DeepSeek: Builds Own AI Inference Chip

DeepSeek quietly advances its own inference-focused AI chip project that began a year ago, hiring engineers and contacting foundries without public postings. The move follows the January 2025 R1 model release that erased $600 billion from Nvidia market value in one day and now directly challenges both Nvidia and Huawei Ascend dominance in China’s $50 billion AI chip market. Nvidia already recorded a $4.5 billion inventory charge in Q1 2026 with another $8 billion expected in Q2 after export controls slashed its China revenue share from 20-25 percent to 9 percent. DeepSeek itself shifted to Huawei Ascend 950 processors for its April 2026 V4 model before reversing its no-funding policy to secure $7 billion at a $52-59 billion valuation specifically to fund hardware development. The effort mirrors broader AI industry impact as OpenAI, Google and Meta also pursue custom silicon to cut Nvidia dependence while US curbs limit access to advanced foundries and high-bandwidth memory.