Sundar Pichai Says Alphabet Can't Build AI Capacity Fast Enough, and Anthropic Has Secured 5 Gigawatts of It.

Alphabet said it remains supply constrained in AI computing capacity despite committing to up to $205B of 2026 capex, but it is monetizing the scarcity through Anthropic deals covering 1GW+ of Google TPU capacity in 2026 and ~5GW online starting in 2027. Google Cloud revenue rose 82% YoY to $24.8B and operating income more than tripled to $8.8B, with margin up to ~36% and cloud backlog growing to ~$514B (from $462B). While Alphabet is investing heavily ahead of returns (2026 capex guidance raised to $195B–$205B; Q2 capex $44.9B vs cash from ops $39.1B), the company is effectively pre-selling contracted AI infrastructure and strengthening its future cloud earnings profile.

Analysis

Alphabet’s AI capacity is functioning less like a traditional cloud sale and more like a long-dated annuity backed by scarce supply. That matters because the market usually pays up for scarcity only when utilization is visible; here the better setup is that capacity is being pre-sold before it is built, which should support cloud multiple expansion if management keeps converting backlog into margin rather than just revenue. The cleanest second-order winner is GOOG/GOOGL; the cleaner beneficiary outside the name is AVGO, because every additional custom-silicon deployment deepens the non-NVIDIA AI stack and reinforces Broadcom’s role as the picks-and-shovels partner.

The main risk is duration mismatch. Alphabet is funding capacity ahead of cash generation, so the short-term accounting optics can look worse before they look better, especially if capex stays elevated and financing costs rise. The falsifier is not the next headline deal; it is any evidence over the next 1-3 quarters that cloud margin inflects down, backlog growth slows, or Anthropic’s demand is not strong enough to absorb the committed build-out. Over 6-18 months, the real threat is a supply catch-up that turns scarcity into commoditization and compresses returns on capital.

Consensus is underweight the embedded option value of being both supplier and investor, but likely overweight the certainty of monetization. The market is treating AI infrastructure as if all gigawatts are equal; in practice, the winner is the platform with pricing power and contract duration, while the loser is any customer whose growth trajectory does not keep pace with reserved capacity. That makes this a good name to own on pullbacks, but a poor one to chase after capex enthusiasm has already been marked up.

Key Decisions for Investors

• Long GOOG/GOOGL on weakness into the next cloud print; target a 3-6 month hold if cloud margin stays above ~30% and backlog keeps compounding, with thesis invalidation if margin slips back toward the low-20s.
• Add AVGO as the cleaner second-order AI infrastructure beneficiary; the trade works best if management commentary confirms continued TPU/custom-silicon demand over the next 1-2 quarters.
• Relative-value: long GOOG/GOOGL vs. AMZN for 1-3 months if the market starts rewarding AI monetization visibility over raw capex scale; cut if AWS reaccelerates or Google Cloud growth decelerates sharply.
• Avoid chasing NVDA on this headline alone; this is supportive for total AI spend, but the incremental dollar is more likely to accrue to internalized silicon and cloud platform economics than to merchant GPU share.
• Watch item, not trade: if Anthropic financing or revenue momentum weakens over the next 2 quarters, reduce GOOG/GOOGL exposure because the market will re-rate the backlog quality, not just the size.