As AI hype grows, cloud computing becomes more expensive
Price shock: cloud providers raise rates as AI demand surges
Alibaba Cloud (阿里云) and Baidu Intelligent Cloud (百度智能云) this month announced broad price increases on AI compute and high‑end storage — effective April 18 — following an earlier, steeper move by Tencent Cloud (腾讯云). Overseas peers were first: AWS raised EC2 training instance prices in January and Google Cloud later lifted some AI‑infrastructure fees. Why the pivot after nearly two decades of falling cloud prices? Simple: AI has turned compute from a commodified utility into a scarce strategic resource.
Tokens, agents and a global compute squeeze
The technical story is about token consumption. AI agents and large multimodal models drive orders of magnitude more token throughput than conventional chatbots, and that demand is concentrated on high‑end GPUs and fast storage. It has been reported that IDC projects explosive token growth through 2030, and industry sources say large domestic players are hoarding capacity — ByteDance (字节跳动) alone has reportedly amassed tens of thousands of H20 GPUs — squeezing what little spare supply cloud operators can sell. Geopolitics compounds the squeeze: Nvidia’s H100/H200 family remains dominant, and U.S. export controls and global supply constraints have tightened access to top‑tier accelerators.
Rising component costs force margin reckoning
On the cost side, manufacturers are raising prices across the semiconductor and parts chain. TrendForce has reportedly revised DRAM and NAND price forecasts sharply upward for early 2026, and suppliers such as Samsung have announced double‑digit NAND increases; Murata’s notice of MLCC hikes for AI server applications underscores broader component stress. For cloud providers that long competed on scale and low margins, these are not cyclical blips but structural cost shifts. As a result, vendors are reallocating scarce AI compute toward token‑intensive, higher‑value services and moving from “selling resources” to “selling intelligence” — a commercial refocus that helps explain why only AI compute and high‑end storage lines are being repriced while basic VM rates remain stable.
What does this mean for enterprises? Expect higher bills for large‑scale model training and inference, and a faster consolidation of strategic compute inside hyperscalers and a few well‑capitalized customers. In short: the economics of cloud are being rewritten by AI, and the change will reverberate through supply chains and global tech competition.
