DeepSeek
Chinese AI lab (Hangzhou) founded 2023 by Liang Wenfeng. Releases open-weight LLMs that shocked the market with GPT-4-comparable quality at a fraction of training cost.
Major milestones: DeepSeek-V2 (June 2024, MoE), DeepSeek-V3 (December 2024, 671B params with 37B active, $5.5M training), DeepSeek-R1 (January 2025, reasoning model open source under MIT, beat OpenAI o1 on several benchmarks). Nvidia stock crash on January 27, 2025 (-17%) when the market grasped the implication for AI compute. Models are open but training data + recipes are secret.