OpenAI and Microsoft – Microsoft on Top

The water becomes much clearer.

• The signing of an MoU between OpenAI and Microsoft represents a step forward in resolving the corporate governance issues that are plaguing OpenAI, and the endgame is becoming much clearer.
• I have long been concerned with how OpenAI is run (see here), as there are huge conflicts of interest between the non-profit arm and the for-profit subsidiary.
• There are also provisions in the current agreement that are contentious lawsuits waiting to happen.
• This is why there have been fraught negotiations for some time (see here) between the two companies, but it looks now as if a broad agreement has been reached.
• A lot more needs to be worked out in terms of what Microsoft has access to and what exclusivity it gets, but the big disagreements appear to have been settled.
• This paves the way for OpenAI to become a properly run company and potentially to go public at some point in the future if it ever makes any money.
• The non-binding MoU itself has not been made public, but one of the key terms is that the non-profit part of OpenAI will end up with a stake worth around $100bn (20%-30%), which, crucially, implies that it will have no control over the company.
• Instead of a for-profit company labouring with a non-profit conflicting entity sitting on top of it and controlling it, the for-profit company will be free and have the non-profit as a large shareholder.
• Microsoft will probably end up with a similar but somewhat larger stake, a say in how the company is run, and certain access to the company’s technology.
• This should remove almost all of the conflicts that have been bothering me since the near-implosion in 2023 (see here), reducing the risks of implosion down to running out of money, just like everyone else in this space.
• It will also free up management to get on with running the company and developing its product instead of fighting with Microsoft, which is something it badly needs, given how the competition is snapping at its heels.
• The deal is far from closed, but the final arrangement is expected to be completed by the end of the year.
• This is crucial as this is the deadline to receive the rest of the $40bn investment from SoftBank, which the company badly needs to continue building compute capacity in the form of AI datacentres.
• Given the deal that OpenAI has signed with Oracle and its commitments elsewhere, this will barely scratch the surface, but it is crucial to the company nonetheless, given that it is due to arrive soon.
• Hence, I am confident that the deal will be hammered out and signed on binding terms comfortably before the end of the year and that SoftBank’s billions will arrive.
• The net result is that I am now far less concerned with the stability of OpenAI, as the conflicts of interest and its fights with Microsoft will have been solved.
• This means that it will be business as usual, although I am pretty sure that Microsoft will have won substantial concessions from OpenAI as it is holding all of the cards in this negotiation (see here).
• This also removes a great deal of risk from Microsoft, as it is now much safer to base its technology on OpenAI, as the risk of non-financial implosion has been reduced to almost zero.
• I see the financial risk as almost zero, as in the worst-case scenario, Microsoft could bail OpenAI out or purchase the company outright, which I suspect is still an option.
• This could occur when the current frenzy dies down and OpenAI has much more difficulty in raising money, meaning that it can no longer meet its contractual commitments.
• It is at this point that Microsoft (or SoftBank, although I don’t think it could afford it) could step in and buy up OpenAI at a discount to the last round, as very often happens in these sorts of corrections.
• Hence, if there is no correction (unlikely) the road is now open to IPO, but I think that the most likely situation is that OpenAI gets into a cash crunch and is purchased by Microsoft.
• I think that the same will happen with Anthropic, Mistral et al, underlining once again the best place to be positioned in AI is in the picks and shovels trade (see here), which is exactly where I remain.