The US is losing its lead in AI

What It Covers

The Vergecast examines the narrowing AI gap between the US and China, analyzing how Chinese firms like DeepSeek, Moonshot, and Alibaba have reached near-parity with Silicon Valley's frontier models, and what this means for US policy, tech regulation, and the competitive strategies of OpenAI, Anthropic, and Google.

Key Questions Answered

• •China's AI timeline: The "six months behind" benchmark that once described China's lag behind US frontier models now represents the maximum possible gap, not the average. For everyday consumer-facing models, Chinese labs are either already at parity or will reach it within weeks. Businesses evaluating AI vendors should factor Chinese alternatives into procurement decisions immediately.
• •Distillation as competitive shortcut: Chinese firms DeepSeek, Moonshot AI, and MiniMax collectively generated approximately 16 million fraudulent exchanges with Anthropic's Claude to extract training data. This technique — feeding a target model mass queries and using responses to train a rival system — allows resource-constrained teams to compress years of development into months without building foundational capabilities independently.
• •Chip export controls losing effectiveness: The core US strategy of restricting advanced semiconductor exports to slow China's AI development was undermined when DeepSeek produced a competitive frontier model using fewer and less advanced chips. Policymakers now face pressure to develop alternative containment strategies, as hardware restrictions alone cannot prevent Chinese labs from achieving model parity.
• •AI labs leveraging China threat for regulatory relief: US frontier labs — OpenAI, Anthropic, and Google DeepMind — are using China's AI advances as political leverage to resist domestic regulation. The Trump administration's executive orders explicitly prioritize speed over safety guardrails, and labs argue that any regulatory constraint risks ceding ground to Chinese competitors, a framing that has gained traction in Washington.
• •Open-source models creating enterprise pressure: Enterprise customers of major AI labs are actively evaluating open-source model alternatives to reduce costs, routing different query types to different models to save millions in API fees. If Chinese labs continue producing cheaper, efficient frontier-quality open-source models, US labs face simultaneous competitive pressure from abroad and customer defection at home.