DeepSeek targets AI chips to cut reliance on Nvidia and Huawei

Published: 2026-07-08T09:31:25.000Z

DeepSeek, the Hangzhou startup that stunned the AI world last year, is reportedly trying to build its own chip, aimed at loosening its dependence on Nvidia and Huawei. The effort is still early, and the processor is being designed for inference rather than training new models.

DeepSeek, the Hangzhou startup that stunned the AI world last year, is reportedly trying to build its own chip. According to a Reuters report citing three people familiar with the matter, the company has spent roughly the past year quietly developing a semiconductor of its own, aimed at loosening its dependence on Nvidia and Huawei for the silicon that runs its models. The effort is still early, with DeepSeek said to be in talks with chip-design, foundry and memory partners and hiring chip engineers discreetly, without the public job postings that would tip off rivals. Crucially, the chip is being designed for inference rather than training. Training is the enormously demanding process of building a model in the first place, while inference is the comparatively lighter work of running that finished model to answer users, and it is the fastest-growing slice of AI computing as these tools spread into everyday use. Inference chips can be cheaper and less power-hungry than the general-purpose graphics processors that dominate training, which makes them a sensible first target for a newcomer. It would also put DeepSeek in step with a global trend, since OpenAI recently unveiled its first custom inference chip with Broadcom and Anthropic has been weighing a similar move. For DeepSeek the calculation carries an extra, geopolitical charge. American export controls bar Chinese firms from buying Nvidia's most advanced chips, and Beijing has been leaning on its national champions to build homegrown alternatives, so an in-house processor is as much a matter of resilience as economics. The company trained the foundation model behind R1, whose startlingly cheap performance rattled US tech stocks at the start of 2025, on an Nvidia chip that Washington later restricted. Yet the obstacles are steep, since designing a competitive chip takes years and deep pockets, and separate US curbs have cut China off from the most advanced foreign foundries and from the high-bandwidth memory that inference chips depend on. One analyst was blunt that without access to leading-edge manufacturing, DeepSeek has little hope of selling silicon beyond China. The Gulf watches all this with unusual interest, because DeepSeek's efficient, low-cost models have found an eager audience there. Regional players have hosted its systems, and cost-conscious deployments across the UAE and Saudi Arabia are drawn to exactly the kind of cheap inference DeepSeek preaches. The complication is that the region is caught between two technology spheres, courting Chinese efficiency while remaining bound to American hardware and subject to Washington's pressure over Chinese kit, a tension the UAE already navigated when securing its big Microsoft partnership. A cheaper, China-made inference chip would be tempting for the Gulf's sovereign AI ambitions, but the same export-control web that constrains DeepSeek would shape whether that silicon could ever be deployed in Riyadh or Abu Dhabi.