← Back to feed

China is drafting a plan to spend approximately 2 trillion yuan ($295 billion) over five years building out national AI infrastructure, according to Reuters and Bloomberg reporting on June 9. The plan is authored by key government agencies and sits in the final stages of inter-ministry review.

The framing is deliberate: AI compute is being classified alongside railways, power grids, and telecom networks — critical national infrastructure operated by the state rather than a privately competitive market.

The Structure

State-owned telecoms — primarily China Mobile and China Telecom — would operate the bulk of the new data center capacity. Both companies already run hyperscale cloud and edge networks; this plan formalises AI compute as their next mandate.

The domestic chip requirement is the figure that matters most for the global semiconductor industry: the plan calls for at least 80% of AI chip technology to come from local suppliers. Huawei’s Ascend line is the only Chinese AI chip at scale. The implication is that the plan would create a captive procurement market for Ascend hardware at a scale that no US export control can directly touch.

For context, the Biden and Trump administrations have spent three years restricting Nvidia’s most capable chips from reaching China. This plan assumes those restrictions hold and routes around them by mandating domestic alternatives — at 80% penetration across a $295 billion program.

The Numbers

$295 billion over five years is $59 billion annually. For comparison:

• The US CHIPS Act committed $52 billion total, not annually, across chip manufacturing and research
• Google Cloud’s Q1 2026 revenue was $20 billion — in a single quarter, and not all of it compute
• Nine major global cloud providers are projected to spend $830 billion combined on AI infrastructure in 2026 (TrendForce)

China’s plan, if executed at stated scale, would represent roughly 35% of what the nine largest global cloud spenders combined will invest in a single year — deployed as a directed state program rather than distributed commercial capital.

What the State-Telco Model Means

The decision to use state telecoms as operators rather than Alibaba, Tencent, or ByteDance is not neutral. It keeps AI compute outside the jurisdiction of China’s private tech sector — the same sector that has faced regulatory crackdowns since 2020.

It also means capacity allocation decisions will go through government channels. Which models run, which data gets processed, and which workloads get prioritised will ultimately be subject to the same industrial policy apparatus that managed the previous wave of Chinese tech build-out.

Western frontier labs operate in a commercial market where any developer with a credit card gets API access. The Chinese model being assembled is closer to a national compute utility — rationed, directed, and procured domestically.

The Timing

The plan follows four months of pressure from US export controls that effectively ended Chinese access to Nvidia B200 and H100 hardware at scale. It also follows xAI, OpenAI, and Anthropic collectively announcing over $200 billion in US compute investments in the same period.

China’s $295 billion is a response to the realisation that the gap in raw compute is widening — and that closing it through the private market is no longer viable given export constraints. The state program is the workaround.

Execution risk is real. China’s large infrastructure projects often overrun timelines, and Huawei’s Ascend chips remain roughly one to two generations behind Nvidia’s current hardware on training performance. But a five-year window and $295 billion creates significant room to close that gap.