Sources: Baidu ordered AI chips from Huawei in 2023 to shift away from Nvidia; a source says Baidu ordered 1,600 Ascend 910B chips, worth ~$62M, for 200 servers
Context & Ripple Effects
Baidu's Huawei purchase sits against a 2023 procurement cycle in which major Chinese internet companies had also placed large Nvidia chip orders amid concern over tighter US export limits. The order shows Baidu was building an alternative supply path rather than relying on a single accelerator vendor.
The move foreshadowed a broader shift: subsequent coverage reported that Chinese cloud and internet groups were moving some demand toward Huawei chips, while Nvidia introduced China-specific H20 products. Later reports of Alibaba and ByteDance testing Huawei's newer chip for improved CUDA compatibility suggest this early purchase helped establish a customer base for a more usable domestic platform.
First-order effects
- Baidu gains 200 servers based on Huawei's Ascend 910B, immediately diversifying the hardware available for its AI workloads and reducing its dependence on Nvidia supply.
- Huawei gains a sizable named-customer deployment for Ascend, providing commercial validation beyond its own product ecosystem.
Second-order effects
- Baidu's order adds weight to the later shift by major Chinese platforms toward Huawei chips, increasing pressure on Nvidia to retain demand with products permitted in China, including its China-specific H20 offering.
- A larger installed base of Ascend systems gives Chinese AI operators and software teams a reason to invest in adapting workloads across more than one accelerator stack.
Third-order effects
- If similar purchases persist, China’s AI infrastructure market could become structurally more multi-vendor, with domestic accelerators serving as a second source rather than merely a contingency option.
- The longer-term competitive question shifts from chip availability alone to software portability and ecosystem compatibility, a direction reflected in later tests of Huawei's newer chip for CUDA compatibility.
The trend: Export-constrained AI buyers are turning second-source accelerator procurement into a broader push for heterogeneous, domestically supported AI compute stacks.