DeepSeek founder Liang Wenfeng says the company exists to maximize AGI odds, not profit
智东西 · wechat · 2026-07-23
DeepSeek founder Liang Wenfeng lays out a full AGI-first worldview
A long transcript from an investor call shows Liang explaining DeepSeek’s strategy across company mission, open source, model roadmap, competition, and compute.
- Mission and management: DeepSeek was not founded to maximize profit. Liang says the company is organized around a shared AGI vision rather than KPIs, and that keeping the team stable is its only core interest.
- Pricing and open source: DeepSeek prices APIs to recover hardware costs in about 10 months and argues that open source is compatible with paid APIs as long as profits stay reasonable. He says the strongest future model may also be open-sourced.
- AGI roadmap: Liang frames current AI progress as a sequence of stages: CoT, then agents, then continuous learning, then self-improvement. He says agentic systems are not the endpoint, and that the next real bottleneck is long-horizon learning.
- Product strategy: He says DeepSeek should not rush consumer products or commercialization, because AGI progress is still the highest-return use of effort. Video generation and world models are not on the main path unless they directly help intelligence.
- Compute and China: Liang says the gap between China and the US is mainly resources, not talent. DeepSeek is aggressively buying Nvidia GPUs, while also betting on domestic hardware ecosystems. He argues CUDA’s moat is eroding and claims Huawei’s 950 supernode could eventually stand in for Nvidia’s GB200/GB300 class systems.
- Industry outlook: He predicts China will become one of the three major poles in global AI, with cost as its biggest structural advantage, and says the domestic base-model market will likely consolidate into only a few major players.
Related event: DeepSeek's Liang Wenfeng Pledges Sole Focus on AGI Over Profits(18 posts)→
More from AGI Musings
- Superintelligence will be maximum good, not stupid or evil, argues Patterson — davidpattersonx · 2026-09-11
- Mathematician Daniel Litt Launches Problem Repo to Track Human vs AI Progress: 15 Problems, 1 Solved — littmath · 2026-09-11
- Should AI models be taught morality? Breakout incidents expose missing ethical training — Pfungus_ · 2026-09-11
- SoftBank's Masayoshi Son predicts 100 trillion self-replicating AIs: "humans' era as top life form is ending" — Puzzleheaded-King584 · 2026-09-11
- We are witnessing the unreasonable effectiveness of inference-time scaling — sqcai · 2026-09-11
- The AlphaFold lesson: AI-solved math may mean fewer mathematicians needed — kiki-le-koala · 2026-09-11