Nvidia's AI chip shipments into China fell by roughly half, and for the first time domestic non-GPU accelerators outsold GPU-based products in the Chinese market, according to a BigGo Finance report on the reshaping of the country's AI silicon sector.

That is a striking reversal for a market Nvidia once treated as a reliable multi-billion-dollar revenue pool. It is also a marker the industry will read carefully: the share that GPUs have held almost unchallenged in AI training and inference has, in China at least, now been ceded to a different class of hardware.

What changed in the ranking?

The report frames the shift as a first. Domestic non-GPU accelerators — application-specific AI chips rather than graphics-derived processors — have overtaken GPUs in China's AI chip market for the first time. Nvidia's shipment volume, meanwhile, halved.

The two facts are connected. Washington's export controls have progressively narrowed what Nvidia may legally sell into China, pushing the company through successive downgraded product tiers for that market. Each restriction round has widened the opening for local vendors, whose accelerators now sit on the winning side of the ledger.

Readers should note what the report does and does not quantify. It gives the direction and the rough magnitude — Nvidia shipments down by half, non-GPU accelerators ahead for the first time — but does not publish absolute unit volumes, revenue splits by vendor, or a named domestic champion. Treat the ranking claim as a reported market-share shift rather than an audited shipment table.

Why non-GPU designs gained ground

The logic is straightforward. GPU architectures from US vendors face licensing walls; non-GPU accelerators from Chinese designers do not. When the restricted product is the imported one, buyers reroute.

China's largest AI buyers — cloud providers and model developers facing their own compute procurement pressures — have spent two years qualifying domestic alternatives. A first-place finish for non-GPU accelerators suggests that qualification effort has matured from pilot projects into volume purchasing.

What does this mean for Nvidia?

The halving of shipments quantifies what earlier signals implied. Nvidia has repeatedly warned that export controls would cost it meaningful China revenue. A fifty percent shipment decline converts that warning into a number.

The company still sells compliant products into the market, and demand for AI compute in China has not shrunk — it has been redirected. Nvidia's problem is not that Chinese customers stopped buying accelerators. It is that they increasingly buy someone else's.

How firm is the trend?

One reporting period is a data point, not a trend. Non-GPU accelerators holding the lead depends on domestic vendors sustaining supply, software support and yields at scale — historically the hard part for any challenger to GPU-based ecosystems.

Still, a first overtaking event is how displacement begins. If domestic non-GPU accelerators hold their lead through subsequent quarters, Nvidia's addressable share of the world's second-largest AI market will settle well below its historical position, and Chinese chip designers will have their first defensible proof point that the GPU's dominance in AI compute is not a law of nature.