Eben@CANSLIM ResearchMorgan Stanley forecasts Rmb8.5tr of total China AI capital expenditure over 2026-30E, enough to roughly triple domestic compute capacity to 81GW by 2030E. The bank’s Asia Pacific team remains Overweight AI infrastructure (Alibaba, Kingsoft Cloud, VNET), AI labs (MiniMax, Zhipu) and GPU localisation. The report’s central tension is financing: Rmb2.0tr of that capex must be raised offshore, and domestic inferencing ROIC starts materially below US levels.
Key Takeaways
- Morgan Stanley forecasts Rmb8.5tr of total China AI capex across 2026-30E, with hyperscaler and internet capex alone reaching Rmb1.2tr in 2027E — roughly 16% of US peers.
- Domestic compute capacity is projected to expand to 81GW by 2030E, with 34GW of net additions from hyperscalers over 2026-30E and 47GW on a global scale.
- Morgan Stanley rates Alibaba (BABA.N), Kingsoft Cloud (KC.O) and VNET Group (VNET.O) Overweight within AI infrastructure, alongside AI labs MiniMax (0100.HK) and Z.AI (2513.HK).
- Self-built GPU IaaS economics show roughly 44% operating margin, ~13% ROIC and a three-year cash payback; rented infrastructure carries no capex but only about 20% operating margin.
- Domestic inferencing ROIC is modelled at just 0.3-17.9% depending on performance versus foreign servers, versus 29% for the foreign-equivalent case.
What Morgan Stanley’s “China’s AI Path” Report Says
In a report titled “China’s AI Path — Rmb8.5tr capex to 3x Compute”, Morgan Stanley’s Asia Pacific equity analysts — led by Gary Yu, with Tom Tang, Yang Liu, Charlie Chan, Rebecca Xu, Lydia Lin, Eddy Wang and associates — set out a five-year build-out case for Chinese AI infrastructure. The industry view is Attractive, and the report is dated September 16, 2026.
Subscribe to continue reading
Become a paid subscriber to get access to the rest of this post and other exclusive content.