News

Microsoft and OpenAI restructure their partnership: extended IP rights, $250B Azure commitment, and AGI verification panel

Oct 28, 2025

Key Points

  • Microsoft extends IP rights through 2032 including post-AGI models, while OpenAI commits to $250 billion in Azure services over time, structurally loosening compute exclusivity.
  • OpenAI gains freedom to release open-weight models, partner with third parties on non-API products, and serve US national security customers on any cloud provider.
  • An independent expert panel will verify AGI declarations, but AGI's definition remains contested and panel composition unspecified, leaving key terms open for renegotiation before OpenAI's potential IPO.

Summary

Microsoft and OpenAI have restructured their partnership with material concessions on both sides, clearing the path for OpenAI's conversion to a for-profit entity and eventual public offering.

Microsoft's IP rights extend through 2032 and now include models developed post-AGI, subject to safety guardrails. Research IP, the confidential methods used in model development, remains protected until either an independent expert panel verifies AGI or through 2030, whichever comes first. Microsoft's research IP rights exclude model weights, inference code, fine-tuning code, and data center hardware or software. This distinction preserves Microsoft's ability to build competing systems using non-research IP.

OpenAI has committed to purchasing an incremental $250 billion in Azure services. The revenue share agreement that sparked earlier private skepticism from venture investors persists until AGI is verified, with payments stretched over a longer period.

OpenAI can now develop products jointly with third parties. API products built with those partners must run on Azure exclusively. Non-API products can run anywhere. OpenAI also gains the right to release open-weight models meeting certain capability thresholds and to provide API access to US national security customers regardless of cloud provider. Microsoft loses its right of first refusal to be OpenAI's compute partner.

An independent expert panel will verify any AGI declaration. The definition of AGI itself remains contested. Tyler Cowen recently argued it has already been achieved, while others point to "spiky intelligence" and the need for broader capabilities before true general intelligence arrives. No panel composition is specified.

If Microsoft develops AGI using OpenAI's IP before the panel declares AGI, the resulting models face compute thresholds significantly larger than current leading systems. This mechanism gives Microsoft a path to AGI R&D that does not trigger immediate IP conflict, but only if the system is substantially larger than what exists today.

The deal resolves the corporate structure complexity that had hung over OpenAI since its 2019 nonprofit founding. The conversion from LLC to Public Benefit Corporation, necessary for a future IPO, had stalled partly because this partnership agreement needed updating. Investors like Josh Kushner at Thrive had deployed capital into OpenAI's for-profit subsidiary despite structural uncertainty. This agreement removes that obstacle.

Many terms appear interim. Definitions of research IP, thresholds for open-weight release, AGI verification criteria, and the mechanics of product partnerships all leave room for renegotiation before OpenAI goes public. The deal provides a pathway forward, not a permanent settlement.