Stanford's annual AI Index report reveals the US-China AI gap has effectively closed. In May 2023, OpenAI's GPT-4 led Chinese models by over 300 Arena points; by March 2026, Anthropic's Claude Opus 4.6 leads China's top model by just 2.7%. China now dominates in publication volume, citations, patent output, and robot installations, while the US retains an edge in frontier model development. The report also finds AI adoption is outpacing the PC and the internet, but companies are spending hundreds of billions on infrastructure with uncertain returns.
AI chip maker Cerebras Systems is making its IPO paperwork public today, aiming to raise approximately $3 billion at a valuation north of $35 billion. The filing comes after a failed 2025 attempt that was derailed by regulatory concerns over UAE investor G42. A $10 billion multi-year compute deal with OpenAI — the largest non-Nvidia AI infrastructure contract ever — strengthens the company's revenue story heading into the listing on Nasdaq under ticker CBRS.
A finding highlighted in both the Stanford AI Index and Nature reports that the best AI agents perform only about half as well as PhD-level human experts on complex scientific tasks. Despite rapid progress in agentic AI workflows, the study pours cold water on claims that AI agents are ready to autonomously conduct end-to-end scientific research. The gap underscores that current AI excels at narrow, well-defined tasks but struggles with the open-ended reasoning real science demands.
Mozilla's for-profit arm MZLA Technologies launched Thunderbolt, a self-hostable AI client designed to give enterprises an alternative to Microsoft Copilot, ChatGPT Enterprise, and Claude Enterprise. Built on Deepset's Haystack framework, it supports MCP and ACP orchestration protocols, connects to Anthropic, OpenAI, Mistral, and local models via Ollama. It ships under MPL 2.0 on all major platforms. The play is data sovereignty — your AI, your infrastructure, your data stays on-prem.
A PwC survey of 1,217 senior executives across 25 sectors finds a stark AI winner-take-most dynamic: the top 20% of companies generate 7.2x more AI-driven revenue and efficiency gains than the average competitor. The differentiator is not AI spend but business model redesign — leaders are twice as likely to rebuild workflows around AI rather than layer it onto existing processes. The strongest predictor of returns is "industry convergence," using AI to expand beyond traditional sector boundaries.
The U.S. General Services Administration has launched a "million hours challenge" using its internal AI tool USAi, after losing nearly 40% of its workforce since October 2024. About 400,000 hours of automatable work have already been identified. The agency is following an "eliminate, optimize, automate" playbook, with roughly half its remaining staff already using GSAi daily. It is the most aggressive federal AI automation initiative to date and a live test of whether AI can compensate for severe staffing cuts.
