TL;DR

Nvidia, AMD resume China AI chip sales with US revenue share; Chinese open LLMs gain globally; Anthropic, OpenAI advance model capabilities.

Highlights

• Nvidia and AMD secured US export licenses to resume AI chip sales to China, agreeing to a 15% revenue share with the US government; legal and market uncertainties remain 1.
• Chinese firms DeepSeek and Alibaba are rapidly growing global adoption of open-source LLMs, raising US policy and security concerns 2.
• California finalized rules requiring transparency and bias assessments for AI-driven hiring tools; Workday faces a bias lawsuit 3.
• OpenAI is backing Merge Labs, a brain-computer interface startup valued at $850M, to compete with Neuralink 4.
• CoreWeave’s Q2 revenue rose 207% to $1.21B, but losses widened; AI infrastructure demand remains strong with a $30.1B backlog 5.
• Anthropic expanded Claude Sonnet 4’s context window to 1 million tokens, outpacing GPT-5’s 400,000-token limit 6.
• OpenAI halved GPT-5 latency, improved medical reasoning, reinstated GPT-4o after user backlash, and added Google connectors plus a low-cost “ChatGPT Go” tier in India 78.
• Microsoft is offering multimillion-dollar packages to recruit Meta AI talent, escalating the AI hiring competition 9.
• YouTube launched a US pilot for AI-based age verification, responding to regulatory pressure and privacy concerns 10.
• Figure’s Helix robot autonomously folded laundry, demonstrating generalization to new dexterous tasks with additional training data 11.
• IBM and Google announced plans to build million-qubit quantum computers by 2030; global competition in quantum hardware is increasing 12.
• A Lancet study found a 20% drop in doctors’ colonoscopy detection rates after routine AI use was withdrawn, highlighting clinical “de-skilling” risks 15.

Commentary

US policy on AI chip exports to China is shifting toward a revenue-sharing model, with Nvidia and AMD now able to resume certain sales in exchange for a 15% cut to the US government 1. This move offers a partial reprieve for US chipmakers but increases operational complexity and may prompt further regulatory scrutiny 1. At the same time, Beijing is encouraging domestic firms to reduce reliance on US chips, adding competitive pressure in a key market 1.

Chinese open-source LLMs, notably from DeepSeek and Alibaba , are gaining traction globally due to permissive licensing and commercial flexibility 2. This trend is prompting concern among US policymakers about the erosion of US influence over AI standards and the potential security risks associated with widespread open-source deployment, especially around prompt-injection vulnerabilities 2.

Regulatory oversight is tightening: California’s new rules on AI-driven hiring require transparency, opt-outs, and bias risk assessments, reflecting growing legal scrutiny as seen in the Workday lawsuit 3. Enterprises using AI for HR functions should prepare for increased compliance burdens and documentation requirements 3. YouTube ’s AI-powered age verification pilot and UK police guidance on facial recognition reflect the broader trend of AI regulation intersecting with privacy and civil liberties 1013.

On the infrastructure and product side, CoreWeave’s revenue surge and backlog growth highlight robust demand for AI compute, though profitability remains a challenge 5. Advances in model capabilities—such as Anthropic’s 1M-token context window 6 and OpenAI ’s latency and medical reasoning improvements 7—are driving differentiation in the developer and enterprise market. OpenAI ’s introduction of new connectors and regional pricing, and the reinstatement of GPT-4o after user feedback, signal a focus on user retention and global market expansion 8.

Finally, the AI talent market is highly competitive, with Microsoft targeting Meta researchers with aggressive offers 9. OpenAI ’s investment in Merge Labs 4 and Figure’s progress in robotics 11 point to continued expansion into hardware and embodied AI. The Lancet study on clinical “de-skilling” from AI use underlines the need for careful integration of automation in sensitive domains 15.