This issue is dominated by OpenAI, from a massive funding expansion to broader moves around compute, product strategy, and safety. Elsewhere, infrastructure keeps pulling to the center as companies chase chips, power, and security, while new research highlights how AI usage patterns and research workflows are shifting in practice.
OpenAI secured an additional $10 billion commitment from investors including a16z, DE Shaw Ventures, MGX, TPG, and T Rowe Price. That pushes the company's cumulative fundraise above $120 billion. The company is also moderating spending while targeting roughly $600 billion in total compute spend through 2030.
Read Full ArticleOpenAI is shutting down the Sora app, API, and website amid legal pressure. At the same time, it is reportedly in advanced talks with Helion Energy for large-scale fusion power capacity and is broadening its strategy around compute, health research, and safety. ChatGPT is also adding shopping and file-management features, alongside open-source teen safety prompts.
Read Full ArticleMETR ran a tabletop exercise to examine research workflows and bottlenecks as model releases accelerate. The group focused on how hard it may become to stay current as the number of new evals grows. The write-up looks at what workflows emerged, where the bottlenecks showed up, and how much faster researchers might actually get.
Read SourceAnthropic's latest Economic Index report says Claude usage has diversified. The report finds a decline in the share of high-value tasks and an increase in lower-wage personal queries. It offers another data point on how mainstream AI use may be broadening beyond premium knowledge work.
Read SourceEpoch AI says a difficult hypergraph problem that would take an expert human about one to three months was solved by two people using GPT-5.4 Pro. The write-up includes a full transcript of the interaction. It also notes that GPT 5.4 xhigh, Gemini 3.1 Pro, and Claude Opus 4.6 max solved the problem as well.
Read SourceAI-powered SIEM with agentic threat detection, launched alongside moves to improve secure agent deployment.
Inference engine that streams a 397B MoE model from SSD on a 48GB MacBook Pro through a custom Metal pipeline.
to expand compute capacity in one of the sector's biggest infrastructure bets.
with a personal CEO agent in development and broader use of internal tools tied to daily work.
with $100 million from Brett Adcock and plans for personalized AI paired with dedicated hardware.
raising fresh scrutiny around trust, verification, and security claims in AI vendor marketing.
as reliability and output quality become stronger competitive factors in consumer AI.
Get the latest AI news, research insights, and practical implementation guides delivered to your inbox daily.
By subscribing, you agree to our Terms of Service and Privacy Policy.