AI build-out hits snag as Google and Microsoft put off raising capex forecasts

Microsoft cut its full-year capex forecast to $170 billion from $175 billion, while Alphabet kept guidance at $195 billion–$205 billion; both companies’ reported capex figures came in below analysts’ estimates. Alphabet missed earnings and revenue estimates, reported more than $9 billion of negative free cash flow, and fell about 10% in extended trading, while Microsoft beat estimates as Azure sales grew 48% and its shares were cheered. Delayed GPU capacity and funding concerns for non-investment-grade neoclouds add risk, although hyperscaler capex is expected to exceed $1 trillion next year and Microsoft sees spending accelerating in the first half of 2027.

Analysis

The key distinction is timing and financing—not whether AI demand exists. Delayed third-party capacity can defer revenue conversion for AI providers even while hyperscalers’ long-run spending plans remain intact. That favors platforms able to fund and monetize capacity internally: Microsoft’s Azure growth supports resilience, while weaker external providers face a more immediate utilization and refinancing test. If credit tightens, the likely second-order effect is consolidation and a larger share of AI workloads returning to hyperscaler-owned infrastructure, not necessarily a collapse in total AI investment.

The market may be overreading capex undershoots as demand weakness: component and deployment delays can shift spending into 2027. But the counter-risk is that higher costs and negative free cash flow make returns on incremental compute less attractive, especially if capacity arrives after customer demand or financing has weakened. Company-level attribution of the delays to CRWV, IREN, or NBIS is unconfirmed; treat their selloff as sector risk repricing, not proof of a specific execution failure.

Near term, Amazon’s capex and capacity commentary is the next read-through. Over 1–3 months, watch cloud growth, delivery schedules, utilization, and neocloud borrowing terms. Over 6–18 months, the likely divergence is between cash-generative platforms and capital-dependent infrastructure providers. The thesis weakens if hyperscaler capex guidance is cut alongside slowing cloud growth, or if neocloud funding costs ease while delivery and utilization improve.

Key Decisions for Investors

• Prefer a relative-value position: long MSFT versus a basket of CRWV, NBIS, and IREN, sized conservatively. Microsoft’s cloud growth and financing capacity offer a more robust way to express AI demand than leveraged capacity buildouts. Treat the neocloud short leg as a hedge rather than a clean single-name thesis; the article does not identify which providers caused the delays.
• Wait for Amazon’s report before adding broad hyperscaler exposure. An intact or higher capex outlook with credible delivery plans would support the view that spending is deferred, not cancelled; a cut paired with weaker cloud demand would falsify it and argue for reducing the long leg.
• Keep GOOG exposure event-driven rather than treating one weak quarter and negative free cash flow as proof of a structural AI-spending problem. Reassess on subsequent cloud growth, free-cash-flow trajectory, and evidence that outside capacity is arriving; sustained deterioration across those measures would invalidate the wait-and-see stance.
• Monitor financing terms, maturities, and customer concentration at CRWV, NBIS, and IREN before increasing shorts. Easier credit, improving utilization, or confirmed delivery catch-up could trigger sharp rebounds; deteriorating access to capital would strengthen the consolidation-risk thesis.