Alibaba Cloud today informed users it will increase prices for many services by up to 34 percent.
“Due to the surge in global AI demand and rising supply chain costs, the procurement costs of core hardware in the industry have increased significantly,” states a price adjustment notice posted on Wednesday. “After careful evaluation, we have decided to adjust the prices of services,” it adds.
Updated price lists record dozens of services and instance types for which costs will rise by five percent. The cost of higher-end instances powered by GPUs will rise by 25 to 34 percent. Even instances running Alibaba’s own software, like the PolarDB cloud-native database, cop the price rise.
The cost of using accelerators in the Chinese cloud will rise by five to 30 percent. The increases even apply to Alibaba’s homebrew Pingtouge Zhenwu 810E, a parallel-processing ASIC for AI applications said to match Nvidia’s dumbed-down-for-China H20 GPU.
Alibaba’s notice says customers who purchased relevant services before April 18 2026, “will not be affected by this adjustment in your current order/billing cycle; the new price will apply at the start of your next renewal cycle.”
That appears to be good news for some Alibaba Cloud customers, as the outfit allows customers to subscribe to cloud services for one-or two-year terms, and in increments of three, six, and nine months.
Alibaba’s price hikes are probably reasonable in the context of surging memory prices which create unavoidable cost increases.
Hiking prices for compute, however, looks a little opportunistic given last November Alibaba Cloud said it can’t install servers fast enough to keep up with demand, and was rationing access to GPUs for customers who spend more money. The Chinese cloud has also previously claimed to possess resource optimization prowess that let it operate more efficiently than rival clouds.
Alibaba Cloud is not the only hyperscaler to have hiked prices of late: AWS classily did it on an early January weekend, sneaking out news of a 15 percent hike for certain machine-learning-centric resources. ®
Source: The register