Unredacted Messages Expose Microsoft’s Hand in OpenAI’s Board
Newly unredacted trial exhibits from Musk v. Altman deliver a rare, granular look at how Microsoft executives helped shape the post-crisis OpenAI governance structure. In November 2023, as Sam Altman pushed to regain control after his brief ouster, an intense text thread unfolded between Satya Nadella, CTO Kevin Scott, Microsoft president Brad Smith, and Altman. That exchange shows Microsoft leaders informally screening candidates for a revamped OpenAI board. Scott reacted with a “strong, strong no” to former Google Cloud CEO Diane Greene, while Nadella objected to both Greene and longtime Amazon-linked executive Bing Gordon over their ties to direct competitors in AI. By contrast, Belinda Johnson, former Airbnb COO, drew praise, and Nadella personally proposed former Gates Foundation CEO Sue Desmond-Hellmann, who ultimately joined the board. The messages reveal a partner-investor acting as a de facto gatekeeper over who would oversee OpenAI’s most sensitive decisions.
Satya Nadella’s IBM Warning and Microsoft’s One-Way Door
Satya Nadella’s courtroom testimony offers the strategic backdrop to Microsoft’s OpenAI board activism. In an internal email shared at trial, he compared the OpenAI deal to Microsoft’s early PC-era relationship with IBM, warning he did not want Microsoft to become the next IBM while OpenAI evolved into the next Microsoft. Nadella described the multibillion-dollar investment decision as a “one-way door”: Microsoft could not justify building two parallel supercomputers, one for its own teams and another for OpenAI, without incurring huge opportunity costs. That meant committing scarce computing resources and “outsourcing essentially a lot of the core IP development,” creating deep dependency on OpenAI’s research and models. To hedge that risk, Nadella emphasized securing access to OpenAI’s intellectual property while simultaneously building internal capabilities. His concern about dependency helps explain why Microsoft treated OpenAI governance as a strategic lever, not a peripheral issue.
Boardroom Vetoes Reveal How Tech Partners Exert Control
The November 2023 board battle illustrates how tech partnership control can extend far beyond equity stakes. While Nadella testified that Altman and other OpenAI insiders sought Microsoft’s input and could have ignored it, the text thread shows Microsoft’s preferences carried real weight. Nadella vetoed Greene and Gordon because their backgrounds linked them to rivals in cloud and AI, a sign that competitive alignment mattered as much as boardroom expertise. At the same time, Microsoft’s leaders enthusiastically floated other names: Brad Smith praised Anne Sweeney and Leslie Kilgore as steady, practical voices; Scott compiled a long list of veteran operators and even half-jokingly offered himself. Ultimately, a framework emerged around Bret Taylor, Larry Summers, and Adam D’Angelo, with Altman reinstated as CEO but off the board. The episode underscores how a strategic investor can subtly filter who governs a supposedly independent AI lab.
OpenAI’s New Governance Structure and the Power of Dependency
Following the 2023 turmoil and subsequent restructuring, OpenAI’s governance structure shifted to a foundation that now holds a minority stake in the for‑profit public benefit corporation. That foundation board includes chair Bret Taylor, Adam D’Angelo, Sue Desmond-Hellmann, and other high-profile figures from academia, industry, and security, with Altman ultimately returning as a director. Microsoft portrays this setup as proof of a mutually beneficial relationship that created one of the largest nonprofits in the world and enabled tools like ChatGPT and Copilot. Elon Musk’s lawsuit, however, argues that Microsoft’s protective stance over its investment and IP access undermined OpenAI’s original nonprofit mission. Nadella’s admission that the nonprofit had no full-time staff or visible research output through early 2026 adds fuel to concerns about mission drift. The Musk v. Altman filings suggest that dependency in AI partnerships can blur the line between support and structural capture.