OpenAI's boardroom drama and Microsoft's $13bn exposure

Sam Altman was pushed out of OpenAI over a weekend and reinstated days later, an episode governance scholars will study for years as farce. For Microsoft, which has put $13 billion into the artificial intelligence start-up, it was a near miss. The company had committed billions of dollars of its shareholders' money to a business whose board owed nothing to equity investors.

How Microsoft came to be exposed

The relationship began modestly. Microsoft invested $1 billion in 2019, the year OpenAI created its capped-profit arm to raise outside money, and it added more in 2021. The big commitment came in January this year, a few weeks after ChatGPT was released to the public and became the fastest-spreading consumer product anyone could remember. That multi-year, multibillion-dollar deal brought the total to the figure now quoted.

The money bought a licence and a customer. OpenAI's models run on Microsoft's Azure cloud, and Microsoft has threaded them through its products, from the Bing search engine to GitHub Copilot, a coding assistant. On 1 November it began selling Microsoft 365 Copilot to large businesses at $30 per user per month, a bet that office workers will pay for software that drafts their emails and summarises their meetings. Much of the investment case Mr Nadella has put to shareholders this year runs through a company he does not control.

Five days in November

The board fired Mr Altman on Friday 17 November, saying only that he had not been consistently candid with it. Greg Brockman, the president, lost his seat as chairman and resigned within hours. Microsoft, the largest outside investor, was given almost no warning. Mira Murati, the chief technology officer, was named interim chief, and by Sunday night she too had been replaced, by Emmett Shear, a former boss of Twitch.

On Monday Microsoft said it would hire Mr Altman and Mr Brockman to run a new research group, and nearly all of OpenAI's staff signed a letter threatening to follow them unless the board resigned. Among the signatories was Ilya Sutskever, the chief scientist, who had sat on the board that voted for the firing and now said he regretted it. That threat decided the matter. Late on Tuesday OpenAI agreed to take Mr Altman back, with a new board chaired by Bret Taylor, a former co-chief executive of Salesforce, alongside Larry Summers and Adam D'Angelo, the only director to survive from the old board.

A board investors cannot touch

OpenAI's structure is odd even by Silicon Valley standards. A commercial arm with capped profits answers to a non-profit board devoted to an esoteric safety doctrine, and investors get neither votes nor seats. Satya Nadella handled the crisis deftly, securing Altman's return and a new board. The exposure has not gone away, and nothing hedges it. A company the size of Microsoft should not rent its most important intellectual property from a non-profit run on ideology.

The directors owe their duty to the non-profit's mission, which is to make sure that artificial general intelligence benefits humanity, and owe none to shareholders. OpenAI's own documents tell investors to regard their money in the spirit of a donation and warn that the company may never make a profit. Microsoft signed up knowing those terms. The weekend showed that they are enforceable, and that four directors could put a large part of Microsoft's product plans at risk in a single afternoon.

The licence has a trapdoor

The most awkward clause concerns general intelligence itself. OpenAI says its board alone decides when such a system has been built, and that the system would then fall outside the commercial licences Microsoft relies on. The same kind of board that mishandled a personnel matter so badly would therefore rule on the question that determines much of what Microsoft's investment is worth. No accountant can put that risk on a balance sheet, and no covenant protects against it.

There is no agreed test for general intelligence, which makes the clause harder to price, not easier. A board that believed its mission required it could declare the threshold crossed early and cut Microsoft off from the models it most wants. A board anxious about safety could equally argue that commercial deployment itself breaches the charter. Either decision would be made by people with no financial stake in the outcome and, as the past week showed, a limited appetite for explaining themselves.

Leverage without control

The strongest case for Microsoft is that it held the cards anyway. Its licence lets it use OpenAI's models, those models run on its Azure cloud, and it showed that it could offer a home to the people who build them. Mr Nadella got what he wanted without casting a single vote, and investors approved: Microsoft's shares closed at a record high on the Monday it announced the hires. That leverage depended, however, on hundreds of researchers following one man out of the door, which is loyalty rather than a contract. Rebuilding the work inside Microsoft would also have cost time, and time matters while Google and others race to match GPT-4.

The loyalty had a financial edge. OpenAI's staff were reported to be weeks away from selling shares to outside investors at a valuation of about $86 billion, a sale that a collapsing company could not have completed. Employees who threatened to walk were also protecting a payday, which made their threat credible this time. A future dispute in which staff side with the board, or split, would leave Microsoft with far less to bargain with.

Owners with a different aim

Foundation ownership need not be a flaw. The Novo Nordisk Foundation controls Novo Nordisk, and its stated objectives include providing a stable basis for the commercial activities of the companies it owns. OpenAI's charter ranks the mission above the business and even commits the organisation to stop competing with, and start assisting, a safety-minded rival that gets close to general intelligence first. Shareholders in Copenhagen know what their controlling owner wants, while Microsoft has just learned that it cannot be sure.

America has its own lesson in how a charitable owner can surprise the market. In 2002 the trust that controls Hershey, set up to fund a school for poor children, announced that it wanted to sell the chocolate company, and then abandoned the plan after an outcry in Pennsylvania and intervention by the state's attorney-general. Investors who had bid up the shares learnt that the trust answered to its beneficiaries and to state officials, and to them only through the share price. OpenAI's non-profit sits in the same legal category, with the same indifference to minority capital.

What would change the risk

The new board is an interim one and is expected to grow. Microsoft's position improves if it gains a seat, even a non-voting one, or if the charter is rewritten so that directors must weigh the interests of the capped-profit company's investors. The independent inquiry into the firing that Mr Shear promised during his brief tenure may also show whether the old board had a case or merely lost its nerve. Absent those changes, the governance risk sits exactly where it sat before 17 November.

Microsoft got its man back this time. Until it has a legal say in how OpenAI is governed, its $13 billion rests on the goodwill of a board it does not control, and the next coup may end less neatly. Expect Microsoft to push for an observer seat within months and to keep building its own models as insurance; if it gets neither, the discount investors should apply to its AI story will reappear at the first sign of boardroom friction.