Why OpenAI’s non-profit dream collapsed: FT lays out the 2017 turning point

TL;DR:

• Disclosed OpenAI emails in the ongoing Musk v Altman trial trace the precise moment — Google’s 2017 GPU-cluster breakthrough — when the non-profit AGI model became commercially untenable.
• The Ilya Sutskever email to Musk explaining the shift is the document at the centre of the FT’s analysis: combining GPU chips at scale meant a single large experiment, not many small ones, became the path to frontier progress — and money became vastly more important than non-profit talent advantages.
• OpenAI’s operating company is now valued at $852 billion (£640 billion) and is reportedly preparing for an IPO; the original non-profit retains around 20% of shares and the right to hire and fire directors, though the 2023 board ouster of Altman tested how meaningful that power is in practice.

The FT’s Simon Mundy uses the trial’s disclosed documents to reframe the standard “OpenAI sold out” narrative as a more specific technological one. The 2015 founding statement — drafted jointly by Musk and Altman — promised AGI development “unconstrained by a need to generate financial return”. Sutskever’s 2017 email explains why that became impossible: Google’s chip-clustering capability turned single, expensive, hardware-intensive experiments into the only viable research path, and OpenAI’s non-profit structure could not raise the capital to compete.

The 2018 inflection that produced today’s structure

By early 2018, Musk had concluded OpenAI would have to switch to a for-profit model to stay in the race with Google, and proposed a Tesla takeover. The OpenAI board declined, Musk cut formal ties, and the entity began the restructuring trajectory that has now produced the for-profit-with-non-profit-supervisory-board model. Most recently, October’s rejig scrapped the investor return cap that had been designed in 2019 to reassure on mission alignment — a quiet but decisive change.

UK angle: pension funds and AGI exposure

For UK readers, the practical thread runs through the IPO. OpenAI’s investors now include Nvidia, Amazon, SoftBank and Microsoft, with Microsoft the single largest shareholder. UK institutional investors hold material indirect exposure through Microsoft equity already, and several UK pension and sovereign-style funds have been approached on pre-IPO secondary lines. A Musk win in the current trial — or even a partial verdict on restructuring fairness — would force disclosure adjustments that could shift IPO timing and pricing. UK pension trustees considering the prospectus, under the AI-trustee duty-of-care framework that TPR set out this month, will be reading the FT’s analysis carefully.

Looking forward

Closing arguments in the Oakland trial are set for this week, with a nine-person jury deliberating shortly after. The deeper question the FT raises — whether existing corporate forms can govern technologies that demand vast resources and carry significant public consequences — is the live policy question the King’s Speech Regulating for Growth Bill set out in this week’s UK package will eventually have to answer. The non-profit experiment is over; the alternative governance models are still being designed in real time.