Morgan Stanley: hyperscaler AI capex to hit $3.4T across 2027-2028 with leases included
FinanceYF5 · x · 2026-10-21
Morgan Stanley projects that, once operating leases are counted, hyperscalers and neoclouds will spend $1.58T on AI infrastructure in 2027 (up 62%) and $1.81T in 2028 — nearly $3.4T over two years, billed as the largest infrastructure buildout in history.
Key points:
- 2026 spending estimated at $978B, accelerating sharply thereafter
- By 2028: Google $434B, Amazon $314B, Microsoft $310B, Meta $303B — over $1.36T from the big four alone
- Heavy use of operating leases means cash-capex-based estimates significantly understate the real scale
- Money flows across the whole supply chain: GPU, HBM/DRAM, advanced packaging, networking, optics, power and cooling (NVDA, AVGO, MU, VRT, plus neoclouds like NBIS and CRWV)
- This may not be a short capex cycle: inference, agents, robotics and physical AI are still early — but firms without strong balance sheets and durable financing risk being squeezed out
More from Venture
- Anthropic Hits $16.8B Quarterly Revenue as Growth Slows to 45% — victor_explore · 2026-10-21
- Senja cofounders met in person after 5 years, having built a $850K ARR business fully remote — nico_jeannen · 2026-10-21
- Salesforce to acquire Braze for $5.66B, betting on real-time customer data — emmanuelvivier · 2026-10-21
- Netflix down ~50% from $569B peak as YouTube runs away with US TV viewing time — asymco · 2026-10-21
- From Musk's SpaceX Shares to Anthropic's $48B Fund: AI Companies Are Kicking Off UBI — marcvanderchijs · 2026-10-21
- Indie hacker Gefei55: if Build In Public is your only play, you deserve to earn nothing — gefei55 · 2026-10-21