Autoresearch: CSP AI capex cycle — peak or sustained? Q1 2026 earnings verdict

Generated by /autoresearch on 2026-05-25. Synthesized across 2 rounds (early-exit after R2 — R3 confirmed no new material). All synthesis from search snippets (WebFetch blocked system-wide, HTTP 403). No Grokipedia anchor (403). See Provenance. Context: vault/projects/stock-market

Summary

The Q1 2026 earnings season delivered a definitive answer: the hyperscaler AI capex cycle is not peaking — it is accelerating. The four largest cloud providers (Amazon, Google, Microsoft, Meta) collectively committed $650–725 billion in 2026 capex, a ~62–64% increase from 2025's record $388–400B. Goldman Sachs projects the cycle continues through 2027 (consensus has been wrong low for two consecutive years). The binding constraint has shifted: it is no longer chips or capital — it is power infrastructure. Grid interconnection queues run 4–7 years in key markets. Only 5GW of the 16GW announced US 2026 pipeline is actually under construction. This creates a durable, multi-year investment theme for grid equipment (GE Vernova, Eaton, Hubbell), power/cooling infrastructure (Vertiv), and electrical construction (Quanta Services, MYR Group). Demand is supply-constrained at the cloud layer (Microsoft has $80B in Azure orders it cannot fulfill due to power), which means the capex cycle for physical infrastructure has a long visible runway.

Findings

Q1 2026 earnings: unanimous capex raises

Every major hyperscaler raised its 2026 capex guidance in Q1 2026 earnings:

Company	2026 Capex Guidance	vs Prior	Notes
Amazon	~$200B	Up sharply from $125B 2025	Largest single commitment
Alphabet	$180–190B	Up from $175–185B	CFO flagged additional increase planned 2027
Meta	$125–145B	Up from $115–135B	Raised citing component pricing + DC costs
Microsoft	~$110–120B	Up from $90B 2025	Q3 FY2026 single quarter capex +84% YoY; Azure >$37B annual run rate

Combined: $650–725B for 2026. (NextWeb Q1 2026 earnings; Yahoo Finance hyperscalers $700B)

Azure supply-constrained: demand exceeds supply

Microsoft has $80 billion in Azure orders that cannot be fulfilled due to power constraints. PTU waitlists are 6–9 months in some regions. Northern Virginia and Texas Azure regions have been limiting new subscriptions due to space, rack, and server shortages. Management confirmed demand exceeds capacity. This is structurally bullish for physical infrastructure buildout — the demand pipeline is visible and funded. (Windows Forum, Azure supply crunch)

Power is the binding constraint

The bottleneck is no longer chips or capital — it is power infrastructure:

• Grid interconnection queues in Northern Virginia, Phoenix, and Dallas: 4–7 years
• A campus joining the NOVA interconnection queue in May 2026 cannot realistically draw utility power before 2030–2033
• Against a 16 GW announced 2026 US pipeline, only 5 GW is actually under construction
• 30–50% of US data center projects planned for 2026 are now expected delayed or cancelled
• Uptime Institute: total global AI data center power load hits 10 GW by end of 2026
• Analysis to PJM governors warns of 49 GW US generation shortfall by 2028
• US utilities planning $1.4 trillion in AI data center power infrastructure

(tech-insider.org, 7GW capacity crisis; Data Center Frontier, gigawatt bottleneck; tech-insider.org, utilities $1.4T)

Goldman Sachs: cycle through 2027 and beyond

Goldman's model: capex projected at $1.15T from 2025 through 2027 (more than double the $477B spent from 2022–2024), growing to $1.6T in annual capex by 2031. Consensus has been wrong-low for two consecutive years (consensus projected ~20% growth each year in 2024 and 2025; actual exceeded 50% both years). Data center occupancy projected to peak above 95% in late 2026, with moderation starting 2027 as new capacity comes online — but this is occupancy, not capex; the capex for capacity that comes online in 2028–2029 is being placed now. (Goldman Sachs, $500B+ 2026; Goldman Sachs, tracking trillions)

