DeepSeek Founder Pledges Strongest Model Open Source, Says China Trails U.S. AI by 12-18 Months
Updated
Updated · geopolitechs.org · Jul 23
DeepSeek Founder Pledges Strongest Model Open Source, Says China Trails U.S. AI by 12-18 Months
3 articles · Updated · geopolitechs.org · Jul 23
Summary
Liang Wenfeng told investors DeepSeek will open-source even its strongest model, framing openness as central to its AGI mission rather than a competitive concession.
Ten-month hardware payback guides API pricing, Liang said, as DeepSeek avoids chasing consumer traffic or enterprise fads and treats commercialization as a byproduct of AGI research.
20,000 H-equivalent GPUs now back DeepSeek, but Liang said China remains 12-18 months behind the U.S. and is advancing with roughly one-twentieth of American compute.
Agents are this year's step and continual learning the next, Liang said, with DeepSeek aiming for models that can improve themselves before moving toward embodied intelligence.
A few hundred million dollars in ARR is possible this year if GPU supply expands, but Liang said China's crowded base-model field will consolidate even if domestic models match foreign peers this year.