Nvidia Prepares Blackwell-Based B30A Chip for China
Nvidia is designing the B30A, an AI chip based on the advanced Blackwell architecture, specifically for the Chinese market.
It Offer a more capable chip than H20 while staying within U.S. export limits.
Key Features
Cutting-edge Build: Utilizes a single-die Blackwell design less complex than the dual-chiplet B300 to deliver strong performance while meeting compliance requirements.
High-Bandwidth Memory & Connectivity: Equipped with HBM3E memory (~144 GB)—a step up from the H20’s 96 GB—plus NVLink for fast data interconnects.
Performance Sweet Spot: Significantly more powerful than the H20, but designed to deliver around 50% of the performance of the flagship B300.
Complementary Product: Nvidia is also prepping the RTX6000D a Blackwell-based inference chip for China, designed with reduced specifications to fit export compliance.
It’s expected to start sampling in September.
Benchmark Performance
The B30A outpaces Nvidia’s current China-approved H20 chip.
Delivers roughly half the compute power of B300, balancing performance and regulatory acceptance.
Access & Availability
Nvidia aims to ship samples to Chinese clients as early as next month, contingent on U.S. government approval.
Wider availability may extend into late 2025 or early 2026.
Strategic Goal
Nvidia’s strategic goal with the B30A chip is to preserve its access to the massive Chinese AI market, even as U.S. export restrictions and geopolitical tensions intensify.
By creating a China-specific version, Nvidia ensures compliance with regulations while maintaining business continuity, protecting its market presence, and meeting rising AI demand without violating international trade policies.
News Gist
Nvidia is developing the Blackwell-based B30A chip tailored for China to comply with U.S. export restrictions.
The AI chip balances high performance with regulatory limits, aiming to preserve Nvidia’s market presence amid rising geopolitical tensions and increasing demand for AI solutions.
