TL;DR

US House moves to freeze state AI laws; Nvidia tops $3.45T; OpenAI acquires Windsurf.

Highlights

• U.S. House narrowly passed a bill freezing state AI regulation for 10 years, shifting oversight to the federal level and nullifying 20+ state AI statutes; Senate debate pending 1.
• OpenAI acquired developer tools startup Windsurf for $3B; Anthropic cut Windsurf’s Claude 3.x API access days before deal confirmation 213.
• Microsoft restricted key Visual Studio Code extensions to its own platform, impacting third-party editors and developer tools 213.
• Nvidia surpassed Microsoft as the world’s most valuable company ($3.45T) amid strong AI chip demand, record Blackwell GPU shipments, and major cloud deals 3.
• CoreWeave shares rose 260% post-IPO to a $72B market cap, driven by AI compute demand and Nvidia ’s investment; revenue projected to exceed $16B by 2027 4.
• CoreWeave , Nvidia , and IBM set a new MLPerf Training v5.0 record, training Llama 3.1 (405B params) on 2,496 Blackwell GPUs 5.
• OpenAI rolled out lightweight ChatGPT memory for free users globally; paid tiers retain advanced memory and new Codex features 6.
• Tesla filed for "Robotaxi" trademarks and plans an autonomous ride-hailing launch in Austin amid ongoing NHTSA FSD crash investigations and regulatory scrutiny 7.
• The New York Times signed a multi-year AI content licensing deal with Amazon (Alexa); legal action against OpenAI and Microsoft continues 8.
• Chinese lab DeepSeek’s latest model is suspected of training on Google Gemini 2.5 Pro outputs; Gemini 2.5 Pro outperformed GPT-4 in coding and legal tests 9.
• Meta, Google DeepMind , Nvidia , and Cornell quantified LLM memorization at 3.6 bits/parameter and introduced methods to reduce inference costs 10.
• Microsoft -backed Builder.ai collapsed after revelations of "AI washing," inflated revenues, and a major data breach; regulatory probes underway 11.
• Hugging Face launched SmolVLA, an open-source, low-cost robotics VLA model; H Company released open-source Holo-1 (3B/7B) vision-language-action models 14.
• Google open-sourced a full-stack Gemini 2.5 research agent stack, enabling advanced research assistant development 15.

Commentary

The U.S. House’s move to freeze state-level AI regulation for a decade, pending Senate review, signals a significant shift in the regulatory landscape 1. While this could streamline compliance for AI companies in the near term, it leaves a gap in oversight as federal standards remain undefined. Companies operating in highly regulated or consumer-facing sectors should closely watch for Senate amendments and prepare for a period of regulatory uncertainty.

Nvidia ’s rise to the top of global market capitalization and CoreWeave ’s dramatic post-IPO gains reflect sustained demand for AI compute infrastructure 34. The record-setting MLPerf benchmark using Blackwell GPUs and large-scale cloud deals (e.g., with Saudi PIF’s Humain) highlight the scale and pace of investment in next-gen AI hardware 35. However, the collapse of Builder.ai over “AI washing” and financial misrepresentation underscores the need for transparency and due diligence in both technical claims and business operations 11.

The developer ecosystem is experiencing increased consolidation and platform control. OpenAI’s acquisition of Windsurf and Anthropic’s abrupt API restrictions point to growing competition and vertical integration in the developer tools space 213. Microsoft ’s move to restrict VS Code extensions to its own platform further raises barriers for independent tool providers 213. In contrast, open-source releases from Google and Hugging Face are expanding access to advanced research agents and robotics models, providing alternatives to closed ecosystems 1415.

On the data and research front, the New York Times’ licensing deal with Amazon and ongoing legal actions against OpenAI and Microsoft highlight the evolving market for AI training data and the importance of content provenance 8. Allegations of synthetic data use in DeepSeek’s latest model, along with new research quantifying LLM memorization and efficiency, reinforce the need for transparent, auditable model development pipelines 910.