Confirmatory: $725-750B combined hyperscaler capex 2026 unchanged; Meta raised to $125-145B; transformer/switchgear shortages pushing $150-200B from 2026 to 2027-2028. No major new signals since May 15.

Autoresearch: CSP capex cycle — May 18 2026 incremental update

Generated by /autoresearch on 2026-05-18. Synthesized in 1 round from search snippets. No Grokipedia anchor. Treat as raw material. Context: vault/projects/stock-market

Summary

No major new developments since the May 15 comprehensive CSP capex update. The $725–750B combined capex figure for 2026 (top-4 to top-5 hyperscalers) remains unchanged. Incremental: Meta explicitly raised its 2026 capex guide to $125–145B (from $115–135B) during Q1 2026 earnings, citing "higher component pricing and additional data center costs." The new data point most relevant to the existing wiki is the electrical bottleneck: 25–50% of U.S. planned data center projects face delays/cancellations due to transformer, switchgear, and battery shortages (transformers now 5-year lead times vs. 2–3 years pre-2020). This pushes $150–200B in capex from 2026 into 2027–2028 — a structural tailwind for the datacenter-construction hypothesis but means near-term silicon demand may be slightly softer than the headline capex implies.

Findings

Capex numbers: $725–750B confirmed, Meta raised guide

Combined top-4 hyperscalers (MSFT, Alphabet, Amazon, Meta) guiding to ~$725B for 2026 per Tom's Hardware, with top-5 (adding Apple/Oracle) reaching ~$750B per CreditSights. Meta specifically raised its full-year 2026 guide from $115–135B → $125–145B, citing higher component pricing (memory/HBM) and additional DC costs. This is the clearest post-May 15 update.

Capex intensity at unsustainable-looking levels: 54% of Meta revenue, 47% MSFT, 46% Alphabet, 25% Amazon. All guided higher despite cash flow pressure. The supply-constrained narrative holds: "markets are supply-constrained, not demand-constrained" per all four companies.

Electrical bottleneck: $150-200B deferred to 2027-2028

A new quantitative finding: between 25% and 50% of U.S. data center projects planned for 2026 face delays or cancellations, primarily due to transformer, switchgear, and battery shortages. Transformer delivery stretched from 24–30 months pre-2020 to as long as 5 years today. This pushes an estimated $150–200B in 2026 capex into 2027–2028.

Implication: the 2026 capex headline is real demand, but the physical conversion of that capex into deployed compute is partially deferred. This reinforces the picks-and-shovels trade (electrical infrastructure: VRT, PWR, HUBB, FIX) as a multi-year thesis, not a 2026 event.

Contradictions and open questions

Capex intensity (47–86% of revenue for some) raises the Damodaran-style concern about whether returns on invested AI capex justify the spend. Enterprise AI pilot fail-rate (IBM: 95%) is still cited as a medium-term risk.
The electrical bottleneck suggests the $725B headline spend does NOT all hit silicon demand in 2026 — the sequence is: commitments now → construction delays → deployments in 2027-2028.

Provenance

Rounds run: 1 of 3 (early exit — no material new data since May 15 to justify further rounds)

Round 1:

"hyperscaler capex 2026 Microsoft Amazon Google Meta guidance update May 2026"
"AI data center spending 2026 peak slowdown supply constrained demand May 2026"
"hyperscaler capex Q2 2026 earnings guidance raised AI infrastructure May 2026"

URLs fetched: 0 (all 403). Search-snippet-based synthesis.

Key source URLs:

Tools used: WebSearch (3 searches). WebFetch all 403. Generated: 2026-05-18