DeepSeek gives away the engine behind its cheap AI
DeepSeek open-sourced DeepInfer, the production inference framework serving 155 trillion tokens per day, under Apache 2.0 — with NVIDIA and Huawei Ascend support from day one.
2026-10-212 views
DeepSeek open-sourced DeepInfer, the production inference framework serving 155 trillion tokens per day, under Apache 2.0 — with NVIDIA and Huawei Ascend support from day one.
DHS invokes FASCSA for the first time against an AI company, ordering federal agencies to remove DeepSeek and any third-party product built on it.
A 763B open-weight budget model tops four agentic benchmarks while costing $0.18 per task.
DeepSeek doubles its second funding round to 100 billion yuan at an unchanged $74.5B valuation, with Tencent, CATL, SMIC and Geely among investors.
Chinese AI startup DeepSeek is closing a $12-15B funding round backed by Tencent and CATL, targeting a 2027 IPO at a $75B valuation.
DeepSeek doubles annualized revenue to $1B and raises $7.45B at a $74B valuation, positioning as the most valuable AI company outside the US.
DeepSeek V4.1-Flash uses a new Causal Encoder-Decoder architecture to deliver near-flagship performance on 1/4 the HBM — and is replacing V4-Pro.
DeepSeek plans a 160,000-chip Huawei Ascend 950DT data center in Inner Mongolia — the largest known Ascend cluster ever, and the boldest move yet by a Chinese AI lab to replace NVIDIA hardware.