster than the revenue it produces, and the gap widens instead of narrowing, investors stop treating spending as a growth signal and start treating it as a bottleneck. That is the exit signal. Not a specific P/E number, not a Fed meeting. The moment the spending stops looking like an investment and starts looking like a bill.

Alphabet's negative free cash flow quarter is the closest thing we have to a date stamp on that flip.

One company failing that test gives you a single-stock crash. Meta, this quarter, more or less. What worries me is the version where half the sector fails it in the same fortnight, and there is no Microsoft left to reassure anyone. That is not a correction. That is a panic.


The part that should sound familiar

Nothing about the Asian Financial Crisis started with bad revenue. Thailand had factories. Korea had exports. The growth was real, the savings rates were high, the fundamentals people quoted at the time were genuinely there. What broke was the funding structure underneath. Money that could leave in ninety days was paying for assets that took ten years to earn out.

Look at what is happening now. Data centers, land, power contracts, chips. Long assets. Judged by a market that reprices you every ninety days on a quarterly report. And increasingly, free cash flow gets consumed, funded with debt rather than cash from operations, which is the exact shift the BIS flagged as a pressure point in its 2026 Annual Economic Report. Their wording: the five largest hyperscalers are set to spend over a trillion dollars on AI capex across 2025 and 2026; those commitments are outpacing earnings and free cash flow, and some firms are issuing debt to cover the gap.

That is a maturity mismatch. It is the same shape. Different asset, different continent, same problem.

I am not saying we are heading for 1997. The hyperscalers have balance sheets that Thai finance companies could only dream about, and that matters. But the thing that made 1997 dangerous was not that the projects were bad. It was that the funding could disappear faster than the assets could pay off. If AI capex keeps sliding from cash flow onto debt, that is the ingredient getting added.

Where this goes

The honest answer is that Q2 was a warning, not a verdict. Demand is real, the beats were real, and I would be lying if I said the bears got their evidence.

But the market's tolerance visibly narrowed. Alphabet posted 82% cloud growth and got punished for what it spent to get there. That is not the behavior of investors who believe the returns are coming. That is the behavior of investors who have started counting.

Next week I want to get into how all this is being financed, because Nvidia just signed up half of Wall Street to turn GPUs into tradable securities, and Larry Fink compared it out loud to the birth of the mortgage-backed securities market. Which, given what this blog is about, I have some thoughts on.

Next: Are GPU-Backed Securities the New Mortgage-Backed Securities?

Sources


Location: Malaysia

Comments

Post a Comment