This article first appeared on GuruFocus.

China’s AI chip race is entering a new phase, and Cambricon is stepping straight into the vacuum left by Nvidia’s (NASDAQ:NVDA) forced retreat. The chip designer is preparing to lift production to roughly half a million AI accelerators in 2026, a leap that could position it as one of the biggest domestic beneficiaries of Beijing’s push to replace foreign technology. The plan includes as many as 300,000 units of its Siyuan 590 and 690 chips, built primarily on SMIC’s N+2 7-nanometer process. Investors have already begun to price in the shift: Cambricon shares rose 2.8% in Shanghai, while SMIC gained 3.9% in Hong Kong and Hua Hong Semiconductor added 3.1% as expectations grew around a deeper reallocation of China’s AI spending.

But the story gets more interesting when you layer in Cambricon’s operational momentum. The company reported a 14-fold surge in revenue in the September quarter, its market value has climbed nine-fold since 2021, and ByteDance now accounts for more than half of all orders with potential new business from Alibaba (NYSE:BABA) in the coming years. The production targets, however, depend heavily on securing enough capacity at SMIC, where yields for Cambricon’s top chips are about 20%, meaning most silicon dies remain unusable. At the same time, rivals are stepping up: Huawei is preparing to double output of its most advanced AI chips next year, and Moore Threads is debuting in Shanghai as another contender pushing into China’s accelerator ecosystem.

Even with this surge in local suppliers, demand could still outpace availability. Bloomberg Intelligence points to persistent constraints, from low 7-nm yields at SMIC to dependence on high-bandwidth memory still sourced from SK Hynix and Samsung. Yet Cambricon’s rising influence gives it bargaining power as geopolitical pressure pushes Chinese AI firms closer to homegrown designs. With Beijing’s self-reliance drive drawing an estimated $98 billion in government and corporate spending this year, companies from DeepSeek to Alibaba could increasingly rely on domestic accelerators instead of Nvidia’s offerings especially with Nvidia blocked from China and the Trump administration only considering whether H200 sales might be allowed. The result could be a market where China’s AI hardware ecosystem continues to diverge, creating the contours of a possibly bifurcated global AI landscape.