Alibaba Raises AI Compute Prices Up to 34% as Chinese Demand Outpaces Supply

Alibaba Cloud raised prices on its T-Head AI chips by 5 to 34% this week, with parallel file storage costs climbing 30% simultaneously. The company cited demand that has outpaced its supply capacity — a direct result of US export controls that have effectively removed Nvidia H100 and H200 chips from the Chinese market.

The Zhenwu 810E, Alibaba’s flagship AI training chip, saw the largest price increase in the T-Head line. For Chinese enterprises that have been building AI infrastructure around T-Head hardware, these increases arrive without an obvious alternative. Nvidia’s latest data center GPUs are off the table. Huawei’s Ascend line has capacity constraints of its own. Alibaba has pricing power it didn’t have two years ago.

This is what supply-constrained markets do. US export controls were designed to slow Chinese AI development by cutting off access to leading-edge compute. They’ve partially succeeded at that goal — Chinese frontier model development does lag on raw compute access. But the secondary effect is a Chinese cloud market that has shifted its infrastructure dependency from American hardware to domestic alternatives, creating oligopolistic pricing dynamics that didn’t exist before.

Alibaba isn’t alone. Baidu reportedly implemented similar price increases on its AI compute offerings in the same window. When the two largest Chinese cloud providers raise prices simultaneously, it signals not coordination but a shared capacity ceiling. Both are selling into more demand than they can currently serve.

The 30% storage increase is a separate issue. Alibaba Cloud Parallel File Storage serves AI training workflows that require high-throughput access to large datasets. Storage infrastructure has been a secondary bottleneck — less discussed than compute, equally real for teams running large-scale training runs.

For Western AI companies watching the China market, the pricing shift has two implications. First, Chinese model developers are now paying a hardware premium that didn’t exist in 2023, which affects the cost structure of competing in that market. Second, Chinese cloud providers have strong incentives to keep optimizing their domestic chip designs — the premium they’re currently charging funds the R&D required to close the gap with Nvidia.

The US-China AI hardware decoupling, roughly three years into its implementation, is producing exactly the market structure you’d expect: higher prices, accelerated domestic investment, and a technology gap that’s narrowing from both ends.