OpenAI is now a Bedrock vendor

THE AI BRIEF

Today's signal: Microsoft ended OpenAI's cloud exclusivity, and within 48 hours, OpenAI was live on AWS. Plus five shorter reads on Musk, Manus, NVIDIA, Joby, and Google's Pentagon deal.

THE READ

The amended Microsoft deal turned a 6-year exclusive into a multi-cloud reality in less than 48 hours. The procurement implications start now.

On Monday, Microsoft and OpenAI announced an amended agreement that ends Microsoft's exclusive cloud rights to OpenAI's models and replaces them with a revenue-share. By Tuesday afternoon, AWS had OpenAI's frontier models, Codex, and a new product called Bedrock Managed Agents live in limited preview on Bedrock. Reuters reported the move clears the way for OpenAI's previously contested $50B compute deal with Amazon, and gives Microsoft a continuing cut of OpenAI revenue in exchange for the loosened terms.

What I'd watch for buyers is the plumbing, not the headline. OpenAI on Bedrock inherits the AWS enterprise stack: IAM, PrivateLink, guardrails, encryption, and CloudTrail logging. Codex authenticates with AWS credentials, and usage of both OpenAI models and Codex applies toward existing AWS cloud commitments.

For any company already on AWS, the friction of running a real OpenAI workload just dropped to roughly the friction of turning on a new Bedrock model family. Security review, data residency story, and budget posture stay where they already are.

Microsoft is still the closest thing to a default OpenAI deployment surface, especially for shops on Azure and Microsoft 365. The calculus that "OpenAI = Azure, Anthropic = AWS, Gemini = Google Cloud" is gone, and frontier models are becoming a feature of the cloud you already use rather than a reason to pick a different one. That changes how IT, procurement, and security teams should think about RFPs that lock a specific cloud to a specific model family.

Two things to watch. One: how AWS prices Bedrock Managed Agents against OpenAI's direct API and Microsoft's Azure equivalents, because that's where the actual lock-in math lives. Two: whether Anthropic gets reciprocal Azure access in any form, because if it does, the "model on every cloud" pattern is the new default, and pretty much every multi-year cloud-AI commitment signed in the last 18 months is worth a second read.

FREE WORKSHOP

I'm running a free 30-minute session on how you can build a real workflow inside ChatGPT Workspaces tomorrow. RSVP here:

ALSO WORTH KNOWING

Elon Musk took the stand in the OpenAI trial
Musk testified for two hours Tuesday in Oakland, accusing OpenAI's executives of converting a nonprofit they had no right to take private. The case is the public venue where OpenAI's for-profit conversion, IPO path, and Microsoft ties get litigated.

China ordered Meta to unwind its $2B acquisition of Manus
Beijing blocked the deal after a months-long antitrust probe, citing concerns about losing technology that Manus's Chinese-origin team had developed. Cross-border AI M&A involving Chinese-rooted companies is now an antitrust question in two jurisdictions, not one.

NVIDIA released Nemotron 3 Nano Omni as an open-weight multimodal model
The 30B mixture-of-experts model handles vision, audio, and text in a single system and runs on one GPU with 3B active parameters. NVIDIA is using model releases to define what efficient agentic infrastructure looks like on its hardware.

Joby flew an electric air taxi from JFK to Manhattan
The 15-minute eVTOL flight to West 30th Street Heliport was NYC's first point-to-point electric air-taxi flight, part of a 10-day FAA pilot program. The regulated infrastructure for low-emissions urban transport is moving from deck to flight log.

Google signed a classified AI deal with the Pentagon over employee opposition
The agreement allows the Department of Defense to use Google's AI on classified work for "any lawful government purpose," announced the same day 600+ Google employees wrote Sundar Pichai opposing it. Eight years after Project Maven, the defense-AI line has moved.

WATCHING TOMORROW

Microsoft, Alphabet, Meta, and Amazon all report after close tonight. Four companies collectively committed to spending $645B on AI infrastructure this year. The number to watch is whether the capex guidance holds or creeps higher without the cloud growth to back it. Apple follows Thursday. I'll have the operator read in Thursday's issue.

Back tomorrow,
Haroon