In just three months, NVIDIA's H20 AI chip went from being a highly anticipated solution in China to a discontinued product. Designed to navigate U.S. export controls, the H20 ultimately could not overcome regulatory restrictions and security concerns, marking a dramatic exit from one of the world’s largest AI chip markets.

1. The Collapse of H20 in China

In April 2025, the U.S. imposed stricter export controls on AI chips, effectively blocking H20’s access to international markets. NVIDIA’s quarterly report reflected the immediate impact: $4.5 billion in inventory losses and $8 billion in potential revenue evaporated, while China’s $17 billion annual revenue base began to shake.

Even when the U.S. eased restrictions in July, a 15% revenue-sharing requirement made H20 a low-profit compliance product.

The decisive blow came from security scrutiny. The Cyberspace Administration of China uncovered hidden features in H20 firmware, including automatic connections to overseas data centers and remote control interfaces capable of disabling computing clusters. These features, aligned with U.S. monitoring requirements, raised national security concerns domestically.

Domestic enterprises, especially those with state-owned backgrounds, paused purchases. With local alternatives already available, H20 lost its market foothold.

2. The Geopolitical and Technical Implications

H20’s sudden discontinuation reflects the global struggle for computing sovereignty. U.S. policies sought to enforce performance restrictions and embedded monitoring, while the 15% revenue-sharing clause highlighted attempts to leverage technological dominance into economic gains.

However, these measures accelerated the rise of domestic AI chips in China. Today, Chinese chip makers match or surpass H20 in key metrics such as computing power, energy efficiency, and supply chain security.

With security becoming the top procurement priority, foreign brands can no longer rely solely on price-performance advantages. NVIDIA’s goal of $50 billion annual revenue in China now faces serious challenges as domestic alternatives gain momentum.

H20’s case also exposed a trust crisis. Evidence of firmware backdoors undermined NVIDIA’s claims, signaling the end of the “inspection-free era” for foreign chips. In the digital economy, chips are critical infrastructure, and products with potential remote access are increasingly rejected.

3. B30A: Upgrade or Rebranding?

Following H20’s exit, NVIDIA introduced the B30A chip, based on the Blackwell architecture. It offers 1.3× the computing power of H20, high-bandwidth memory, and NVLink support.

Yet performance restrictions remain: 50% of capabilities are capped, and the 1.2 TB/s bandwidth control appears designed to comply with U.S. export rules rather than fully exploit the chip’s potential.

NVIDIA’s move is driven by China accounting for 13% of global revenue and the allure of a $50 billion market opportunity. However, to regain trust, performance gains alone are insufficient. The Chinese market now demands transparent, secure, and backdoor-free solutions. With domestic alternatives offering equal or superior performance, foreign brands’ technology dominance is being challenged.

4. Conclusion

The discontinuation of H20 signals more than a product exit—it marks a reshaping of global AI computing power. As China accelerates domestic chip innovation, attempts to profit from technological monopoly and covert control are increasingly futile. The competition for computing sovereignty will define the rise of Chinese AI chips, representing a pivotal shift in the global tech landscape.