Nine Cloud Giants Will Spend $830B on AI Infrastructure in 2026 — TrendForce Lifts Growth Forecast to 79%
The nine largest cloud providers — Google, AWS, Meta, Microsoft, Oracle, ByteDance, Tencent, Alibaba, and Baidu — are on track to spend $830 billion on capital infrastructure in 2026, according to a TrendForce report published May 6. That figure represents 79% year-on-year growth, a significant upward revision from TrendForce’s earlier estimate of 61%.
The revision follows a wave of earnings guidance increases across North American hyperscalers, each citing stronger-than-expected AI demand as the driver.
The Per-Company Breakdown
AWS is the biggest spender at $230B+, up more than 50% year on year, fuelled by AI cloud services and generative AI workloads. Microsoft has raised its 2026 outlook to $190B, implying 130% growth — of which roughly $25B reflects rising component costs across GPUs, memory, and networking. Google increased its guidance range to $180–190B (previously $175–185B), with growth exceeding 100%. Meta revised upward to $125–145B from $115–135B, representing about 85% growth.
TrendForce notes the pace of expansion among North American providers now outstrips the global average by a wide margin.
Strategy Divergence at the Chip Layer
Not all four US hyperscalers are spending the same way. TrendForce identifies a clear split: Microsoft and Oracle are going heavy on GPU clusters; Google is doubling down on custom TPUs; AWS is running a dual strategy across both GPU and ASIC; Meta is still testing its MTIA custom silicon at scale before committing.
This divergence matters. Labs and enterprise customers that deploy on a specific cloud increasingly inherit that provider’s chip architecture — GPU-native workloads and TPU-optimised inference paths are not interchangeable at production scale.
Electricity Crossover — 2026 Is the Year
Total global data center installed power capacity is forecast to reach approximately 155 GW in 2026, up 29% year on year. More significantly, AI servers are projected to surpass conventional general-purpose servers in total electricity consumption this year, crossing a threshold that has structural implications for grid planning, power purchasing agreements, and cooling infrastructure.
The gap will widen further in 2027–2028 when platforms such as GB300, Rubin, and the next generation of ASIC-based AI servers enter mass production. TrendForce flags HVDC power systems and liquid cooling suppliers as direct beneficiaries.
What the $830B Buys
The spending is concentrated in three areas: high-performance GPU clusters, in-house ASIC development, and next-generation hyperscale data centres designed for high-power-density workloads. By end-2025, the top five North American CSPs collectively operated 800–900 data centres globally, with AWS holding the largest footprint.
The Chinese contingent — ByteDance, Tencent, Alibaba, Baidu — is also accelerating overseas data centre deployment to sidestep domestic energy constraints and serve global markets, a trend TrendForce expects to intensify through 2027.
Key Numbers
| Provider | 2026 CapEx | YoY Growth |
|---|---|---|
| AWS | $230B+ | 50%+ |
| Microsoft | $190B | ~130% |
| $180–190B | 100%+ | |
| Meta | $125–145B | ~85% |
| Top 9 Combined | ~$830B | 79% |
Global data center installed power: ~155 GW (+29% YoY). AI server electricity consumption: overtakes general-purpose servers for the first time in 2026.