Meta's CapEx surged from $12.9B in Q1 2025 to $21.4B in Q4 2025, a 66% sequential ramp. Analyst commentary implies 2026 full-year CapEx could exceed $100B. The $21B CoreWeave agreement and gigawatt-scale Prometheus cluster confirm Meta as the largest non-cloud AI infrastructure buyer.
↑NVDA accelerates GB300 shipments to Meta; LRCX etch tool orders rise for new fabs
↓NVDA GB300 order book thins; LRCX etch tool orders decelerate on slower fab buildout
CoreWeave and leased capacity commentary
$21B expanded agreement through 2032 (announced Q1 2026)
↑CoreWeave accelerates NVDA GPU purchases; owned+leased capacity both expanding
↓CoreWeave deal substitutes for owned CapEx; net GPU demand flat vs prior trajectory
Gigawatt-scale cluster deployment timeline
Prometheus cluster (1GW+) coming online in 2026, Hyperion scaling to 5GW over several years (Q2 2025 call)
↑Eaton, Vertiv power equipment orders accelerate; NVDA rack-scale demand confirmed
↓Eaton, Vertiv power equipment orders slip right; NVDA rack-scale demand pushed to 2027
What peers said
AmazonAmazon's forward guidance notes potential volatility from memory chip supply and resource constraints as material ris..., also seeing stable on ai server capacity
2021 signals · 56 high-qualityResearch read-through · not a trade recommendation