OpenAI and Microsoft prove ‘breaking up isn’t hard to do’

Home AI Infrastructure News OpenAI and Microsoft prove ‘breaking up isn’t hard to do’

A multibillion-dollar makeover clears the way for SoftBank’s $22 billion investment, new funding rounds, and a possible IPO in the future

In sum, what to know:

More freedom: OpenAI is freed from the constraints of the capped-profit model that scared away some investors, and Microsoft is freed from the constraints of the “AGI clause” that stifled its pursuit of AGI.

For-profit arm: OpenAI’s non-profit arm remains as the OpenAI Foundation, while its new for-profit arm, OpenAI PBC, is free to pursue fundraising and a corporate structure more conducive to an IPO.

Microsoft valuation boost: With a 27% stake in the newly created OpenAI PBC, which is said to be worth about $500 billion, Microsoft’s market capitalization surpassed $4 trillion.

In a seemingly amicable conclusion to a high-profile dispute with Microsoft, OpenAI is both a nonprofit and a for-profit, with its non-profit OpenAI Foundation having a 26% stake in the for-profit OpenAI Group PBC. That amounts to about $130 billion of the PBC’s current $500 billion valuation. Additionally, profit caps for investors will be replaced with conventional equity stakes, enabling future liquidity exits. Those changes may help attract the type of larger-scale funding that advanced AI development requires.

Not only is it remarkable that OpenAI is now free from the shackles of the previous capped-profit model, but also that Microsoft is free to independently develop AGI, leveraging OpenAI’s research. The previous agreement’s clause said Microsoft had the right to use OpenAI’s technology “until OpenAI achieves AGI.” That determination was previously going to be left to OpenAI’s board. Now, the existence of AGI will be determined by an independent panel — the composition of which is unknown, as is the criteria by which the panel will determine if AGI has been achieved. Regardless of if and when CGI is achieved, Microsoft’s exclusive rights to pre-AGI research and models will be extended through 2032, however, its 20% revenue share with OpenAI will cease at the point AGI is achieved. This makes this deal truly unique and groundbreaking, as well as confusing.

Another component of the new agreement is that OpenAI’s API products will remain a core part of Microsoft Azure OpenAI Service, but that OpenAI will be able to use other cloud providers for non-API products (as has already begun with its recent $300 billion compute and data center deal with Oracle).

In addition, Microsoft will continue to be one of OpenAI’s biggest financial backers, with a 27% stake in the OpenAI Group PBC. Some say this deal gives Microsoft 10x what it invested in OpenAI, with the for-profit transition pushing Microsoft’s valuation beyond $4 trillion. To give context, Microsoft invested $1 billion in 2019 and another $10 billion in 2023. In addition, Microsoft provided OpenAI’s compute for its large-scale AI models — a grip that loosened with OpenAI’s recent deals with Oracle, as well as Google and others involved in the Stargate project.

For OpenAI, the deal also clears the way for the $22 billion investment by Softbank, just one example of the doors that will potentially open now that OpenAI can move toward a more traditional corporate structure to attract the capital needed to scale its operations. That could move it toward a possible IPO — something that wasn’t really possible under the capped-profit structure.

Going public might be the best chance OpenAI has to close the gap between what it consumes and what it earns. Earlier this week, CEO Sam Altman revealed during a public livestream that OpenAI has $1.4 trillion in financial obligations to use or develop 30 GW of data center capacity. Though that is a projection and not an actual debt, Altman did allude to big infrastructure spending on the horizon, stating during the livestream that he’d like “OpenAI to add 1 gigawatt of compute every week,” which by some estimates would mean a capital cost of more than $40 billion.

Hyperbole aside, $115 billion in expenditures are expected by 2029, due to the expected AI infrastructure costs to power ChatGPT. How does that measure up? OpenAI revenues for this year will be around $13 billion, and some reporting indicates OpenAI wouldn’t be profitable until 2029, when it’s predicted the company could make $125 billion in revenue. At that point, OpenAI’s projected annual revenue could outpace its massive compute and development costs. It depends on whether enough individuals, enterprises, and institutions will be willing to eventually pay for ChatGPT. Of its 800 million users, only about 5 million business users and 20 million individuals currently pay for the service. With this new restructuring, it’s possible OpenAI could rapidly develop new ways to monetize, and possibly make its models far more efficient, and hence less costly, than they have been.

What you need to know in 5 minutes

Join 37,000+ professionals receiving the AI Infrastructure Daily Newsletter

This field is for validation purposes and should be left unchanged.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More