Activity in China's AI accelerator business is picking up again: Nvidia has ordered a halt to production of its H20 chip., the model designed to comply with U.S. export restrictions. The company is reorienting its roadmap in the region and focusing on its next product family.
The decision comes after growing regulatory pressures and doubts about adoption of H20 among major Chinese customers. Nvidia, however, insists that it manages its supply chain dynamically to adapt to market conditions and emphasizes that its products do not include hidden access points.
Reasons for the pause and orders to suppliers
According to various industry sources, Nvidia has informed key partners, including Samsung Electronics and Amkor Technology, to stop work related to H20. In the case of Amkor, which is in charge of advanced packaging, there is even talk of an increase in work-in-process inventory awaiting final instructions.
The company declined to provide additional details, simply stressing that “constantly manages its supply chain”. In parallel, an immediate reaction was observed in the markets: Nvidia shares fell by after-hours operations after the orders were announced, reflecting the uncertainty about future demand for this accelerator.
Warnings in China and debate on security
In the last weeks, Chinese authorities recommended large local technology companies to avoid H20 or justify its purchase, citing national security concerns. At the government level, the instruction would have been especially strict, while encouraging the use of domestically manufactured semiconductors.
Among the concerns aired was the possibility of remote access or disabling functions. Nvidia flatly denied the existence of any backdoor, recalling that such a mechanism would also put critical infrastructure at risk in other countries. Separately, it has been documented that some Shipments incorporated tracking devices in the packaging, although this would not be a widespread practice.
Roadmap Recalibration: Blackwell B30A
While H20 slows down, Nvidia accelerates the relay with Blackwell B30A, a variant adapted to regulatory limits based on the B300 Blackwell Ultra. The key difference is in the single-die encapsulation (vs. the B300's twin-die), a tweak that cuts performance roughly in half to better fit export regulations.
This approach preserves the most valuable elements of the home's ecosystem: CUDA support and optimized librariesFor engineering teams in China, maintaining tools and workflows minimizes migration costs and allows for rapid iteration despite the power reduction from Hopper.
Impact on the market and local competition
The withdrawal of H20 opens a gap that Huawei and other Chinese suppliers They try to cover it with their own proposals. In practice, two paths could coexist: Nvidia's mature software stack as a de facto standard, and the national alternative with better cost and technological sovereignty in an increasingly sensitive environment.
Washington has recently allowed the resumption of sales of lower-end accelerators to China under strict conditions, including a 15% commission on revenue in certain cases, while Beijing increases scrutiny of H20O purchases. On the stock market, the new context has boosted to some local semiconductor values, as customers evaluate risks, prices and availability.
What changes for customers and supply
For buyers who already had the H20, the pause generates uncertainty about schedules and deploymentsNvidia is also carrying unsold inventory and has previously recorded a large accounting charge related to chips destined for China, so inventory rationalization is now a priority.
Management maintains that it continues in dialogue with Washington on the future of its specific offer for China and recalls that, as both governments recognize, the H20 was not conceived for military use or government infrastructureThe immediate key will be to ensure a stable supply of the Blackwell family that meets standards without compromising on the software and support that data centers require.
In a scenario marked by geopolitical tensions and export limits, Nvidia pivots from H20 to B30A with the intention of maintaining a presence in a key market while attempting to clear regulatory doubts. For Chinese AI operators, the decision forces them to recalibrate their roadmaps: continue with Nvidia's platform for its software and tools, opt for domestic hardware, or combine both paths to ensure the computing power required by the next generation of models.
