DeepSeek reportedly working on an in-house AI chip to cut dependence on Nvidia and Huawei

DeepSeek, the Chinese artificial intelligence startup known for building and deploying large models, is reportedly developing its own AI chip to reduce reliance on external semiconductor suppliers including Nvidia and Huawei, according to a Reuters report cited by financial news outlets on Tuesday.

The report, attributed to multiple people familiar with the matter, characterizes the project as an effort to improve hardware independence. For AI labs, that usually means lowering procurement risk and potentially avoiding bottlenecks tied to availability, export rules, and negotiating leverage with chip vendors that dominate accelerated computing.

Several secondary write-ups describing the Reuters reporting say DeepSeek’s chip effort is aimed at “inference,” the phase of model usage where an AI system applies a trained model to produce outputs for users, rather than at training. Inference chips tend to emphasize throughput and energy efficiency for serving workloads, an area where AI developers often want more tailored performance than what comes bundled with off-the-shelf data center GPUs.

Nvidia’s position in the AI hardware stack has come under particular scrutiny from users and competitors as governments and regulators move to constrain access to advanced compute. If a frontier-model company is able to run efficiently on internally sourced chips, it can potentially reduce its dependence on the fastest and most geopolitically sensitive hardware and diversify its supply chain.

The story also underscores the growing interest among Chinese AI firms in designing specialized silicon. Huawei and other domestic players already supply chips for AI-adjacent workloads, but the reported goal suggests DeepSeek wants options beyond buying from any single vendor, including by building a chip ecosystem that matches its own model and serving patterns.

Nvidia did not respond in the cited reports, and DeepSeek did not publicly confirm details of the chip program in the material reviewed. The reporting likewise did not lay out a timeline for availability, expected performance, or whether the chip would be used only for DeepSeek’s internal deployments or offered to external customers.

For investors watching Nvidia, the immediate takeaway is not that DeepSeek is replacing Nvidia tomorrow, but that the most capable model builders are actively exploring alternatives. That dynamic can matter over time by affecting demand assumptions for large-scale GPU clusters, particularly if more inference and serving workloads move toward custom silicon.