Internet/e-Commerce: Raising Internet capex ests: A look into Internet capex, capacity and monetization

研报英文原文证据摘录

Internet/e-Commerce: Raising Internet capex ests: A look into Internet capex, capacity and monetization

ata center capacity. We estimate the Big-3 MTIA: Meta Training and Inference Accelerator

Internet mega caps had ~27GW capacity exiting 2025, which will grow to 39GW in 2026 TAM: Total Addressable Market

and 57GW in 2027. We estimate Amazon will add the most capacity in 2026-2027 at LLM: Large Language Model

15GW, followed by Google at 9GW and Meta at ~6GW, with Amazon’s capacity additions

at a lower cost than peers (at around $24bn in AWS capex per GW per 2025 disclosures).

Recent capacity deals suggest growing capacity value

Assuming 70% of Amazon’s AWS capacity is used for Cloud (vs 30% for core), we

estimate AWS revenue per GW of capacity at $10.6bn in 2026, with Google Cloud at

$15.7bn. These revenues/GW are well below recent capacity deals by Anthropic and

Google with SpaceX at up to $50bn per GW (for specialized AI capacity), which give us

optimism on future capacity related revenue upside. For Meta, we estimate the company

could have up to ~23GW of capacity by 2030, and if 40% is available for AI enterprise

sales, we estimate a $110bn potential enterprise opportunity (assuming $12bn/GW).

Internet hyperscaler valuation analysis

For our valuation analysis, we back out est. valuations for core advertising and retail

revenues (at 12-15x implied P/Es) and look at valuations imbedded in hyperscaler stocks

per unit of GW capacity. While our analysis is subject to numerous assumptions &

excludes the impact of projects like Leo and Reality Labs., we think our analysis

identifies the limited implied value the Street is giving to Meta’s capacity build. Our

analysis suggests an implied value for Google at $110bn per 2028 Cloud GW (assuming