Nvidia’s New AI Chip Faces Huawei’s Rising Dominance in China

Nvidia is gearing up to compete with Huawei to retain its position in China’s thriving AI chip market.

Nvidia’s forthcoming AI chip for China is a bold strategic move, marking its third effort to comply with regulations while countering fierce domestic competition.

Despite ongoing geopolitical challenges from successive U.S. administrations, Nvidia remains committed to the Chinese market. The company plans to launch a scaled-down Blackwell-based processor tailored to meet export restrictions and rival advanced local competitors, particularly Huawei.

A strategy driven by necessity

Reuters sources indicate Nvidia’s new chip will be priced between $6,500 and $8,000, a significant drop from the $10,000-$12,000 range of the now-restricted H20 model. Production is set to begin in June, highlighting the urgency of Nvidia’s China AI chip strategy.

This price cut involves notable compromises. The chip will utilize Nvidia’s RTX Pro 6000D architecture with standard GDDR7 memory, moving away from the high-bandwidth memory used in premium models.

Notably, the chip bypasses Taiwan Semiconductor’s advanced CoWoS packaging, a choice that reduces both performance and production complexity.

After the April ban on H20 sales, Nvidia faced a $5.5 billion write-down on its Chinese inventory and commitments. Initial attempts to adapt the H20 for continued sales in China proved unviable under current U.S. export regulations.

Growing domestic rivalry

Huawei’s rise as a formidable competitor has reshaped China’s AI chip market. Its Ascend 910C and 910B processors have gained traction with major Chinese tech firms like Tencent, Baidu, and ByteDance, particularly for inference tasks, where they deliver strong performance.

The competition extends beyond chips to full infrastructure solutions. Huawei’s CloudMatrix 384 rack system directly challenges Nvidia’s Blackwell GB200 NVL72 setup, showcasing Huawei’s aim to dominate the entire AI hardware ecosystem.

Market trends show a shift in dynamics. Reports suggest H20 chips have been sold at discounts exceeding 10% compared to Huawei’s Ascend 910B, underscoring Nvidia’s struggle to maintain pricing power against local rivals.

CEO Jensen Huang admitted the decline, noting that Nvidia’s market share in China has roughly halved due to recent export restrictions.

High financial stakes

The financial impact is significant. Huang estimates China’s AI chip market could reach $50 billion, with Nvidia recording over $17 billion in Chinese sales in 2024. These figures explain Nvidia’s persistent focus on developing AI chips for China despite regulatory hurdles.

Sources indicate Nvidia is not relying solely on this chip. The company is reportedly planning a second Blackwell-based variant for China, with production slated for September. This multi-faceted strategy reflects Nvidia’s determination to maintain a foothold in China through varied products tailored to different customer needs and regulatory constraints.

A challenging road ahead

Nvidia’s latest China AI chip venture raises critical strategic questions. Can intentionally reduced-performance hardware compete with rapidly advancing local alternatives? Will Chinese customers accept these compromises as domestic options continue to improve?

Huang’s recent statement that “China is catching up fast. We’re neck and neck” in AI capabilities suggests the competitive gap is closing rapidly. Coupled with strong government support for China’s semiconductor industry, the market poses serious questions about Nvidia’s long-term prospects in China.

See also: Can the U.S. effectively enforce a global AI chip restriction?

Want to explore more about AI and big data from industry experts? Visit the AI & Big Data Expo in Amsterdam, California, and London. This comprehensive event is co-located with leading conferences like the Intelligent Automation Conference, BlockX, Digital Transformation Week, and Cyber Security & Cloud Expo.

Discover other upcoming enterprise tech events and webinars hosted by TechForge here.