The headline act was OpenAI and Microsoft tearing up the deal that defined commercial AI for the past three years. On Monday, the two companies announced a sweeping overhaul: Microsoft loses exclusive rights to OpenAI's models, retaining only non-exclusive IP access through 2032. Its revenue share drops to 20% of OpenAI revenue until 2030, subject to an undisclosed cap. The controversial AGI clause — which would have restructured the partnership once artificial general intelligence was achieved — is gone entirely. Hours later, Amazon CEO Andy Jassy announced that OpenAI's models would be available on AWS Bedrock "in the coming weeks." The practical effect is that OpenAI can now sell everywhere: Azure, AWS, and potentially Google Cloud. It also clears the legal path for the $50 billion Amazon investment announced in February. For Microsoft, the trade-off is more freedom too — it no longer shares revenue on Azure-served OpenAI models, and can more aggressively build its own model stack.

The same week, OpenAI released GPT-5.5 — its most agentic model yet. Where GPT-5.4 was a reasoning leap, 5.5 is built to carry entire workflows: writing and debugging code, researching the web, analysing data, creating documents, and moving across tools until a task is done. It matches 5.4's per-token latency while performing at a significantly higher level of intelligence. OpenAI says it has more than 900 million weekly active users and is approaching $2 billion in monthly revenue. The model rolled out to Plus, Pro, Business, and Enterprise users, with a Pro-tier variant for heavier workloads. The timing — dropping a flagship model the same week as the Microsoft restructure — felt deliberate: a show of independence.

Anthropic had a different kind of week: money coming in, creative tools going out. Google announced plans to invest up to $40 billion in Anthropic — $10 billion now at a $350 billion valuation, with another $30 billion tied to performance targets, plus 5 gigawatts of compute capacity over five years. This landed days after Amazon committed up to $25 billion in a separate deal. Between the two, Anthropic has lined up roughly $65 billion in new capital from hyperscalers alone, while its annualised revenue has hit $30 billion. On the product side, Anthropic launched nine Claude connectors for creative tools — Blender, Adobe Creative Cloud, Ableton, Autodesk Fusion, SketchUp, Splice, and others. The idea: describe what you want in plain English and the connector chains the right tools together. Anthropic also became a Blender Development Fund patron, backing the open-source 3D suite.

Google, meanwhile, expanded the Pentagon's access to its AI after Anthropic's continued refusal to work with the Defence Department — a gap Google is now openly stepping into. The company also announced that Gemini will begin rolling out to cars with Google built-in, replacing the current Google Assistant in millions of vehicles. And Apple reported a quiet surprise: Mac revenue hit $8.4 billion in Q2, beating Wall Street's expectations on the back of AI-driven demand. Developers and enterprises are buying more Macs to run local AI workloads, a trend Apple hadn't fully anticipated.

A few more threads worth noting. Novo Nordisk formalised a strategic partnership with OpenAI to integrate AI across drug discovery, clinical trials, manufacturing, and supply chain — with pilot programmes launching this year and full integration by the end of 2026. On the security front, a sobering reality check: three leading AI coding agents — Claude Code, Gemini CLI, and GitHub Copilot — were all hit by a single prompt injection technique hidden in pull request descriptions, enabling remote code execution. The vulnerability scored a 9.6 on CVSS. Separately, Nvidia-backed legal AI startup Legora hit a $5.6 billion valuation after a $50 million Series D extension, intensifying its battle with Harvey. The vertical AI race is spreading fast — pharma, law, defence, creative tools — and this week made it clear that the platform layer underneath is being rebuilt in real time.

Signals

Week in Review: May 1