China’s reported Nvidia chip “stop-buy” and the end of its Google probe — the real signal for AI supply chains

Beijing has reportedly told firms to halt purchases of Nvidia AI chips, even as it formally ends an antitrust probe into Google. The mixed signals look tactical, not contradictory — easing one pressure point while hardening another.

We’ve tracked the pivot for weeks: regulators discontinuing the Google case while increasing scrutiny on Nvidia aligns with China’s push for supply sovereignty. For context, our earlier piece on Beijing’s shift is here, and our analysis of Nvidia’s restricted-spec parts and tepid reception in China — think RTX 6000D — helps explain the policy friction.

What’s confirmed vs reported

• Reported stop-buy: Reuters relayed foreign-ministry comments after reports that Chinese regulators ordered firms to stop buying Nvidia AI chips. The ministry said it was “willing to maintain dialogue,” without confirming specifics.
• Probe ended: Separately, Beijing has formally ended its Google antitrust investigation, per Financial Times reporting cited by Reuters.
• Market read-through: Nvidia’s China-specific RTX 6000D has reportedly seen limited traction with major firms, reflecting the performance compromises of export-compliant SKUs.

Why this is happening now

China is optimising leverage in parallel: a policy olive branch toward one U.S. firm while tightening the supply choke-points that matter more to its AI build-out. If “restricted-spec” accelerators are framed as anticompetitive or strategically risky, domestic buyers have a policy pretext to delay or cancel — and to accelerate local alternatives.

Implications for buyers and builders

• Procurement volatility: Expect extended lead times, last-minute cancellations, and a shift to domestic GPUs where software permits.
• Model road-maps change: Teams may prioritise efficient architectures and smaller training runs — a trend underscored by DeepSeek’s $294k R1 claim.
• Software moat pressure: CUDA lock-in remains strong, but policy can force migration; tracking ROCm and local stacks will matter more than benchmark deltas.

What to watch next

1. Any formal circulars or procurement guidance that clarifies the scope/duration of purchase restrictions.
2. Import/export scrutiny on PCIe data-center SKUs and workstation cards that edge into AI workloads.
3. Domestic accelerator rollouts and price/perf disclosures, plus software ecosystem maturity.