NVIDIA B30A AI Chip for China: Blackwell, H20 and Export Controls
NVIDIA is reportedly developing a new Blackwell-based AI accelerator for the Chinese market, known as the B30A, with performance expected to significantly exceed the H20 while remaining within U.S. export restrictions.
The reported B30A is part of NVIDIA’s broader effort to maintain access to China’s rapidly growing AI market as export controls reshape the company’s product strategy. At the same time, NVIDIA faces increasing competition from China’s domestic semiconductor industry.
🚀 NVIDIA’s B30A Targets China’s AI Market #
According to industry reports, NVIDIA is preparing the B30A AI accelerator specifically for China. Unlike NVIDIA’s highest-end Blackwell products, the B30A is expected to be configured around export-compliant performance limits.
The reported chip could deliver approximately 50% of the performance of NVIDIA’s flagship B300, putting it substantially ahead of the H20 currently available to Chinese customers.
The B30A is reportedly expected to include:
- High-bandwidth memory (HBM) for high-throughput AI workloads
- NVLink for high-speed communication between processors
- A Blackwell-based architecture optimized around U.S. export requirements
- A single-die configuration, according to industry reports
NVIDIA could begin providing B30A test samples to selected Chinese customers as early as September, although specifications and availability may still change before commercial launch.
⚖️ U.S. Export Controls Reshape NVIDIA’s China Strategy #
NVIDIA’s China product roadmap has been heavily influenced by evolving U.S. semiconductor export restrictions.
Earlier reports indicated that NVIDIA and AMD reached an arrangement involving a 15% share of certain China-related chip sales revenue in exchange for export licenses. The reported arrangement added another layer of complexity to NVIDIA’s efforts to sell advanced AI accelerators in China.
The H20 has also faced scrutiny inside China. Chinese state media and other commentators have raised concerns about potential security risks associated with NVIDIA’s AI products, creating an additional challenge for the company beyond U.S. regulations.
NVIDIA has therefore been developing multiple China-specific products rather than relying on a single accelerator.
One example is the reported RTX 6000D, which is aimed primarily at AI inference. It is expected to use conventional GDDR memory rather than HBM, with reported memory bandwidth of approximately 1,398 GB/s. Its lower specifications could allow NVIDIA to offer it at a more competitive price.
Limited RTX 6000D shipments were reportedly expected in September.
🔄 Why NVIDIA Is Bringing the H20 Back to China #
The H20 became a central part of NVIDIA’s China strategy after tighter U.S. export restrictions affected the company’s ability to sell more advanced AI accelerators in the country.
Under the reported licensing arrangement, NVIDIA could resume H20 sales after agreeing to pay a portion of China-related revenue to the U.S. government. Former U.S. President Donald Trump described the H20 as an older-generation product, a characterization that helped explain why its export could receive approval under revised conditions.
Analysts have also suggested that NVIDIA could increase H20 pricing to compensate for the additional licensing cost while protecting margins.
However, regulatory approval does not automatically guarantee strong demand. Chinese companies with particularly strict security requirements may remain reluctant to deploy foreign AI accelerators, especially when domestic alternatives are becoming increasingly capable.
🇨🇳 China’s Domestic Semiconductor Industry Is Accelerating #
NVIDIA’s challenges in China extend beyond export controls. The country’s semiconductor industry is rapidly expanding across AI computing, servers, PCs, automotive electronics and data-center infrastructure.
The automotive sector is particularly important. Chinese automakers are increasingly developing or adopting locally designed processors to reduce dependence on foreign semiconductor suppliers.
Examples reported in the market include:
- Xpeng developing its own Turing automotive processor
- NIO using its Shenji chip
- Chinese automakers working with domestic suppliers including Horizon Robotics, Huawei HiSilicon and Black Sesame
Foundry capacity is also expanding. SMIC has increased its presence in automotive semiconductors, with its reported market share rising from below 3% in 2020 to approximately 10% in 2025.
