Message from Copenhagen: Microsoft Has de Facto Control of OpenAI
I recently came across an interesting article by Frederik Hovmark Pedersen, a scholar from Copenhagen Business School. It caught my attention because it supports a conclusion — or suspicion, I guess — that I’ve been reluctantly articulating recently. Until now I made the assertion, without much evidence or scholarly support, that profit was controlling charity in OpenAI’s joint venture with Microsoft and other private investors. My evidence was mostly intuition mixed with cynicism. But Pedersen states that despite OpenAI’s formal adherence to nonprofit governance, “de facto control resides with Microsoft, undermining the formal governance structure. Moreover, the case demonstrates that foundation-like ownership can, under certain circumstances, be vulnerable and might not be a good fit for a startup like OpenAI with significant capital requirements and enormous commercial potential.”
The inmates have taken over a highly profitable charitable asylum. What Pedersen is really saying — the sad part that is most relevant to this blog — is that jot and tittle compliance with Revenue Ruling 98-15 will not stop profit-takers from taking control of a joint venture. Maybe the Service was right all along about human nature. Years ago, it was categorial that profit and charitable tax exemption could not coexist in a single entity. The instinct to pursue profit is hard-wired. It will always prevail over the instinct for charity. I was hoping otherwise. Silly of me.
In November 2023, the Board of Directors of OpenAI fired its popular CEO, Sam Altman, because it had lost confidence in him, and concluded that he was hindering the board’s ability to properly exercise its duties. The decision outraged employees and investors, including Microsoft which has a $13bn investment in OpenAI, and ignited a brief governance crisis. After mounting pressure from these key stakeholders, the crisis ended with Sam Altman returning as CEO and most board members resigning. The turmoil brought the importance and impacts of corporate governance to the attention of business and startup communities alike, and drew attention to OpenAI’s somewhat unusual ownership structure: a self-owning and self-governing non-profit organisation controlling a for-profit company, where investors have a contractual claim on a portion of the profits but no formal control, and the ultimate decision-making authority is the board of the non-profit, which appoints and removes itself, has no personal financial interest, and has a fiduciary obligation to the non-profits mission of building general-purpose artificial intelligence that safely benefits humanity. This arrangement resembles the ownership found at enterprise foundations, and thus provides an interesting opportunity to study this relatively rare type of organisational control. This case therefore seeks to understand corporate governance at OpenAI and the turmoil in November 2023, to deduct what other companies and boards can learn from it. The case shows that although corporate governance of OpenAI has been intentionally designed to advance the mission and protect this from financial obligations, in particular the short-term interests of investors, de facto control resides with Microsoft, undermining the formal governance structure. Moreover, the case demonstrates that foundation-like ownership can under certain extreme circumstances be vulnerable, and might not be a good fit for a startup like OpenAI with significant capital requirements and enormous commercial potential.
It might not yet be true that Microsoft is really in control, though only an optimist would deny what seems inevitable. But Pedersen’s observation supports a notion I’ve previously entertained. He almost makes the case that OpenAI employees — those with such rare skills that they have OpenAI by the short hairs — might even be disqualified persons. Those employees have profit interests beyond fixed salaries. Their stock in the joint venture essentially makes them private investors in OpenAI. Here is the lesson to be learned from the joint venture, according to Pedersen:
Bearing in mind that we do not know what the results and effects of the recently filed lawsuit against OpenAI and Sam Altman will be, observers can still deduct several lessons from the OpenAI case. The first being that formal control does not necessarily equate to actual control. The board had the right to fire Sam Altman, perhaps even the obligation to do so, but how the events unfolded afterwards show that they were not in control of the situation. The reason being that they effectively did not control OpenAI’s two most valuable assets: its employees and its technology. Most employees turned out to be loyal to Sam Altman rather than the company, and prepared to leave if he was not reinstated, providing the important lesson for firms relying on highly skilled labour while possessing few physical assets, such as many tech companies, that this grants employees’ significant leverage. Meanwhile, Microsoft to a large extent controls the technology by providing the capital and infrastructure necessary to develop and run OpenAI’s models and having exclusive licensing rights to the IP, creating a dependency which means that although formally the board of OpenAI is not accountable to Microsoft, it would appear that informally it is. Here is a lesson for non-profits and for-profits alike in the risk of becoming overly dependent on one single stakeholder.
darryll k. jones