Chinese AI startup DeepSeek is developing a proprietary chip focused on inference the stage where trained models generate responses in a bid to reduce its dependence on Nvidia and Huawei hardware, Reuters reported, citing three people familiar with the matter. The project, which sources say began about a year ago, remains in early stages as the company holds discussions with chip‑design firms, foundry partners and memory suppliers while quietly recruiting semiconductor design engineers.
According to the report, the chip is intended for inference rather than training, reflecting industry demand as more compute is devoted to running models in production environments. DeepSeek has historically relied on Nvidia GPUs for training and has also optimized models to run on Huawei’s Ascend accelerators, but the startup’s move toward custom silicon would give it direct control over performance, cost and supply sequencing.
The initiative coincides with DeepSeek’s first major external fundraising, which Reuters and other outlets say could total about 50 billion yuan ($7.4 billion), with Tencent and CATL among the largest prospective investors and founder Liang Wenfeng contributing personal funds to the round. Sources caution that chip development is capital‑intensive and can take years to yield commercial results; DeepSeek has not publicly commented on the project.
Industry observers say inference chips are an increasingly contested segment as AI deployments scale, but Reuters’ sources note DeepSeek has not disclosed a foundry partner, prototype or performance benchmarks, and the effort appears limited to early partner talks and recruiting so far. The company’s move follows earlier hiring drives for semiconductor design talent reported in 2025 that signaled potential plans to build proprietary processors.
DeepSeek’s decision to pursue in‑house inference hardware comes amid broader efforts by Chinese AI firms to secure supply‑chain autonomy and manage reliance on foreign vendors for critical compute infrastructure, according to the reporting. The startup’s fundraising and hiring activity suggest the effort is strategic, but concrete technical or commercial milestones have not been disclosed publicly.
Read Article: Google to Count Android Backups Toward Google Account Storage
