The Anthropic RSP story is the headline, and it should unsettle anyone who took "safety-first" at face value. Anthropic built its identity — and its valuation — on the promise that it would be the responsible lab. Dropping flagship safety commitments while simultaneously publishing fascinating alignment research (the Persona Selection paper) captures the contradiction perfectly: the research arm is doing rigorous work on understanding what these models actually are, while the business arm is dismantling the policies that were supposed to constrain what they become.

Meta's $60B AMD play is the biggest compute story of the year so far. This isn't about AMD vs. Nvidia in the abstract — it's about whether any single company should have the leverage Nvidia currently holds over the entire AI industry. Meta is spending nation-state money to create a credible alternative. If AMD delivers on this contract, the Nvidia premium starts eroding by Q4.

The margin miss story from The Information is the quiet bombshell. OpenAI and Anthropic — the two most-funded AI companies on Earth — can't hit their own gross margin targets. Inference is expensive. Scaling is expensive. The "we'll figure out margins later" playbook works until your investors start asking when "later" arrives. This is the financial pressure that explains the Anthropic safety retreat: when margins are tight, principles become luxuries.

Mercury 2 at 1,000+ tokens/second via diffusion is technically significant. If diffusion-based LLMs can match autoregressive quality at 10x the speed, the entire inference cost equation changes — which is exactly what OpenAI and Anthropic need right now. Watch this space closely.

The METR productivity study update is the kind of honest, unsexy work that matters more than most announcements. If AI coding tools deliver less productivity gain than advertised, the enterprise spending wave may decelerate. Or — more likely — companies will buy the tools anyway because the narrative is too powerful to resist.