US export controls are boosting Chinese AI chip sales

Three of China’s leading AI chip makers, Biren Technology, Cambricon Technologies, and Hygon Information Technology, have reported record quarterly revenue, challenging the standard story around US export controls.

Biren said Q1 revenue reached RMB 640 million, up 60% year over year. Cambricon reported 52% top-line growth. Hygon said AI chip sales to enterprise customers nearly doubled.

Those numbers point to something bigger than a simple rise in AI demand. Since tighter US restrictions hit Nvidia and AMD sales into China, domestic suppliers have gained a stronger position with major buyers who still need accelerators for cloud and model workloads. The result looks increasingly like accidental industrial policy: export controls aimed at slowing China’s AI stack are also helping Chinese chip companies secure demand at home.

Protected demand, not technical parity, is driving China’s AI chip gains

The key shift is commercial, not technological.

China’s domestic AI chip leaders still trail Nvidia at the high end, especially on cutting-edge training performance.

Yet they no longer need full parity to win business. They need enough capability to serve customers who face shrinking access to US chips, and in many cases the priority is easier AI deployment, not just raw speed.

That dynamic is showing up in procurement. Tencent, Alibaba Cloud, and Baidu have publicly moved toward greater use of domestic chips, and Chinese cloud providers have been signing longer-term deals with local vendors. Biren said it won its largest-ever cloud cluster deployment from a top-three hyperscaler, a contract that, in an earlier market, likely would have gone to Nvidia. Cambricon and Hygon have also pointed to growing design-ins for AI workloads, including training and inference tasks that previously depended on US GPUs. That shift also fits a broader pattern where the real challenge is not merely buying silicon, but turning chips into working capacity.

Policy support is reinforcing the shift. Chinese authorities have given domestic vendors a clearer path into public-sector and strategically sensitive computing projects through procurement guidance, financing support, and local-preference frameworks. That matters because export controls bite most sharply at the point of access. Advanced foreign chips have become harder to buy, more expensive to source, and riskier to deploy through unofficial channels. For many Chinese customers, the practical choice is increasingly domestic, especially as companies like Alibaba launch new AI chips to meet rising demand.

This creates a powerful feedback loop. Protected demand gives local chip makers more volume. More volume supports more revenue. More revenue can fund product development, software work, and customer support that would have been harder to finance in a fully open market. Biren and Hygon have each discussed next-generation chip plans, while Cambricon is working with China’s domestic manufacturing ecosystem, including SMIC. None of that proves China has closed the performance gap. It does show that sanctions can help create the market conditions local firms need to scale, even as rivals abroad prepare to invest tens of billions more to lead in AI chips.

The Chinese AI chip market is being reshaped less by outright leapfrogging than by a home-market advantage enforced by geopolitics. In practical terms, that advantage already matters for cloud deployments, public contracts, and capital allocation. A market once dominated by superior foreign products is becoming one where domestic supply is favored, financed, and embedded.

What to watch next

Three signals will show whether this is a lasting realignment or a temporary boost.

First, watch product performance and manufacturing consistency.

If Biren’s BR100 or Hygon’s S3000 series starts to narrow the real-world gap with Nvidia’s A100 or H100 in large Chinese deployments, the case for durable domestic substitution gets much stronger. Manufacturing yields matter too. If Chinese fabs can sustain reliable 7nm-class or better production, local supply becomes easier to scale.

Second, monitor procurement rules and policy direction.

If local-content preferences expand into new cloud, model development, or secure-computing projects, domestic chip demand could become even more entrenched. If major buyers begin pressing for exceptions or showing frustration with local alternatives, that would suggest the current momentum has limits.

Third, follow the revenue and deployment data.

If Biren, Cambricon, and Hygon keep posting strong growth while US chip sales into China remain constrained, that would support the idea that a parallel Chinese AI compute market is taking shape. If growth slows, inventories rise, or major hyperscalers stall deployments, the market may be hitting performance or operational ceilings.

Export controls designed to slow China’s AI hardware progress are also helping build a protected customer base for Chinese chip makers. Whether that turns into a durable competitive ecosystem will depend on execution, yields, and software maturity. But the first commercial proof points are already here, and they show how quickly geopolitics can redraw a technology market.