OpenAI has partially unwound its exclusive collaboration structure with Microsoft (MS), allowing the company to offer its artificial intelligence (AI) models on other clouds such as Amazon Web Services (AWS).
On the 27th (local time), OpenAI and MS said they amended their existing agreement to convert MS's exclusive license to use AI models into a nonexclusive license. As a result, OpenAI can expand supply of the GPT models, which had been available only on MS Azure, to AWS, Google Cloud, and others.
However, OpenAI will keep MS as a primary cloud partner. Except where MS cannot support necessary features, OpenAI products will launch on Azure first.
MS will continue to hold intellectual property (IP) licenses for OpenAI's models and products through 2032, as before. However, the license terms will shift from exclusive to nonexclusive. In return, MS will no longer make separate payments for revenue from OpenAI models sold through its own distribution channels.
The condition that OpenAI will allocate a certain share of its model and product sales revenue to MS through 2030 was also maintained. The share is known to be around 20%. Notably, the previous clause that would have ended revenue sharing if OpenAI achieved "artificial general intelligence (AGI)" before 2030 was deleted. AGI refers to general intelligence at or above the human level, but the lack of a clear definition has fueled ongoing debate.
The revision is seen as resolving friction that arose during recent efforts to pursue collaboration between OpenAI and AWS. Earlier, OpenAI sought to offer the enterprise AI service "Frontier" on AWS infrastructure, and MS raised the possibility of a contract breach and said it was even considering legal action.
Immediately after the revision, AWS announced plans to offer OpenAI models to customers on its cloud platform. AWS CEO Andy Jassy said on the professional networking site LinkedIn that "there was a very interesting announcement from OpenAI this morning," adding that "in the coming weeks, customers will be able to use OpenAI models directly in Bedrock," AWS's AI model platform.
Experts say MS, too, could reduce infrastructure burdens and improve resource allocation efficiency under the amended deal. Analysts at the U.K. financial group Barclays told Reuters, "MS no longer needs to build every data center required for OpenAI, allowing it to invest more in Copilot and other areas."
MS has been investing in OpenAI since 2019, more than three years before OpenAI launched ChatGPT. When OpenAI restructured last October into a public benefit corporation (PBC) capable of pursuing profit, MS was recognized with an equity stake of about 27% and has maintained its position as the second-largest shareholder after the OpenAI Foundation.