Tencent Warns US Offshore Chip Curbs Will Slow China AI Training
Tencent President Martin Lau warned that Washington's licensing requirement for advanced chips shipped to foreign data centers serving Chinese firms will slow model training across China and may delay releases, ultimately pushing every Chinese developer onto domestic silicon. CSO James Mitchell said inference, the bulk of Tencent's cloud work, already runs on domestic chips. The warning came as Tencent's Q3 capex surged more than 6x to $12.2B, or 39% of revenue, holding net profit growth to 0.8%.
Full Report
Rohan Paul (@rohanpaul_ai) · 2026-11-19 06
Tencent President Martin Lau said Washington’s block on offshore Nvidia access will slow model training across China.
The US Commerce Department now requires licenses for advanced chips headed to any foreign data center that could serve Chinese firms.
Many Chinese labs trained flagship models on Blackwell chips that can’t ship to China but could be rented abroad.
Lau said training will take the hit, possibly delaying model releases, and tighter curbs will eventually push every Chinese developer onto domestic chips.
Chief Strategy Officer James Mitchell said inference, most of Tencent’s cloud work, can already run on domestic chips the rule doesn’t touch.
The warning came as Tencent’s third-quarter capital spending jumped more than 6X to $12.2B, or 39% of revenue. That spending helped hold net profit growth to 0.8% as revenue rose 8%, missing forecasts.
Mitchell said spending sits at peak levels that could last through 2027.
