China builds $1T AI compute grid as token demand hits 140T/day
Key insights
- China's daily AI token calls exceeded 140 trillion in March 2026, a 1,000x increase from early 2024.
- China's NDRC projects more than $1 trillion in computing network investment in 2026 alone.
- Beijing's 15th five-year plan formally reframes telecoms operators as core AI infrastructure providers.
Why this matters
A state-mandated $1T compute buildout treated as utility infrastructure creates a procurement and capacity floor that private cloud markets cannot match in pace or scale, directly affecting how Western AI firms benchmark their own infrastructure investment cycles. Telecoms operators being legally repositioned as AI infrastructure providers establishes a regulatory template that other governments may adapt, reshaping how hyperscalers and national carriers negotiate compute access globally. The 1,000x token demand growth in 15 months also validates that inference-side compute, not just training, is the constraint that will define competitive AI deployment at national scale through at least 2027.
Summary
China's daily AI token calls hit 140 trillion in March 2026, a more than 1,000x increase from the start of 2024, and Beijing is treating that growth as an infrastructure emergency requiring a national response.
The NDRC is now accelerating construction of a nationwide computing power network explicitly modeled on the electrical grid, connecting data centers, supercomputing facilities, and intelligent-computing clusters across regions. State media are framing compute as a public utility, and the 15th five-year plan documents call for a "multilayered computing infrastructure system" that treats telecoms operators as core AI infrastructure providers rather than communications businesses.
Essentially: (China's NDRC, major state telecoms operators) are fusing AI compute into the national utility stack.
- Investment in these networks is expected to exceed 7 trillion yuan ($1T) this year alone, per NDRC projections.
- Token demand growth of 1,000x in roughly 15 months signals that Chinese AI deployments have moved from pilot to mass adoption at a pace that outran existing data center capacity.
- Telecoms operators, recast as AI infrastructure providers, now face mandatory build-out obligations similar to grid operators.
The initiative reframes sovereign AI capability as a utilities question, putting China's compute buildout on a policy and funding trajectory that has no direct equivalent in the US or EU.
Potential risks and opportunities
Risks
- US export-controlled chip restrictions may slow China's buildout timeline, forcing reliance on domestic alternatives like Huawei Ascend, which could create a 12-24 month capacity gap that inflates inference costs for Chinese AI firms.
- Concentration of compute in state-managed utility infrastructure gives regulators direct leverage to throttle or surveil AI workloads, creating compliance risk for any multinational operating AI services inside China.
- If the $1T investment projection reflects planned spending rather than committed contracts, a slowdown in Chinese AI commercial revenue could leave state telecoms operators overextended on infrastructure debt by late 2027.
Opportunities
- Domestic Chinese AI chip suppliers (Huawei HiSilicon, Cambricon, Biren) gain a captive $1T procurement pipeline insulated from foreign competition, accelerating their ability to close the gap with Nvidia on inference hardware.
- Cooling, power, and data center construction vendors operating in China (Vertiv, Delta Electronics, locally listed infrastructure firms) face a multi-year demand surge as the utility-scale buildout requires physical infrastructure at grid parity.
- Western AI firms with inference-optimization software (Groq, Cerebras, SambaNova) could find leverage in markets outside China where governments watch this buildout and seek to replicate the compute-as-utility model without depending on US hyperscalers.
What we don't know yet
- Whether the 140 trillion daily token figure includes all inference workloads or only API-routed calls, which would significantly affect how the demand baseline compares to US or EU figures.
- Which specific telecoms operators (China Telecom, China Unicom, China Mobile) have been assigned build-out targets and on what timeline within the 15th five-year plan period.
- How the NDRC plans to allocate the 7 trillion yuan across regions, and whether western provinces with cheaper power will receive disproportionate data center capacity as with earlier cloud buildouts.
Originally reported by scmp.com
Read the original article →Original headline: China Accelerates National AI Computing Power Network as Daily Token Demand Surpasses 140 Trillion — NDRC Eyes $1T Investment