Sources: Alibaba and Baidu have begun using their own internally designed chips to train their AI models, partly replacing AI chips made by Nvidia
Trouble Ahead for Nvidia? Deborah Sophia / Reuters : Alibaba, Baidu begin using own chips to train AI models, The Information reports Anton Shilov / Tom's Hardware : Top China silicon figure calls on country to stop using Nvidia GPUs for AI — says current AI development model could become ‘lethal’ if not addressed David Cowan / Compact : Nvidia Is a National Security Risk Livemint : Top Beijing Adviser Says China Should Ditch Nvidia For Own Tech X: Amir Efrati / @amir : A spokesperson for Nvidia said: “The competition [in China] has undeniably arrived.” This is not a drill. Question is will it convince USG to loosen restrictions on China sales. [image]
Context & Ripple Effects
Alibaba and Baidu had already been among Chinese platforms shifting some AI-chip demand away from Nvidia’s lower-powered products toward Huawei alternatives in an earlier move toward China-made accelerators. The latest reported deployment takes that strategy from sourcing substitution to in-house chip use for model training.
The shift also follows a period in which China access to Nvidia hardware was constrained and Nvidia was adapting chips for sale under export rules. It matters because training is a core workload: internal deployment can turn chip design into an operating capability rather than a procurement fallback.
First-order effects
- Alibaba and Baidu can reportedly run part of their AI-training workloads on internally designed chips, reducing—but not eliminating—their dependence on Nvidia GPUs.
- Nvidia faces a more direct competitive challenge at two major Chinese AI customers, beyond the prior pressure around restricted and modified products.
Second-order effects
- The move raises the value of software, tools, and engineering needed to make mixed fleets of proprietary and third-party accelerators productive; the relevant constraint shifts from simply obtaining chips to operating scarce AI compute infrastructure efficiently.
- Other Chinese model developers have a clearer incentive to weigh proprietary chips and domestic alternatives against Nvidia offerings, particularly where product availability or policy conditions remain uncertain.
Third-order effects
- If these deployments expand, China’s AI-compute market could become more heterogeneous, with large platforms differentiating through their own silicon and software stacks rather than relying on one dominant GPU supplier.
- This is evidence of an AI hardware strategy split: export constraints and supply risk can accelerate vertical integration, though the report does not establish whether in-house chips will match Nvidia across all training workloads.
The trend: AI infrastructure is moving toward heterogeneous, vertically integrated compute stacks as large platforms seek more control over training capacity and supply exposure.