Amid fierce competition and new export rules, Nvidia is adjusting its strategy for China with a roadmap of AI accelerators tailored to the market and regulatory constraints. The goal is to maintain a presence in an increasingly complex environment, with Beijing promoting local suppliers and Washington controlling chip capacity.
The company is working on Blackwell-based solutions that aim to balance performance and regulatory compliance, while resuming H20 supplies. All this is happening while Chinese regulator scrutinizes security risks of these products and doubts about their use in local data centers are increasing.
What Nvidia is preparing for China: B30A and RTX6000D
According to information reported by Reuters, Nvidia is finalizing a new accelerator known internally as B30AThis processor would be based on the Blackwell architecture with a single-die design and would offer approximately half the power of the B300, the flagship of the company. Even so, it would aim to surpass the H20 (Hopper architecture), designed specifically for the Chinese market.
The B30A would integrate high-bandwidth memory HBM and high-speed link NV Link to accelerate data traffic between GPUs, key capabilities in training loads and AI inference. Nvidia plans send test samples to Chinese customers in the coming weeks, subject to the maintenance of export permits.
In parallel, the firm is preparing an option focused on inference, the RTX6000D, also under Blackwell, designed to meet Washington's thresholds. This model would use GDDR memory and set its bandwidth to 1.398 GB / s, just below the limit that triggered previous restrictions, with Pilot batches planned from September for customers in China.
Nvidia has stated in statements that it is evaluating multiple product configurations to compete if governments allow it, and that their shipments comply with current authorizations for commercial use.
Export limits and the role of H20
Following the 2023 bans, Nvidia developed the H20 as a specific alternative for China, sacrificing some performance to fit into US export rules. In July, Washington authorized the resumption of sales of H20 in the Asian country, which has allowed orders to be reactivated while the new models arrive.
On the political front, and according to public statements by US authorities, Nvidia would agree to transfer 15% of revenues from sales of AI chips to China as part of a conditional exemption; AMD would have been included in that scheme. The sector sees this formula as a way to maintain a minimum flow of hardware into the Chinese market without breaching the limits imposed by national security.
Washington's current strategy is to allow "controlled" chips that do not reach the levels of its most advanced models (such as the B300), with the intention of not accelerating Beijing's technological self-sufficiency and, at the same time, preserving a certain degree of influence over supply.
Supervision in China and growing distrust
The future of these products in the Chinese market depends, to a large extent, on the verdict of the Cyberspace Administration of China (CAC), which evaluates H20 for potential security risks and possible hidden mechanisms. The scrutiny occurs in parallel with information on state media who rate H20 as unsafe for local deployments.
Nvidia, for its part, has defended that Their chips do not incorporate back doors or deactivation mechanisms., in a clear attempt to regain the trust of customers and regulators. The CAC's decision, if unfavorable, could significantly undermine shipments in China, a market the company itself considers critical.
Aside from the ruling, Chinese authorities are pushing major AI model developers to prioritize integrated circuits of national originThis directive, coupled with the narrative about security risks, complicates H20's commercial appeal as an intermediate solution.
Local alternatives: Huawei and other suppliers
Huawei has consolidated its line of accelerators Ascend in recent years with the aim of covering both training and inference. The company works in the Ascend 910D to compete with H100 and, in parallel, in the Ascend 920, which aims to occupy H20's space in the Chinese market. The latter is expected to enter volume production in the second half of 2025, using 6nm technology alongside local partners.
Along with Huawei, actors such as Moore Threads They have unveiled AI-focused GPUs that, on paper, compete in specific areas with solutions from Nvidia and AMD. Although the maturity of the software ecosystem and high-bandwidth memory remain challenges, the Chinese government accelerates its autonomy in semiconductors to reduce dependence on foreign hardware.
Impact for Nvidia and the market
China accounted for around 13% of annual income from Nvidia in the last fiscal year, so losing traction there would be a significant blow. If the CAC limits H20 and the transition to Blackwell Content (B30A/RTX6000D) doesn't work out, the company could see a deterioration in sales just as global demand for AI continues to soar.
To try to avoid this, Nvidia is fitting specifications at the edge of the regulatory framework, advancing samples and scaling up small batches. At the same time, it reinforces its message of safety and compliance, aware that the perception of risk in China weighs almost as much as the raw performance figures.
This pulse, with United States imposing thresholds And China, demanding trust and technological sovereignty, is reshaping the AI GPU market: "Chips for China" are now a category of their own, with customized versions and delivery schedules conditioned by geopolitics.
The evolution in the coming months will depend on three vectors: the CAC's opinion on H20, the Regulatory approval of the B30A/RTX6000D by the US and the ability of local champions to scale their hardware with competitive software and memory.
This scenario keeps Nvidia trying to retain the Chinese market with tailor-made offers and regulatory concessions, while Beijing pushes alternatives and scrutinizes every foreign chip. If Blackwell's new proposals for China pass muster, they could sustain some of the demand; if not, the space will quickly be filled by local solutions already knocking on the door.