How Microsoft’s multibillion-dollar alliance with OpenAI really works

After UK competition officials said they were preparing to investigate Microsoft’s multibillion-dollar alliance with OpenAI last week, the San Francisco-based start-up quietly made a change to its website.

Having previously described the tech giant as a “minority owner” in OpenAI, the wording was changed over the weekend to describe Microsoft as only holding a “minority economic interest”.

The distinction may prove crucial as antitrust authorities attempt to shine a light on Silicon Valley’s most-watched — yet least understood — partnership.

The looming investigation comes after the artificial intelligence start-up was engulfed by chaos last month. Chief executive Sam Altman was fired by OpenAI’s board, only to be reinstated days later after pressure from the AI company’s employees and investors, including Microsoft chief Satya Nadella.

However, neither Microsoft nor OpenAI’s other backers — which include Thrive Capital and Sequoia Capital — own any conventional equity shareholding in the company. Instead, they are entitled to receive a share of its profits from a specific subsidiary of OpenAI, up to a certain limit.

Investors who may choose to participate in a continuing $1bn tender offer that would value OpenAI at $86bn, will get the same deal, according to people familiar with the matter.

The peculiar structure of these deals shows how the altruistic origins of Silicon Valley’s most prominent start-up are now colliding with the huge commercial promise of AI.

OpenAI was founded in 2015 as a not-for-profit organisation, which now owns the for-profit subsidiaries created in 2019 to facilitate Microsoft’s investment. But, at least until recently, many of the details of its structure have been obscure to outsiders.

People familiar with the deal say Microsoft has in total committed up to $13bn to OpenAI, including a $10bn expansion announced in January this year. But until the past week, it was never clear about what kind of return its own shareholders might expect from the investment.

The graphic detailing OpenAI’s complex ownership structure on its website was amended a few days ago without renewing the web page’s timestamp, which indicates that the last update was made in June. OpenAI confirmed the language was altered to clarify Microsoft’s position and did not reflect any change to their underlying agreement.

With regulators circling, Microsoft has also sought to reshape perceptions of its deal. “While details of our agreement remain confidential, it is important to note that Microsoft does not own any portion of OpenAI and is simply entitled to share of profit distributions,” Microsoft said on Friday last week.

Reports before, and for months after, January’s announcement of what the two companies described as a “multiyear, multibillion-dollar investment” had suggested that Microsoft could end up with a 49 per cent stake.

In fact, Microsoft’s billions — which include huge investments in data centre infrastructure as OpenAI’s “exclusive cloud provider” — entitle it to up to 49 per cent of the profit generated by a subsidiary of OpenAI, according to people familiar with the deal.

This “capped profit” subsidiary is majority-owned by a holding company, owned by OpenAI’s employees and other investors. This holding company, in turn, is owned and controlled by a charity controlled by OpenAI’s board, according to the company’s website.

Rather than a direct equity stake, “Microsoft, employees and investors own an economic interest in the for-profit, which entitles them to a share of the profits,” according to a person with knowledge of the arrangement.

OpenAI’s structure originates from its founding mission of “ensuring that safe artificial general intelligence is developed and benefits all of humanity”, rather than only a small group of employees and investors. AGI refers to software that is as intelligent as humans.

The arrangement means all OpenAI investors are subject to an upper limit on the returns they can make, according to people familiar with the matter. How and when those returns are paid out remains unclear.

For the very first investors in the subsidiary, that limit was 100 times their investment, according to a 2019 OpenAI blog post, which said the multiple would be lower for future funding rounds.

“This structure only makes sense because . . . the company thinks the upside from AGI could be so huge, that it could go beyond these profit caps,” said one person familiar with the workings of the company.

In addition to being its exclusive cloud provider, Microsoft’s deal also granted it an exclusive licence to OpenAI’s intellectual property, according to people familiar with the matter. The deal excludes any AGI that OpenAI might eventually create and only OpenAI’s board can declare when that breakthrough has been achieved.

Those IP rights could have been significant if Altman and other OpenAI employees had gone to work at Microsoft, as was proposed by Nadella before Altman’s reinstatement. “A lot of relevant technology could have gone with him,” this person said. “Microsoft is technically a competitor but has all the IP, which puts it in a stronger position.”

Microsoft declined to comment on the details of its agreement with OpenAI. It is unclear how long any exclusivity agreements may last.

Transactions involving non-profit organisations are often exempt from the usual pre-merger notification requirements under the US Hart-Scott-Rodino Antitrust Improvements Act, which may have helped to shield Microsoft’s investment from regulatory scrutiny.

However, the US Federal Trade Commission is now looking into Microsoft’s investment in the company, according to a person familiar with the matter, although no formal investigation has been launched.

Any US probe and that of the UK Competition and Markets Authority, which is expected to begin formally next year, are likely to hinge on whether Microsoft’s influence over OpenAI has grown in recent months, either as a result of January’s expansion of their alliance or last month’s boardroom bust-up.

Ever since the launch of ChatGPT in November last year, OpenAI has been divided internally over its commercial direction, including which products to pursue and invest in.

According to one person close to the leadership’s deliberations, some OpenAI executives felt pressured by Microsoft to commercialise faster, as it “really wanted to be deploying AI products and be an AI leader and make money”.

During the leadership crisis at OpenAI last month, Microsoft was given no warning from the board about the coming ructions and raced to re-establish stability. “They wanted more visibility, but everyone realised that them having an actual board seat would be a non-starter,” this person said.

It was eventually agreed that Microsoft would take a non-voting observer role on a reconfigured board. This would “certainly give [Microsoft] some level of additional influence”, this person added, but suggested that would be a positive step for the start-up’s governance. “There would be more of an expectation that things at OpenAI be done by the book.”