Beijing to pursue AI self-reliance at all costs

Beijing's not kidding around when it comes to AI self-reliance.

On April 29, Executive Vice Premier Ding Xuexiang laid out China's AI strategy at the Digital China Summit in Fuzhou.

The central message: China will build its own AI stack at all costs.

Ding was blunt (Gov.cn):

• "China's AI must take the path of independent innovation to have the confidence to withstand any containment or suppression."
• "[We] should pursue full-chain breakthroughs in core technologies."

Ding's remarks help explain the lack of H200 imports.

• In March, Nvidia restarted H200 production after receiving orders from Chinese customers, and Tencent signaled it would buy.
• Yet, according to US Commerce Secretary Howard Lutnick, not a single H200 has been sold to China (Reuters).
• The Global Times shot back that Nvidia's terms, which require upfront, non-refundable payments, are unfair given the unpredictability of US policy (Global Times).

Our take: Refusing to allow H200s into China is a huge mistake – and amounts to protecting the chip industry at the direct expense of AI model companies.

• Although Chinese models have stayed close to the US frontier, Chinese AI labs are pulling in roughly 100x less revenue than their US rivals.
• Limited compute is capping their growth, and the industry has been warning for months that it does not even have enough chips to serve customers.

The bottom line: China can't build a domestic AI chip industry by starving its only customers.