On its Q2 2026 earnings call, Baidu said Chinese customers prefer homegrown AI accelerator chips for 'supply chain' reasons, framing the shift as buyer led rather than policy driven.
Baidu told investors on its 2026-08-18 Q2 2026 earnings call that Chinese customers are now buying domestically made AI accelerator chips because of supply chain issues, according to the earnings call transcript. The framing turns decoupling from a top-down policy outcome into a buyer's purchasing decision.
AI accelerator silicon, the GPUs and custom chips used to train and run AI models, has been the choke point of US export controls on China, and Chinese demand is now being routed toward a widening set of domestic vendors. Baidu's wording on the call, captured in the earnings call transcript, is a basket: it bundles US restrictions on advanced accelerators, customer-side inventory and lead-time risk, and pressure from Chinese buyers to source domestically.
The shift is real but directional, not a finished break. Baidu still uses foreign accelerators for parts of its stack, and "supply chain" can be read as code for any of the three causes above. The named beneficiary inside Baidu's own walls is Kunlunxin, the Baidu chip unit that filed for a Hong Kong IPO in January 2026 and is positioned to sell into the same domestic demand Baidu is now describing to investors.
Watch the IPO trajectory and any call disclosure of which workloads still run on foreign silicon.