Sources on the key Microsoft-OpenAI clause: OpenAI decides when it hits AGI and “sufficient AGI”, able to make $100B+ in profit, and Microsoft can't build AGI
A key clause in Microsoft and OpenAI's deal embodies the raging divide between AGI true believers and those who think it's still a long ways off.
Context & Ripple Effects
The $100B-plus-profit threshold had already surfaced in reporting on the parties’ undisclosed 2023 agreement, tying an ostensibly technical AGI milestone to a commercial test rather than a purely capability-based one. This account adds that OpenAI, not Microsoft, is positioned to make that determination.
The clause had become a live negotiating fault line: Microsoft was reportedly seeking to remove the provision that could restrict its access, while OpenAI had discussed changing it to preserve investment. The reported ban on Microsoft building AGI makes the dispute about strategic control, not merely licensing terms.
First-order effects
- OpenAI retains contractual leverage to determine whether the AGI and “sufficient AGI” conditions have been met, including the profit-based threshold described in the agreement.
- Microsoft’s access to OpenAI’s frontier technology—and its ability to pursue AGI itself—is constrained by a clause controlled by its principal AI partner.
Second-order effects
- The arrangement gives Microsoft a direct incentive to renegotiate the trigger and governance terms, consistent with its reported push to eliminate the AGI clause.
- Commercial performance becomes consequential to technology rights: the $100B-plus-profit definition can turn revenue and profitability into a trigger for access restrictions, rather than leaving the issue solely to technical evaluation.
Third-order effects
- If such provisions persist, frontier-AI partnerships may increasingly separate capital and infrastructure providers from control over the most advanced models, making contractual access a strategic asset in its own right.
- The dispute illustrates how “AGI” can function as a negotiated legal and commercial boundary, not a settled scientific benchmark; later reporting indicated the access-rescission clause remained in a modified form, suggesting the governance question outlasts any single contract revision.
The trend: AI partnerships are evolving from straightforward investment-and-compute arrangements into negotiations over who controls access, definitions, and commercialization once models cross strategically significant thresholds.