Research cited from Bank of America estimates that Chinese-designed chips accounted for about 9% of the global automotive semiconductor market in 2024, with the share potentially reaching 15%–20% in 2025.
Longer-term projections suggest that Chinese-made automotive chips could capture an even larger portion of the global market over the next five years.
For NVIDIA and other international semiconductor companies, this means that export restrictions could have effects extending well beyond short-term sales. They may also accelerate the development of alternative domestic supply chains.
🧠 B30A vs. H20: What We Know #
The reported positioning of the two accelerators is relatively straightforward.
| Feature | NVIDIA B30A | NVIDIA H20 |
|---|---|---|
| Architecture | Blackwell | Hopper |
| China market | Reportedly designed specifically for China | Previously approved China-focused accelerator |
| AI performance | Reportedly around 50% of B300 | Lower than newer Blackwell products |
| Memory | HBM reportedly planned | HBM |
| High-speed interconnect | NVLink reportedly supported | NVLink |
| Export strategy | Designed around current restrictions | Subject to changing export controls |
The most important difference is architectural generation. The B30A reportedly moves NVIDIA’s China offering from Hopper toward Blackwell, potentially providing better performance and efficiency while remaining below the capability of NVIDIA’s unrestricted flagship products.
Because the B30A was still reportedly under development, its final specifications, performance and commercial availability could differ from early industry estimates.
📈 What the B30A Means for NVIDIA #
The B30A illustrates the increasingly complex trade-offs facing NVIDIA in China.
On one side, China remains a major market for AI computing. On the other, NVIDIA must design products around U.S. export regulations while dealing with growing demand for domestic alternatives.
The company’s China strategy is therefore evolving toward a portfolio that could include:
- B30A for higher-end AI computing
- H20 as a reapproved existing-generation accelerator
- RTX 6000D for more cost-sensitive AI inference workloads
This approach could allow NVIDIA to preserve a presence in different segments while regulatory requirements continue to change.
🔍 Frequently Asked Questions #
What is NVIDIA’s B30A? #
The B30A is a reportedly upcoming NVIDIA AI accelerator designed specifically for the Chinese market. It is expected to use the Blackwell architecture and offer substantially higher performance than the H20 while remaining compliant with applicable U.S. export restrictions.
How does the B30A compare with the H20? #
Early reports suggest that the B30A could be significantly faster than the H20 and deliver roughly half the performance of NVIDIA’s flagship B300. However, final specifications have not been established in the supplied reports.
Why is NVIDIA developing a China-specific Blackwell chip? #
U.S. export controls restrict the sale of certain advanced AI accelerators to China. A dedicated China-market product allows NVIDIA to target Chinese AI workloads while configuring performance and capabilities around applicable regulatory limits.
What is the NVIDIA RTX 6000D? #
The RTX 6000D is another reported China-focused NVIDIA product aimed primarily at AI inference. It is positioned below the H20 in specifications and reportedly uses GDDR memory rather than HBM.
Are Chinese semiconductor companies competing with NVIDIA? #
Yes. Chinese companies are increasingly developing domestic processors and accelerators for AI, automotive, server and other applications. This growing domestic ecosystem could reduce reliance on NVIDIA and other foreign semiconductor suppliers over time.
🎯 Key Takeaway #
The reported NVIDIA B30A represents a significant shift in NVIDIA’s China strategy. By bringing the Blackwell architecture to a China-specific AI accelerator, NVIDIA could offer substantially higher performance than the H20 while remaining within the boundaries of U.S. export controls.
But the B30A is only one part of a larger story. NVIDIA must simultaneously navigate licensing requirements, Chinese security concerns and the rapid expansion of domestic semiconductor alternatives.
If the reported specifications materialize, the B30A could strengthen NVIDIA’s position in China’s AI market—but the long-term competitive landscape will increasingly depend on both export policy and the pace of China’s semiconductor development.