November 20, 2026

AIincider

AI News. No Noise. Just Signal.

Tencent and Alibaba Lean on Chinese Chips as US Curbs Bite

3 min read
Tencent and Alibaba say domestic chips will cushion their cloud businesses from new US export rules, even as model training slows. Read the full breakdown.

China’s two biggest cloud providers delivered the same message to investors this week: new American chip rules will hurt, but not where it counts most. On their third-quarter earnings calls, Tencent and Alibaba said domestic processors can keep their cloud businesses growing on Chinese chips, while conceding that training the next generation of models will take longer, Fortune reports.

The rule that changed the math

The US Commerce Department issued an interim rule on Monday that requires licenses to sell advanced AI chips to data centers in Malaysia, Singapore, Thailand, Indonesia and Brazil, and to any overseas facility supplying large amounts of compute to companies headquartered in China. The target is remote access: Chinese developers renting Nvidia hardware abroad rather than importing it. House Foreign Affairs Committee data cited by Fortune shows regulators had approved roughly 880,000 AI chips for Chinese companies, including 75,000 for Alibaba and 150,000 for Tencent, out of more than 3.4 million that Nvidia and AMD sought to export.

What the two companies said

Tencent posted revenue of 207.3 billion yuan ($31 billion), up 8 percent, and lifted capital spending 527 percent to 81.4 billion yuan ($12.2 billion). Chief strategy officer James Mitchell said the company does not “see an impact” on cloud because that business is mostly inference, which runs well on homegrown silicon. President Martin Lau was blunter about training, calling the restrictions “unfortunate” and warning that the next Hunyuan model will take longer to build. “It sets everyone back in China,” he said.

Alibaba reported revenue of 271.5 billion yuan ($40.5 billion), up 10 percent, with AI and cloud revenue jumping 53 percent to 61 billion yuan ($9.1 billion) and capex up 60 percent. CEO Eddie Wu said progress on domestic computing is “already ahead of our original expectations,” and that production at its T-Head chip unit, which is slated for a Hong Kong listing, should ramp in 2027 or 2028. He also cautioned that cloud growth could cool to 30 to 45 percent next year before rebounding toward 50 percent or more by 2028.

Why it matters

The split between inference and training is the whole story. Serving existing models to customers is now a Chinese-silicon business; building frontier models is not yet. Analysts at Maybank wrote this week that the constraint on China’s AI sector has shifted from access to compute to domestic chip supply, with Huawei’s 2026 output estimated below 4 percent of Nvidia’s and SMIC running above 97 percent utilization, per TechNode Global.

The fallout is spreading. Alibaba shareholders sued on Thursday over alleged ties to Megaspeed, the Singapore company under investigation for chip diversion, and the Financial Times reports TSMC and Samsung have paused some shipments to Chinese customers. The question for 2027 is how fast T-Head and Huawei can close the training gap.

Continue Reading…

Leave a Reply