China’s cloud giants lift AI compute prices as the year‑long price war ends

What happened

Alibaba Cloud (阿里云) and Baidu Intelligent Cloud (百度智能云) this week both posted public notices raising prices for AI compute and related storage products — increases range roughly 5%–34% for compute cards and about 30% for parallel file storage, according to the vendors’ announcements. The move follows a year in which cloud providers competed aggressively on price — Alibaba famously cut China‑region prices by an average of over 20% in early 2024 — and marks a clear reversal toward monetising scarce compute capacity.

Why prices are rising

The shift is driven by rapidly rising demand and constrained supply. It has been reported that enterprise large‑model token consumption exploded last year, with one consultancy estimating a near‑threefold jump in daily token calls in the second half of 2025. At the same time, hardware and storage costs are rising: Counterpoint research reportedly found memory and NAND flash prices up about 90% in Q1 2026, and high‑end GPU supply remains tight. Amid U.S. export controls and surging global appetite for Nvidia (英伟达) accelerators, it has been reported that deliveries of H100/H200 chips are pushed into 2027 — a bottleneck that makes compute a relatively scarce resource.

So what now?

The implications are immediate. Global players have already echoed the shift — Amazon Web Services and Google Cloud raised ML and data‑transfer prices earlier this year — suggesting an industry‑wide move from "burning money for market share" to "monetising compute." For Chinese startups that have relied on ultra‑cheap cloud subsidies, the squeeze will be acute. Who can absorb higher unit costs and who must prove real revenue per token? The answer will favour vendors with self‑developed hardware and full‑stack optimisation, and it may accelerate consolidation as second‑tier providers face a brutal choice: raise prices and risk losing customers, or accept widening losses.