The big-four hyperscalers are on pace for >$700B 2026 capex (+77% YoY) and >$1T in 2027, but Alphabet's raise sold off and AI-bond CDS demand is at records — the trigger is a financing/credit break, not physical glut.

Autoresearch: Megacap Q2 2026 capex raises and the AI-ROI repricing

Generated by /autoresearch on 2026-07-29. Synthesized across 2 rounds (early-exit) from web search across earnings-preview and capex-tracking sources. See Provenance. Treat as raw material — review before promoting. Context: vault/projects/stock-market

Summary

The Q2 2026 megacap prints (Alphabet 07-22; Microsoft/Meta/Amazon reporting this week 07-28→07-31) confirm the AI-capex cycle is still accelerating, not peaking: combined big-four capex is on pace for >$700B in 2026 (+77% from ~$410B in 2025), and Evercore/BofA now see hyperscaler capex exceeding $1T in 2027. But the market's reaction function has flipped — Alphabet's raise to $195–205B (from $180–190B) prompted immediate selling, the clearest sign that "how big" has been replaced by "who can prove monetization." The load-bearing new evidence for the bear leg is in the credit market, not physical oversupply: investment is scaling ~50% faster than revenue, FCF is under pressure, MS/JPM estimate the sector needs ~$1.5T of new debt, ~$100B of AI-capex bonds have printed in 2026, and investors are demanding record CDS protection against them. This sharpens ai-roi-reckoning and routes the actual trigger to mega-issuance-peak-to-ai-capex-derate / ai-terminal-value-uncertainty-to-credit-curve-steepening — a financing break, corroborating the 07-28 framing that the near-term risk is timing, not glut.

Findings

Capex is still being raised — hard

Alphabet raised FY2026 capex to $195–205B (07-22), up from $180–190B a quarter earlier, on Google Cloud +82% YoY to $24.8B and a backlog that jumped +$50B in one quarter to $514B (Value Add VC — Alphabet $205B; Seeking Alpha — AI capex warning).
Microsoft ~$190B CY2026 (incl. ~$25B from higher component pricing — a read-through to memory/HBM inflation, i.e. memory-shock-to-oem-value-transfer / hbm-supply-bottleneck); Amazon ~$200B; Meta $125–145B after its own upward revision (Value Add VC — $725B; S&P Global — MSFT/Meta previews).
Combined big-four >$700B in 2026 (vs ~$410B in 2025); adding Oracle, the top-5 have committed $660–690B, ~75% ($450B) directly AI infrastructure (Yahoo/Futurum — $700B; Futurum — $690B sprint). 2027: >$1T per Evercore/BofA.

The reaction function flipped — raises now sell off

Alphabet's guidance raise "prompted immediate selling, signaling investors now demand clear AI returns over unchecked spending" (Value Add VC). The H2-2026 debate is "no longer 'how big' but 'who can prove monetization'" (Futurum).
Meta's response — Meta Compute (07-01), renting excess GPU + hosting Muse Spark — is a direct attempt to turn surplus infra into a revenue line, i.e. the monetization pressure made concrete (Futurum); ties to meta-cloud-business-fourth-hyperscaler-disruption.

The trigger is credit, not physical glut

Investment scaling ~50% faster than revenue — payback pushed further out each quarter (Forbes — capex-to-revenue gap; Futurum).
FCF among the big five under pressure; MS/JPM estimate ~$1.5T of new debt needed to finance construction; ~$100B of bonds issued in 2026 for AI capex, with investors demanding record CDS protection against them (Futurum; Introl — capex/debt Jan 2026). This is the observable that would fire the financing-break falsifier on mega-issuance-peak-to-ai-capex-derate before any physical vacancy signal.

Contradictions and open questions

Fundamentals still support the spend near-term. Google Cloud +82%, +$50B backlog — the demand side isn't cracking, so the derate is a financing/valuation event, not a demand event yet. The bull (compute-as-financialized-commodity) and bear (ai-roi-reckoning) legs both survive this print.
What's the datable credit trigger? CDS spreads on AI-capex issuers and the pace of the $1.5T issuance are now the machine-checkable metrics to watch — sharper than DC vacancy. Worth a falsifier metric on the derate mechanism.
Component-pricing pass-through (MSFT's ~$25B) confirms the memory/HBM inflation leg is real inside hyperscaler capex — a bullish read for hbm-cowos-as-binding-bottleneck even as it pressures hyperscaler FCF.

Provenance

Rounds run: 2 (early-exit — two broad searches resolved the capex figures and the credit-market tell; no productive round 3).

Sub-questions by round:

Round 1:

What are the Q2 2026 megacap capex guides (MSFT/GOOGL/META/AMZN), and how did the market react?
How does 2026–27 hyperscaler capex compare to cloud revenue and FCF, and where's the financing strain?

Anchor source: none (time-sensitive earnings query).

URLs surfaced (via WebSearch synthesis):

Tools used: WebSearch. Generated: 2026-07-29