Survey: Chinese companies plan to allocate 46% of their AI accelerator budget to domestic products in the next 12 months, up from 30% today, a shift from Nvidia
Chinese companies are ditching Nvidia Corp.'s advanced accelerators in favor of domestic silicon, underscoring how tensions …
Context & Ripple Effects
The survey’s planned increase in domestic accelerator spending follows a multiyear policy and procurement push: major Chinese internet companies were previously urged to curb Nvidia purchases, and were later reported to be testing local alternatives as their Nvidia inventories tightened.
That shift is already visible in market-share data: IDC said Chinese GPU and AI-chip suppliers took nearly 41% of China’s AI-server market in 2025 while Nvidia’s share fell to 55%. The survey suggests buyers expect that substitution to deepen, even as domestic compute availability remains a constraint.
First-order effects
- Chinese buyers expect to redirect a substantially larger share of accelerator budgets to domestic suppliers over the next year, reducing the portion available to Nvidia in China.
- Domestic AI-chip vendors gain a clearer demand signal for their products, while Chinese customers must expand deployment and evaluation of local accelerator stacks.
Second-order effects
- Nvidia faces greater pressure to defend its China business through the products it can sell and through software and ecosystem support, while local suppliers face stronger incentives to improve compatibility, reliability, and supply.
- Large AI developers’ testing and adoption of domestic hardware can pull demand toward associated servers, systems integration, and software tooling built for those accelerators.
Third-order effects
- If planned budgets translate into purchases, China’s AI-compute market could become more structurally split between a domestic hardware-and-software stack and Nvidia-centered infrastructure.
- The reported need to permit limited purchases of Nvidia H200 chips alongside the domestic push indicates that substitution is likely to be uneven: strategic preference for local silicon does not by itself eliminate performance or supply constraints.
The trend: China’s AI-compute market is moving from dependence on imported accelerators toward policy-supported, progressively validated domestic alternatives, with shortages and capability gaps determining the pace.