Deepseek is moving deeper into AI infrastructure by designing its own AI chip, according to Reuters. The effort is still early, but it signals a push to control more of the hardware stack behind its models.
The chip is being designed for inference, the stage when a trained model generates responses for users. That distinction matters because inference is different from training new models, and the source specifically describes Deepseek’s project as focused on the response-generating phase.
Why Deepseek is looking at its own AI chip
Three people familiar with the matter told Reuters that Deepseek’s chip is intended for inference, not for training. In plain terms, the project is about the compute needed after a model has already been built, when people are asking it questions and receiving outputs.
That focus could help explain the strategic value of the work. If Deepseek can build an inference chip that supports its own systems, it could reduce its reliance on Nvidia and Huawei chips. The source does not say that Deepseek has replaced either supplier, only that the move could make it less dependent on them.
For an AI company, dependence on outside hardware can shape cost, availability, and the pace of deployment. The more a company can align its chips with its own models and user workloads, the more direct control it may have over one of the most important parts of running AI services.
The project is still at an early stage
Reuters reports that Deepseek’s chip project remains in its early stages. The company is talking to chip design, manufacturing, and memory companies, which suggests the work is still being assembled across several parts of the semiconductor supply chain.
Those conversations matter because an AI chip is not only a design problem. It also depends on manufacturing capacity and memory, both of which affect whether a design can become a working product at useful scale.
Deepseek has also been quietly hiring chip engineers for months without posting public job listings. That detail points to a deliberate internal buildout, but it does not by itself show how far the company has progressed. The source does not identify the engineers, the size of the team, or a timeline for a finished chip.
What is clear is that Deepseek is not treating the chip effort as a public recruiting campaign. The company appears to be building capability while keeping the project relatively low profile.
Export controls create a difficult path
The route ahead is not straightforward. US export controls have cut Chinese companies off from access to the most advanced chips and memory, according to the source article.
That constraint is central to the story. If a company cannot freely access the most advanced hardware components, designing its own chip becomes both more appealing and more difficult. It may offer a path toward greater independence, but the same restrictions can affect the tools, memory, and manufacturing options needed to make the plan work.
Deepseek’s project therefore sits inside a larger hardware challenge for Chinese AI companies. The company can pursue its own design, but the source makes clear that advanced chips and memory remain constrained by export controls.
Deepseek is not alone in custom AI hardware
Deepseek’s move also fits a broader pattern among major AI developers. OpenAI and Anthropic are also working on their own chips, according to the source.
The source does not provide details about those efforts, but the comparison is important. It shows that custom hardware is becoming a strategic priority for more than one AI company. For businesses running large AI systems, chips are not merely a supplier decision; they are part of the product and operating model.
That does not mean every AI company will successfully build its own chip. It does mean the pressure to optimize hardware for AI workloads is becoming more visible. Deepseek’s inference focus is one version of that trend.
A funding push adds another layer
Deepseek is also raising outside capital for the first time, according to Reuters. The company is looking to bring in $7 billion at a valuation of $52 to $59 billion.
That fundraising effort is separate from the technical challenge, but it gives the chip project a financial backdrop. Designing a chip, hiring engineers, and coordinating with design, manufacturing, and memory companies all require resources and long-term commitment.
The source does not say how much of the proposed capital would go toward the chip project. Still, the timing is notable: Deepseek is exploring a major hardware effort while also seeking outside capital for the first time.
Taken together, the details point to a company trying to expand beyond model development into the infrastructure that supports AI deployment. The outcome is still uncertain, but the direction is clear: Deepseek wants more control over the chips that help power its AI responses.