Alphabet Raised Capex to $205B and the Stock Fell

Alphabet reported the strongest quarter in its recent history on July 22 and the stock went down. That reaction is not irrational, and unpacking it explains more about the state of AI infrastructure than the beat itself does.

The headline numbers were unambiguous. Revenue of $119.8 billion, up 24% year over year, against consensus near $117 billion. Google Cloud at $24.8 billion, up 82% — not a strong quarter for a cloud business, a step-function change in trajectory. Cloud operating margin expanded to 35.6% from 20.7% a year earlier, which is the harder trick: growing that fast while getting dramatically more profitable.

And the backlog. Google Cloud's remaining performance obligations reached $514 billion, up more than $50 billion sequentially in a single quarter. Management expects to recognize just over half of that as revenue within the next 24 months. That is a contracted, signed pipeline larger than the annual revenue of nearly every company on earth.

Shares fell about 4% in after-hours trading.

The line that moved the stock

CFO Anat Ashkenazi raised full-year 2026 capital expenditure guidance to $195–205 billion, up from $180–190 billion. Second-quarter capex alone was $44.9 billion, roughly 107% higher year over year. And 2027, she said, will increase significantly again.

A $15 billion mid-year guidance raise is not a rounding adjustment. Alphabet is now planning to spend, in one year, something in the neighborhood of what it earned in revenue across all of 2016. Roughly 60% of technical infrastructure spending goes to servers — a mix of Google's own TPUs and Nvidia GPUs — with the remaining 40% to data center construction and networking.

The market's discomfort is not really about the money. It is about what the raise implies. Ashkenazi's explanation for the increase was accelerated delivery of capacity to meet demand, and then, in the Q&A, the sentence that did the damage: "We're still in a supply-constrained environment."

Google has been saying some version of this for over a year. The expectation, reasonably, was that $180–190 billion would finally close the gap. Instead the number went up and the constraint stayed. That combination reframes the capex line from investment to tax — a cost of participation that scales with demand rather than a one-time buildout that eventually amortizes into margin.

Renting compute is the real signal

The most consequential disclosure got the least attention. Ashkenazi said Alphabet plans to expand its use of third-party capacity in the third quarter as a bridge while it builds more of its own.

Sit with that. Alphabet — which operates one of the largest private computing fleets in existence, designs its own accelerators, and has been building data centers for over two decades — is renting compute from other providers to serve its cloud customers.

This is not a failure of engineering. It is a statement about physics and lead times. You cannot compress an interconnection queue. You cannot accelerate a gas turbine order book that is sold out for years. You cannot conjure substations, transformers, or the electricians to install them. Alphabet has the capital and the demand and still cannot pour concrete fast enough, so it is arbitraging someone else's already-poured concrete in the meantime.

For anyone modeling the neoclouds — the CoreWeaves and Nebiuses of the market, whose bear case has always been the hyperscalers will eventually build their own and stop renting — this is the counterargument, delivered by a hyperscaler. Structural undersupply lasts longer than any single company's build cycle.

What the backlog is actually saying

The $514 billion figure deserves more scrutiny than it usually gets, in both directions.

Skeptically: backlog is contracted, not collected. Multi-year AI commitments have shown a habit of being renegotiated, and a meaningful slice of hyperscaler backlog across the industry traces back to a small number of very large AI labs whose own revenue is younger than their compute commitments. Concentration risk is real.

Constructively: the sequential jump of more than $50 billion is the number that matters, because it is a flow, not a stock. Enterprises signed that in three months. And Alphabet's own guidance — half of it converting within 24 months — puts a fairly aggressive clock on recognition. This is not a pipeline stretched across a decade to look impressive.

The tension is that both the backlog and the capex are compounding, and Alphabet is essentially telling investors it will keep spending as long as returns look attractive. That is the correct answer if the demand is durable and a catastrophic one if it is not. Nobody, including Alphabet, currently has the evidence to settle it.

The broader read

Three of the largest technology companies in the world are now running capital intensity that looks less like software and more like utilities or telecom in a buildout cycle. The market has spent a year rewarding AI revenue and tolerating AI spending. Alphabet's reaction — a genuine beat on every operating metric, sold off on a spending guide — suggests that tolerance is thinning.

The question shifting into focus is not whether AI demand is real. Google Cloud growing 82% with expanding margins settles that. The question is whether returns on this scale of capital can hold up when the capital keeps escalating and the constraint refuses to clear.

Alphabet just told the market it will spend up to $205 billion this year, more next year, and that it still cannot meet demand. Depending on where you sit, that is either the most bullish sentence of the quarter or the most alarming one.