When the Coding Agents Ran the Cloud Out of Room

Generative Value's quarterly tally of the big cloud providers lands on one theme: demand for compute has run past what even the hyperscalers can supply. The trigger it points to is agentic — autonomous coding agents that consume far more compute than a human developer ever did — and the clearest symptom is price, with rental rates for H100 GPUs pushed to 52-week highs.

The spending behind it is enormous. The newsletter's compilation puts combined hyperscaler capex above $700bn for 2026 and still accelerating, with first-quarter capex alone topping $130bn — up roughly 80% year on year. The growth rates underneath are just as striking: Azure up around 40%, Google Cloud up about 63% with its backlog doubling, AWS up 28%. Most telling, Google described itself as compute-constrained — cloud revenue, it said, would have been higher with more capacity. Outside demand signals, like Anthropic's reported revenue run-rate, are the other side of the same coin.

The core idea The constraint has flipped. For a decade the question was whether enterprises would fill the cloud; now the cloud can't be built fast enough for the agents filling it. Capacity, not demand, is the swing variable — which is why the capex keeps climbing. The figures are the newsletter's own compilation of company earnings.

Where it fits

It's the buy-side mirror of what our AI Build-Out index and Agentic Demand pulses track — the capex super-cycle read straight off the hyperscalers' own numbers, with the agent-driven demand we measure as the cause sitting underneath.