Chinese AI startup DeepSeek has begun developing its own chip designed for inference, according to three people familiar with the matter. The effort aims to reduce the company's reliance on Nvidia and Huawei hardware. The project remains in early stages, with no timeline for mass production disclosed.
The move comes as US export controls tighten access to advanced semiconductors for Chinese firms. DeepSeek, like many domestic players, faces supply chain vulnerabilities that could constrain its AI ambitions. Building an in-house chip is a strategic hedge against geopolitical headwinds.
The specific design targets inference workloads, a step that could lower operational costs and improve energy efficiency for DeepSeek's models. It remains unclear whether the company plans to manufacture the chip independently or through a foundry partner. The startup has not publicly commented on the project.
If successful, the chip could reshape DeepSeek's competitive position in China's crowded AI landscape. It may also pressure Nvidia, which dominates the Chinese AI chip market despite export restrictions. However, developing a competitive chip requires years of engineering and billions in investment.
The counter_argument: Custom chip projects in China have historically struggled with fabrication bottlenecks and software ecosystem challenges. DeepSeek's early-stage announcement risks overpromising on capabilities that may take years to realize, if ever.