Analysis: AI cloud competition shifts to token cost and compute utilization

On November 12, 2026, Zhidongxi published an analysis titled "The Next Battle in the AI Cloud Market: Token Cost and Compute Utilization," arguing that the criteria AI cloud customers evaluate providers by are shifting. Citing an employee at an intelligent computing cloud company, the article notes that over the past six months client priorities have changed noticeably. Customers used to ask how much one GPU cost and how long it could be rented; now the dominant questions are how much one million tokens cost and whether workloads can run reliably to completion. The piece describes this shift as a meaningful upgrade in AI infrastructure.

On market size, the article cites a Gartner forecast projecting that the AI cloud market will reach $267 billion by 2030, roughly CNY 1.78 trillion, with neo-cloud or NeoCloud providers expected to capture 20% of that, a segment corresponding to a market on the scale of CNY 350 billion. On utilization, the article draws on a survey conducted in April of this year by IDC and the China Academy of Information and Communications Technology, covering 250 enterprises. It reports that nearly half of surveyed companies keep intelligent computing resource utilization in the 51% to 70% range, while in sectors such as manufacturing and government services, companies exceeding 70% utilization account for less than 35%.

The piece frames the industry's next challenge as finding ways to use the compute capacity already built so that it is deployed more fully, more flexibly, and more reliably. It adds that over the past two years neo-cloud firms have completed an initial validation of their business models through GPU procurement, cluster construction, and compute leasing, and that competitive benchmarks are now moving along with changes in compute architecture, AI workloads, and customer needs. So far the discussion remains at the level of trend analysis and a stage summary, with no follow-up concrete actions reported.