China reportedly orders tech giants to stop buying Nvidia AI chips

China’s top internet regulator has reportedly told the country’s largest platforms to stop buying Nvidia’s AI chips — and to cancel existing orders — in the clearest sign yet that Beijing wants its tech champions running on domestic silicon. The Cyberspace Administration of China (CAC) delivered the instruction to firms including Alibaba and ByteDance, according to reporting from the Financial Times and Reuters. The order specifically references Nvidia’s RTX Pro 6000D, a China-tailored part that followed earlier, restricted-spec models like H20 designed to comply with U.S. export rules. Nvidia, Alibaba, and ByteDance declined to comment at time of writing. Nvidia shares dipped on the headlines.

On paper, this is just one directive targeting one chip. In practice, it’s a policy pivot: Beijing appears confident that its domestic accelerators are now “good enough” to shoulder more production AI work, at least for near-term needs. That tracks with recent reporting that Huawei is ramping shipments of its Ascend 910C parts — successors to the 910B — and positioning them as H20 alternatives. Sources told Reuters in April that 910C doubles compute and memory versus 910B, even if Huawei wouldn’t confirm specifics. Independent policy research also notes that while Ascend’s raw specs can be competitive with older Nvidia parts (A100-class), software ecosystem gaps and developer tooling still blunt real-world throughput.

Why now? Three forces converged. First, U.S. controls keep tightening the definition of what Nvidia can sell into China; Nvidia has answered with increasingly cut-down SKUs (H20, then RTX Pro 6000D), eroding the value proposition for Chinese buyers. Second, supply security: big platforms in China can’t plan AI roadmaps around waivers and whitelist changes. Third, industrial policy: pushing national champions like Huawei (Ascend) and Baidu (Kunlun) dovetails with Beijing’s broader self-reliance agenda. RAND analysis last month argued some domestic parts may already beat H20 on energy efficiency, even if memory bandwidth and software still lag Nvidia.

For Nvidia, the immediate risk isn’t just one cancelled order cycle — it’s channel substitution. If Alibaba and ByteDance push forward with internal accelerators and local supply chains, those engineering investments won’t reverse overnight even if geopolitics thaw. Nvidia had reportedly prepped a cheaper Blackwell-derived China SKU to keep a foothold in the market — but a regulatory directive for platforms to avoid Nvidia undercuts that plan and narrows the TAM for any “compliant” variants.

The counter-argument is that software lock-in still favors Nvidia. CUDA-centric frameworks and mature libraries translate to faster time-to-train and better utilisation, especially for frontier-scale models. That’s true — and it’s why Chinese platforms have continued to mix Nvidia with domestic parts — but it’s also where the policy hammer matters most. If regulators are prepared to take short-term efficiency hits to achieve supply security and domestic capability, we’ll see forced migration to local stacks, with vendors racing to close the software gap. Analysis from MERICS in July underscored that gap explicitly: Ascend hardware can look strong on benchmarks but is “held back” by software maturity in many applications.

What to watch next:

• Procurement pivots in earnings and supplier disclosures: watch Alibaba Cloud and ByteDance’s infrastructure notes for formal shifts to Ascend/Kunlun.

• Policy follow-through: does CAC’s guidance get echoed by other ministries, and does it expand beyond the RTX Pro 6000D to cover any future Nvidia SKUs?

• Ecosystem velocity: indicators that Ascend’s toolchain and frameworks (e.g., CANN, MindSpore) are closing the productivity gap with CUDA — the real determinant of TCO over 3–5 years.

Bottom line: if this directive holds, Nvidia’s China workaround era is ending. The upside for Beijing is clear — tighter supply sovereignty. The cost is performance drag and developer friction in the near term. For Nvidia, the China strategy now shifts from “comply and sell a bit” to “wait and watch” — while it doubles down on the rest of the world’s insatiable demand.