Picks-and-shovels: Q1 2026 beneficiary scorecard

GE Vernova (GEV): Q1 2026 orders +71% YoY to $18.3B; backlog $163B; $200B backlog milestone pulled forward from 2028 to 2027. Data center-related orders: $2.4B in Q1 alone (already exceeding 2025 total). 20% of 100GW under contract explicitly supporting data center infrastructure. Acquired remaining 50% stake in Prolec GE for $5.3B (adds $5B high-demand transformer backlog). Electrification margin expanded 640bps YoY to 17.6%. Guidance raised. (Yahoo Finance, GEV backlog surges; howtheymake.money, GEV Q1)

Eaton (ETN): Q1 2026 record Electrical Americas sales $3.6B; operating profit $922M; Electrical Americas backlog +44% YoY from March 2025. Boyd Thermal acquisition (March 2026) adds liquid cooling capability. (heygotrade.com, VRT analysis citing Eaton context)

Vertiv (VRT): Q1 2026 revenue $2.65B; EPS +83% YoY; project backlog doubled to >$15B; 2026 guidance raised to $13.5–14B (30% organic growth, 51% EPS growth). Described as the "cleanest public proxy for AI data center power and cooling buildout." (heygotrade.com VRT analysis)

What a capex peak would look like (bear case markers to watch)

The Goldman occupancy model suggests utilization peaks in late 2026 and moderation could start 2027 — that's when the question re-opens. Specific signals to watch for a slowdown:

1. Any hyperscaler reducing 2027 capex guide in Q2/Q3 earnings (currently all pointing up)
2. Azure PTU waitlist shortening (would signal demand is clearing)
3. Power interconnection queue shortening (structural, 3–5 years away)
4. NVidia GPU delivery shortfalls (would slow DC buildout even with capital available)

Contradictions and open questions

• Capex going in vs. capacity coming online: GS occupancy model suggests moderation starting 2027. But the capex being placed NOW is for capacity that becomes operational in 2028–2029. These are two different cycles that can diverge.
• 30–50% of 2026 projects delayed: this is a demand-for-equipment signal (delayed, not cancelled), but actual backlog conversion timing for GEV/ETN/VRT depends on when projects reactivate. If delays are long, near-term revenue recognition may lag backlog.
• Meta investor concern: Meta stock dropped ~6% after raising capex — investors questioning whether revenue is scaling to justify the spend. This is the bear case signal for the whole cycle if it spreads to MSFT/GOOG/AMZN.
• Nuclear/SMR as a long-range power solution: mentioned but described as "tiny in 2026's pipeline composition." This is the right framing for now but could change significantly by 2028.

Provenance

Rounds run: 2 + partial R3 (early-exit R3 — no new material; May 22–25 delta search confirmed no new announcements in the 3-day window, all material from Q1 earnings).

Grokipedia anchor: fetch failed (HTTP 403, consistent with all prior session failures).

Sub-questions by round:

Round 1 (broad survey):

1. What did MSFT/GOOG/AMZN/Meta say about AI capex in Q1 2026 earnings?
2. Analyst forecast revisions — peak or sustained?
3. What equipment/cooling/power/construction companies are most exposed?
4. Azure supply crunch specifics

Round 2 (drill-down):

1. Power/grid bottleneck composition — what's binding first?
2. GE Vernova Q1 2026 specifics and backlog
3. Vertiv and Eaton Q1 2026 specifics

Round 3 (early-exit):

1. Any new developments May 22–25? — Finding: no new material; all relevant data from Q1 earnings season (April–May).

URLs searched (all WebFetch 403, all synthesis from snippets):

Round 1:

• NextWeb — Q1 2026 $650B — snippet
• Yahoo Finance — hyperscalers $700B — snippet
• Goldman Sachs — $500B+ 2026 investment — snippet
• Goldman Sachs — tracking trillions — snippet
• heygotrade.com — Vertiv VRT — snippet
• Windows Forum — Azure supply crunch — snippet

Round 2:

• tech-insider.org — 7GW capacity crisis 2026 — snippet
• Data Center Frontier — gigawatt bottleneck — snippet
• tech-insider.org — utilities $1.4T — snippet
• Yahoo Finance — GEV backlog surges — snippet
• howtheymake.money — GEV Q1 2026 analysis — snippet

Tools used: WebSearch, WebFetch (all blocked), grokipedia-fetch (blocked). Generated: 2026-05-25