Court documents from the Musk v. Altman trial just exposed how nervous Microsoft really was about losing OpenAI to Amazon back in 2017 and 2018, according to The Verge AI. The internal emails, surfaced through litigation and detailed by The Verge AI, capture Microsoft CTO Kevin Scott openly worrying that OpenAI could ‘storm off to Amazon in a huff and shit-talk us and Azure on the way out.’ That single line, written in January 2018, helps explain why Microsoft eventually wrote a $1 billion check.
This is significant because it rewrites the origin story of the most consequential partnership in modern AI. The official version was always strategic conviction. The emails show something messier: hesitation, internal pushback, and a fear of public embarrassment.
What the emails reveal
The timeline laid out by The Verge AI starts right after OpenAI’s Dota 2 bot beat a pro player in summer 2017. Sam Altman replied to Satya Nadella’s congratulations note with a much bigger ask: roughly $300 million in Azure compute at list prices to fund the next phase of research.
That number rattled Microsoft’s leadership.
- Jason Zander, then Azure chief, told Nadella the deal only made sense if it generated ‘$500 million+’ in incremental revenue that couldn’t be earned ‘in a more efficient way.’
- Altman pivoted, proposing a partnership with Xbox around gaming plus shared IP in exchange for sponsorship of the Dota research.
- The Xbox team was curious but wouldn’t fund the research alone.
- Kevin Scott admitted he wasn’t sure what Microsoft would ‘get out of’ the deal, but flagged the PR risk of letting OpenAI walk to AWS.
Scott also conceded a year later, in an email to Nadella and Bill Gates, that he had been ‘highly dismissive’ of OpenAI and Google DeepMind when they were chasing ‘game-playing stunts.’ What changed his mind was OpenAI’s pivot to natural language models and the growing fear that Microsoft would fall behind Google. A month after that email, Microsoft announced its first $1 billion investment.
Why this matters now
The punchline is that Scott’s 2018 nightmare scenario is basically playing out in 2026, just with a delay and a contract renegotiation in between. The Verge AI notes that OpenAI recently restructured its deal with Microsoft to bring its models, Codex, and other tools to AWS. Days before the announcement, OpenAI told employees the Microsoft arrangement had ‘limited our ability to meet enterprises where they are,’ which for many customers means Amazon Bedrock.
That’s the ‘shit-talk’ Scott was worried about, delivered seven years late and with a multi-cloud distribution deal attached.
A few takeaways for anyone tracking the AI infrastructure wars:
- Cloud exclusivity is over. Even Microsoft’s flagship AI partner is now multi-cloud. Expect every other foundation model lab to follow.
- Compute leverage flipped. In 2017, OpenAI was begging for Azure credits. Today, OpenAI dictates terms and AWS, Oracle, and Google line up.
- The ‘PR downside’ framing was prescient. Scott understood early that AI labs were becoming influential voices, not just customers. That insight is now table stakes for every enterprise dealing with model providers.
What stands out here is how thin the margin was. If Zander’s revenue math had won the internal debate, or if Scott had stayed dismissive a few months longer, the entire shape of the AI industry might look different. Microsoft didn’t bet on OpenAI because it saw the future clearly. It bet because the alternative, watching OpenAI badmouth Azure from inside an AWS data center, felt worse.
Full email excerpts and the broader court filing context are available at the original report on The Verge AI.