$700B AI capex bet just flipped your pricing window

The numbers are not subtle. Meta guided $115–135B, Alphabet $175–185B, Amazon ~$200B, and Microsoft is on track for ~$145B annualized — combined ~$655–685B in 2026, a 36–71% jump over 2025. Morgan Stanley puts hyperscaler debt issuance at $400B+ this year, more than 2× last year. Amazon will run free cash flow negative by roughly $17B. This is the first time in a decade that the four largest cloud sellers are spending faster than they earn, and they have to defend it on a public stage today.

Inside hyperscaler sales orgs, that math has already cascaded. Microsoft printed $77.7B in Q1 with Azure +40% and a $13B AI run-rate; Satya now has to defend an Azure +37% guide on $145B of capex. AEs covering you have been given Q2 utilization quotas tied to that public number. Until Memorial Day, every Microsoft, Google, and AWS account exec on your account has more authority to discount, more authority to extend ramp, and more authority to throw in burn-down credits than they will from June onward. Q2 utilization is what underwrites Q3''s narrative — and the ones writing the narrative know it.

Most CFOs are doing the opposite of the right thing. They are reading the capex headlines and deferring renewals into Q3, betting the ''AI bubble'' optics will force concessions later. Wrong direction. Capex doesn''t reverse — it ratchets. Microsoft and Google issued forward capacity guides today; once those data centers come online late Q3, marginal cost flips to monetization mode and the discount window collapses. The contrarian play, the one we are running on three premium-tier renewals this week, is to pull renewals forward, not push them back.

The silicon shift nobody is pricing into renewals: Meta''s MTIA 450 doubled HBM bandwidth and MTIA 500 adds another 50%; Microsoft Maia 200 claims 3× FP4 over Trainium3; Google Trillium is past 100K deployments; custom ASIC growth is 44.6% CAGR, with inference now two-thirds of all AI compute. Within 12 months, your ''AI stack'' silently becomes a ''vendor silicon stack'' — workloads that run cheap on Maia don''t port to Trillium without re-quantization. The only counter is to negotiate model-portability rights this quarter, while capacity providers still treat them as table-stakes concessions instead of hostage-taking levers.

Named moves from the last 14 days that prove the window is real: Anthropic locked $30B of Azure capacity at a disclosed cost rate while simultaneously renegotiating its DoD contract over surveillance clauses. Microsoft and OpenAI dropped exclusivity on April 27 — that single line item will lift Microsoft''s estimated Copilot enterprise margin by ~11%. Meta signed a multi-billion-dollar Amazon Graviton5 deal because its own MTIA 450 won''t ship in volume until Q4. If the people building the silicon are still cross-buying capacity to hedge, the operator who locks single-vendor commits this quarter is buying the same risk they are hedging out of.