Survey: Chinese Companies Accelerating Shift to Domestic Suppliers for AI Chip Procurement
Miles Bennett
A Bloomberg Intelligence survey shows Chinese companies plan to allocate 46% of their AI chip budgets to domestic products over the next 12 months, up from 30% today — the US-China tech decoupling is moving from policy rhetoric to real procurement dollars.
What does 46% actually tell us?
Domestic AI chips currently account for 30% of enterprise procurement; that figure is set to rise to 46% within a year — an increase of more than half.
This means → the domestic substitution push has moved past the "signaling stage" into actual budget reallocation.
The survey covered 60 executives across software, finance, manufacturing, and retail — the people who sign off on AI infrastructure purchases.
Why is Nvidia's share shrinking?
Nvidia's products remain popular among Chinese firms, but supply of its H20 chip — designed specifically for the China market — is tightening.
Beijing has urged domestic tech companies to stop using the H20, so even willing buyers face a closing gate on the policy side.
In plain terms = Nvidia's China problem is not just "being replaced" — supply and policy are squeezing from both ends at once.
Who stands to capture the windfall?
The report names Tencent, Alibaba, and Huawei as best positioned — all three are chip users *and* chip developers.
On the chipmaker side, AI accelerators from Hygon Information Technology (海光信息) and Cambricon (寒武纪) are being actively evaluated by a large share of respondents.
This reflects a beneficiary pool that is heavily concentrated among top-tier players; smaller chipmakers have not yet entered the mainstream procurement conversation.
Where has the bottleneck shifted?
The report notes that China's AI bottleneck is moving from compute chips themselves to securing supply of HBM — high-bandwidth memory chips that enable high-speed data transfer for AI training and inference.
A global memory-chip shortage could constrain expansion at firms like SMIC (中芯国际), while CXMT (长鑫存储) stands to benefit from the gap.
In plain terms = domestic compute alternatives are emerging, but the memory chips that feed data to those processors have become the new chokepoint.
Can a ¥2 trillion buildout hold together?
China is planning roughly ¥2 trillion (about $294 billion) over the next five years for a nationwide data-center network.
At least 80% of the core technology — chips included — is to be supplied domestically.
This means → whether domestic chips can keep pace on both performance and supply reliability will determine if this substitution wave is a temporary procurement tilt or a lasting structural shift.
What does 80% overspend tell us?
The survey found that 80% of executives said total infrastructure spending this year has exceeded budget, primarily because of higher-than-expected AI project costs.
This reflects an urgency to keep pace with AI buildout — firms are spending over plan rather than falling behind.
In plain terms = domestic substitution may be the strategic direction, but right now it's costing more than planned, and the budget pressure is real.
Content is for reference only, not financial advice.