China vs US: Who is winning in AI? | Lex Fridman Podcast

Quick Overview

The competition in AI model development is shifting, with Chinese open-source models, particularly DeepSeek, poised to overtake US dominance due to strong incentives and open development practices, as evidenced by a predicted "flip" in cumulative downloads across platforms like Hugging Face.

Key Points: The "DeepSeek moment" occurred on January 20, 2025, when DeepSeek released R1, an open-source AI matching OpenAI's O1 performance using less compute. Open models from China are projected to overtake US models in cumulative downloads on Hugging Face, with the "flip" happening around September 2025. The AI Chatbot Market Share for January 2026 predicts ChatGPT leading at 80.14%, followed by Perplexity (8.15%) and Google Gemini (7.19%), with DeepSeek at only 0.01%. DeepSeek's architecture (V3/R1) utilizes a Mixture-of-Experts (MoE) approach with more heads and experts compared to Kimi K2 (1 trillion parameters), which uses fewer heads and experts. DeepSeek's business model relies less on API subscriptions and more on open-source adoption and building engineering culture, contrasting with US companies often focusing on proprietary models. The rapid release schedule of DeepSeek models, including V3.2 in December 2025, indicates aggressive development pace in the Chinese open-source ecosystem. Nathan Lambert suggests that the open-source momentum from China, driven by cultural factors and strong incentives, is a significant disruptive force against US proprietary models.

Context: This segment of the Lex Fridman Podcast features a discussion between Lex Fridman and Sebastian Raschka, focusing on the global competitive landscape of AI development, specifically comparing the progress and business models of US-based AI labs (like OpenAI and Google) against rapidly advancing Chinese open-source developers, most notably DeepSeek. The conversation centers on the impact of open-source releases versus proprietary strategies and the projection of which region will lead in open model adoption.