Big-4 hyperscaler 2026 capex $725B (+77% YoY); MSFT $190B, GOOG $190B, AMZN $200B, META $115-135B; Alphabet 'compute constrained'; Goldman Sachs 11 GW shortfall today widening to 45 GW by 2028; cycle is accelerating, not saturating.

Autoresearch: CSP Capex Cycle Status — Q1 2026

Generated by /autoresearch on 2026-06-05. Synthesized across 1 round from 3 web sources. Early exit — question answered definitively. Context: vault/projects/stock-market. Treat as raw material — review before promoting.

Summary

The CSP capex cycle is accelerating, not decelerating. Q1 2026 earnings confirmed the Big-4 hyperscalers (MSFT, GOOG, AMZN, META) will spend a combined $725 billion in 2026 — up 77% from $410B in 2025. Individual guidance: Microsoft $190B, Alphabet $190B (raised $5B during Q1 earnings), Amazon ~$200B, Meta $115–135B. Alphabet CEO Sundar Pichai flagged the company is "compute constrained in the near term." AWS grew 28% YoY (fastest in 15 quarters); Google Cloud grew 63% YoY. Goldman Sachs estimates a 11 GW US data center capacity shortfall today widening to 45 GW by 2028. The Allianz macro team called the AI capex cycle "war-proof" in March 2026. NVIDIA captures ~90% of AI accelerator spend; Q4 data center revenue hit $62.31B (+75% YoY). The conclusion for the ai-capex-to-power-and-materials-cascade and FERC mechanisms is: the forcing function is intact and strengthening.

Findings

Capex Numbers — Q1 2026 Earnings Confirmed

Combined Big-4 2026 capex: $725 billion (+77% from $410B in 2025). Individual:

Company	2026 Capex Guidance	Notes
Amazon	~$200B	Sharp jump from prior years
Microsoft	$190B	Above $152B analyst average; CFO Hood: $25B attributed to component cost inflation
Alphabet	~$190B	Raised $5B at Q1 earnings; Q1 alone $35.7B (+107% YoY)
Meta	$115–135B	Tied to Meta Superintelligence Labs buildout

Source: Yahoo Finance / Tom's Hardware analyst note; Next Web Q1 earnings roundup.

Cloud Revenue Acceleration (Validating Demand)

AWS: $37.59B in Q1 2026 (+28% YoY) — fastest growth in 15 quarters
Google Cloud: ~$20B (+63% YoY)
Sundar Pichai (Q1 2026 earnings): company is "compute constrained in the near term" — capex is not ahead of demand; it is chasing it

Alphabet's quarter-on-quarter capex increase: $35.7B in Q1 2026 vs. ~$17B in Q1 2025 (+107%). This is not a steady-state number; each quarter is substantially larger than the prior year.

Supply Constraint and Physical Bottleneck

Goldman Sachs (cited in multiple 2026 market analyses): US data center capacity shortfall exceeds 11 GW today, projected to widen to 45 GW by 2028. This is the physical-world constraint that translates hyperscaler capex commitments into demand for power infrastructure, grid interconnection (FERC RM26-4-000), and construction materials.

Dell'Oro Group (independent market research): Global data center capex approaching $1T in 2026; projected to reach $1.7T by 2030.

Allianz Economic Research (March 2026): AI capex is "war-proof" — spending commitments are not sensitive to tariff disruption or geopolitical uncertainty given the strategic nature of AI infrastructure competition.

NVIDIA — Accelerator Spend Capture

NVIDIA captures ~90% of AI accelerator spend. Q4 (calendar year) data center revenue: $62.31B (+75% YoY). This is the read-through for the pick-and-shovel position on the capex cycle; NVIDIA's data center growth directly tracks hyperscaler GPU capex.

Open Question Resolution

The prior open question was whether the CSP capex cycle was approaching saturation or continuing to accelerate. Q1 2026 earnings answer this definitively: accelerating. All four major hyperscalers raised or maintained guidance above prior estimates; one (Alphabet) explicitly stated it is supply-constrained, not demand-constrained. The bear thesis on capex deceleration has been repeatedly falsified by sequential quarter data.

Implications for Active Mechanisms

ai-capex-to-power-and-materials-cascade: Forcing function strengthening. $725B capex in 2026 is a 77% increase; the downstream demand for power, grid capacity, cooling, and construction materials is proportional.
FERC large-load interconnection (RM26-4-000): 11 GW shortfall → 45 GW by 2028 validates the urgency of the June 2026 FERC rulemaking.
CEG nuclear baseload: Compute-constraint + power-constraint converging; nuclear baseload with existing interconnection rights is the scarcest solution.
Cooling infrastructure (HON liquid cooling TAM): $725B in AI-optimized data centers creates a proportional sensor/controls TAM upgrade cycle.

Provenance

Rounds run: 1 of 3 (early exit — question answered with high confidence in Round 1)

Sub-questions:

Round 1:

Individual hyperscaler Q1 2026 capex guidance (Microsoft, Google, Amazon, Meta)
AI data center capacity shortfall — supply constraint metrics
Cycle trajectory: acceleration vs. deceleration signal

URLs used (search snippets; no additional fetches needed):

Tools used: WebSearch Generated: 2026-06-05