CHINA DATA CENTERS · PROJECT PROFILE

China Mobile Hohhot: Large-Scale AI Infrastructure and Its Operating Evidence

Hohhot is a documented large-scale AI computing deployment, with operator disclosures that extend beyond a general data center construction announcement.

Hohhot, Inner Mongolia, ChinaSources reviewed 23 October 2026
China Mobile’s Hohhot campus in a September 2026 Digital China report; the image does not identify a particular GPU room.
China Mobile’s Hohhot campus in a September 2026 Digital China report; the image does not identify a particular GPU room. Photo: Digital China / szzg.gov.cn.

At a glance

Operator: China Mobile

Documented status: Entered operation in 2024, corroborated in the operator’s annual report

What the sources establish

An August 2024 China Mobile announcement hosted by SASAC describes the Hohhot AI computing center as deployed with approximately 20,000 AI accelerator cards. It discusses model migration, training optimization and inference deployment as part of the service offering. China Mobile’s 2024 annual report subsequently confirms the operation of ten-thousand-card-scale centers in Hohhot and Harbin. China Mobile via SASAC: Hohhot AI computing center China Mobile Annual Report 2024

The dated operating disclosures are more informative than a proposed-building count, while still leaving important current performance questions open.

Why the software layer matters

Large accelerator deployments need coordinated networking, monitoring, recovery and software support to become useful training infrastructure. An inventory alone does not show how reliably a distributed job runs, how efficiently devices communicate or how quickly faults are resolved.

For buyers and developers, the appropriate next evidence is a workload-specific acceptance record: the model or benchmark, precision, software stack, sustained execution period and recovery behavior. A service can then be compared on its demonstrated operating conditions rather than its device count alone.

Evidence and comparison limits

The approximately 20,000-card figure is a 2024 disclosure, not a new 2026 inventory audit. No conversion to another accelerator platform’s equivalent performance or site MW is made.

What to follow

Follow workload throughput, cluster reliability, software compatibility and newer site-level operating disclosures.

Original information and official project sources

Open the underlying operator material, public-authority disclosure or certification register below. Source dates and reporting periods are separate from this profile’s publication date.

China Mobile via SASAC: Hohhot AI computing center ↗Source date / reporting period: 2024-08-28 · Reviewed 23 October 2026China Mobile Annual Report 2024 ↗Source date / reporting period: 2024 reporting year · Reviewed 23 October 2026

← Back to China Data Centers