DeepSeek charts path to custom AI chips as export curbs limit Nvidia access

DeepSeek, a Chinese artificial intelligence company building language models that rival those of OpenAI and Anthropic, is moving into chipmaking to navigate tightening export restrictions and secure its computing infrastructure.

According to reporting from Reuters, the startup has spent roughly a year pursuing the initiative, establishing relationships with hardware partners and assembling an engineering team for the project.

Targeting inference, not training#

DeepSeek's strategy focuses specifically on data center chips designed for inference—the computational work of running trained models—rather than the more demanding training phase. This direction reflects broader industry thinking about where custom silicon adds the most value and where supply chain independence matters most.

The company's goal encompasses reducing its reliance on multiple suppliers: American chipmaker Nvidia, whose export bans have limited its reach in China, and Huawei, which holds roughly half the Chinese data center chip market. That competitive pressure extends beyond DeepSeek; other major Chinese tech firms including Alibaba and Baidu are similarly exploring silicon development.

A trend across the industry#

Custom chip efforts are not unique to China's AI champions. Earlier this year, OpenAI and Broadcom jointly revealed Jalapeño, OpenAI's first inference-focused silicon offering. Anthropic has also been investigating proprietary chip design, though without public announcements yet.

For American AI leaders like OpenAI, the motivation blends multiple concerns: decreasing dependence on Nvidia's dominance, securing full control over their technology stack similar to Apple's model, and securing computing capacity in a market where demand increasingly outpaces available resources.

Why it matters#

As AI companies worldwide race to scale their models and services, reliable access to compute becomes a strategic asset. Vertical integration—controlling everything from algorithms to chips—increasingly shapes competitive advantage in the industry.