Nvidia is developing a new AI chip for China based on its Blackwell architecture, according to sources briefed on the matter.

The chip, tentatively called the B30A, will use a single-die design with NVLink, and is expected to be more powerful than the H20 Nvidia currently sells in China, but may offer about half the computing power of the dual-die B300 accelerator, the sources said.

Nvidia hopes to send B30A samples to Chinese clients for testing as early as next month, though the chip’s specifications have not been finalized.

Approval from the US regulators remains uncertain amid ongoing concerns in Washington about advanced AI technology exports to China.

🔗 Source: Reuters

🧠 Food for thought

1️⃣ Single-die design reflects strategic trade-offs in chip manufacturing

Nvidia’s choice to use single-die architecture for the B30A chip represents a calculated engineering compromise between performance and manufacturing feasibility.

Multi-die designs allow companies to partition large designs into smaller dies, which enhances manufacturing yield and provides more flexibility in mixing different process nodes1. However, they also introduce challenges in ensuring efficient power delivery and maintaining low-latency connections between dies1.

By opting for single-die construction that delivers roughly half the computing power of their flagship dual-die B300, Nvidia appears to be prioritizing manufacturing reliability and cost control over raw performance.

This approach aligns well with export-restricted markets where regulatory compliance requirements may limit the performance ceiling, allowing Nvidia to focus on production efficiency rather than maximum capability.

2️⃣ Ecosystem lock-in creates substantial switching barriers despite competitive pressure

Nvidia’s strategy of maintaining Chinese developer engagement through its software ecosystem faces real-world validation as companies struggle with platform transitions.

Chinese state-funded AI data centers are mandated to use at least 50% domestic chips, but the switch from Nvidia’s CUDA platform to Huawei’s CANN platform creates significant compatibility challenges for companies adapting existing AI models2.

This software dependency explains why Nvidia argues for retaining Chinese market access—even with restricted chip performance—to prevent developers from fully migrating to rival platforms.

The switching costs extend beyond hardware replacement to include retraining development teams, rewriting existing applications, and potentially accepting performance trade-offs during the transition period.

Recent Nvidia developments