By far the most finicky part of OpenAI’s necessary conversion into a for-profit company was reconciling its convoluted partnership with Microsoft Corp. With the year-end deadline fast approaching, they have made a deal both sides can live with for now, though it sets out the timeline for an eventual split.
Microsoft’s shrewd $13.75 billion investment in 2019 required OpenAI to work with Microsoft and its Azure cloud platform — an arrangement that sent Microsoft’s valuation soaring. But as the artificial-intelligence race intensified, things became untenable: OpenAI needed more computing power than Microsoft was able to build, and Sam Altman, OpenAI’s co-founder and chief executive officer, wanted greater freedom to raise the money to do that. Along the way, frictions between the two companies, and questions about how their two cultures meshed, were starting to emerge. It’s hard to be both partners and competitors at the same time.
By becoming a public benefit corporation, OpenAI hasn’t overcome these challenges, but it has certainly simplified them. Tranches of additional funding from SoftBank Group Corp. can now be accessed, and OpenAI now a has clearer path to going public some day (now that’s an S-1 I’d like to see). It can more freely court the services of other cloud providers without giving Microsoft right of first refusal. It can build “some products” with other companies, reducing its reliance on Microsoft’s questionable ability to smartly incorporate ChatGPT into Office 365, Bing and other products. In other words, the bottlenecks are gone — but then, so are the excuses.
For Microsoft’s part, it has negotiated an agreement that keeps the access its investors will care about while protecting itself against the Altman hype machine. Its stake in the “new” OpenAI is $135 billion. OpenAI will spend $250 billion on Azure cloud computing over an undefined time frame. Microsoft has retrained its intellectual property rights for OpenAI’s models and (non-hardware) products through 2032, even if artificial general intelligence is achieved before then. It retains access to OpenAI’s research until the end of 2030, or when (if) AGI arrives — whichever is sooner.
Here is where we find the most interesting part of the deal and perhaps a foreshadowing of a truly profound argument to come. Even if OpenAI declares it has built human-level intelligence, the related clauses with Microsoft will kick in only after the claim has been “verified by an independent panel.” Deciding who is on this panel, and what benchmarks the members should use, isn’t going to be a small task. How do you assemble credible AI experts without an impossible tangle of conflicts of interest? Staffing and expertise within the AI industry is notoriously incestuous, and it’ll be some task to find those without a strong interest in seeing OpenAI succeed or fail. Declaring OpenAI to have achieved AGI would likely send the values of its competitors plummeting if Wall Street takes the view that AGI might be a winner-takes-all situation, as some suggest it will.
