Open-Source AI Advocacy from Industry Leaders
NVIDIA CEO Jensen Huang publicly endorsed open-weight AI models, arguing that US AI leadership should be measured by ecosystem robustness rather than a single frontier model. He stated that open weights enable accessibility, adaptability, and security through transparency, while closed models create single points of failure with hidden vulnerabilities. He called for policy support including expanded compute access, shared public assets, avoidance of premature restrictions, and application-layer development. Separately, DeepSeek founder Liang Wenfeng told investors that open source aligns with the company's philosophy of earning only reasonable profit, noting that AI's potential 10% GDP share makes exclusivity unsustainable. He outlined DeepSeek's AGI roadmap: coding agents as the current priority, continuous learning as the next critical breakthrough, followed by gradual singularity and embodied intelligence. He stated that China's AI gap is in resources, not talent, and predicted industry consolidation to perhaps two large and two small model companies. Competitive dimensions are cost, time-to-market, and user experience, with cost ranked first. He described DeepSeek as a vision-driven organization where restraint is a deliberate strategy.
Key insights
Open Weights Enhance Security Through Transparency
Huang argued that closed models concentrate risk in single points of failure with vulnerabilities hidden from external scrutiny, while open-weight models allow independent inspection and modification, improving overall ecosystem security.
Reasonable Profit Model Makes Open Source Commercially Viable
Liang stated that if a company only seeks reasonable profit rather than hundredfold returns, open source does not harm the business model. The AI market is large enough that capturing a small share is still enormous, and attempting to monopolize will lead to being 'abandoned by history.'
Continuous Learning Is the Next Critical AGI Milestone
Liang identified that AI currently lacks the ability to learn continuously as humans do. The next-generation model must have this capability, which would then enable AI to accelerate its own research, leading to a gradual singularity before embodied intelligence.
Industry Consolidation Is Inevitable, Cost Is the Primary Competitive Dimension
Liang predicted that too many model companies will converge to a small number, and that cost ranks first among competitive factors, ahead of time-to-market and user experience. DeepSeek's low prices reflect deliberate cost efficiency rather than subsidization.