China’s $295 Billion AI Plan Has a Problem — And It Is Made in China

The world's most ambitious computing buildout is running straight into the limits of China's own chip production

A Plan Built on a Bottlenecks

When Beijing announced its $295 billion, five-year plan to build a national AI data center grid, the global reaction focused on the sheer scale of the ambition. What received far less attention was the fine print — and the fine print contains a structural problem that could significantly slow the entire project.

The plan includes an 80% domestic sourcing requirement, which effectively locks out Nvidia and AMD accelerators from all state-funded data center projects — meaning China's $295 billion buildout will be capped by whatever amount of chips SMIC can physically produce. And SMIC, China's most advanced domestic chipmaker, is already running at the limits of its capacity. The foundry's most advanced stable node remains its N+2 process — roughly equivalent to 7nm — and is currently running above 93% utilization, leaving little headroom as every government-certified Chinese chipmaker competes for the same wafer slots.
The ambition is vast. The supply chain to support it is not — yet.

Huawei Braces for $12 Billion — But Can Barely Keep Up

The immediate commercial beneficiary of the domestic sourcing mandate is Huawei, whose Ascend chip line is the only viable domestic alternative to Nvidia's data center products at scale. The numbers confirm the windfall. Huawei braces for $12 billion in AI chip revenue driven by homegrown AI model demand — but Chinese fabs can barely keep up as Nvidia's market share craters within the region.

The supply constraint is compounded by a second critical bottleneck: memory. Highly limited domestic high-bandwidth memory supply constrains how many Ascend-class accelerators Huawei can assemble — and high-bandwidth memory, which is essential for AI training workloads, remains an area where China's domestic production capability lags significantly behind South Korean manufacturers Samsung and SK Hynix, both of which are subject to US export controls on their most advanced products destined for China.

Chinese chip industry leaders have openly admitted that the country lags five to ten years behind in AI data center chips — a candid acknowledgement that stands in sharp contrast to the confidence of the $295 billion plan, and one that raises serious questions about the timeline on which the plan's ambitions can realistically be executed.

The $295 Billion Paradox — And Why It Still Changes Everything

The gap between China's infrastructure ambition and its chip production reality creates what might be called the $295 billion paradox: a plan so large that it simultaneously accelerates domestic chip development through guaranteed demand while exposing the exact limitations that must be overcome for the plan to succeed.
Half of planned US data center builds have been delayed or canceled, with growth limited by shortages of power infrastructure and parts — including parts from China. This detail is quietly extraordinary: the United States' own data center buildout is constrained by shortages of components manufactured in China — the same country it is trying to outpace in AI. The interdependencies of the global technology supply chain have produced a situation where both superpowers are simultaneously building toward AI dominance and depending on each other for the components to get there.

The $295 billion plan will not be executed on schedule, in full, at the chip volumes it implicitly requires. But its existence — as a state commitment, a demand signal, and a political mandate — will reshape China's semiconductor industry more profoundly than any sanction or export control Washington has yet deployed.

Sources: Tom's Hardware, Reuters, Bloomberg, Huawei investor communications