For years, Microsoft executives have been accustomed to negotiating complex partnership agreement terms with OpenAI. The original contract granted Microsoft exclusive rights to sell OpenAI's artificial intelligence technology to cloud customers.
Two months ago, the developer of ChatGPT unexpectedly announced a partnership allowing the sale of AI services to enterprises through Amazon, Microsoft's primary competitor. OpenAI maintained that its contract with Microsoft permitted this arrangement, while Microsoft contested this view. According to multiple sources familiar with high-level communications between the two companies, diverging interests emerged as OpenAI's revenue surged, prompting Microsoft CEO Satya Nadella and OpenAI CEO Sam Altman, long-time allies, to engage in several rounds of private discussions in recent weeks.
With the involvement of Microsoft's Business Development President Jon Tinter and both companies' CFOs, Sarah Friar and Amy Hood, the two CEOs finalized a plan to avoid potential legal disputes, making mutual concessions on long-standing demands.
The revised agreement, announced on Monday, permits OpenAI to deepen its collaboration with Amazon and, for the first time, authorizes other cloud providers like Alphabet to sell its large models to enterprises. This grants OpenAI operational flexibility comparable to its competitor Anthropic, which already offers AI services across Amazon, Alphabet, and Microsoft's cloud platforms.
In exchange, OpenAI relinquished a key counterbalance clause from the original contract: the right to terminate Microsoft's exclusive rights to sell its top-tier AI products upon achieving Artificial General Intelligence (AGI), a vague technological milestone. This AGI clause, included in the 2019 agreement, had long concerned Microsoft management, who feared OpenAI could trigger it and disrupt the existing partnership.
OpenAI also committed to paying Microsoft 20% of its total revenue until 2030, regardless of whether AGI is declared, with the total payment amount subject to an undisclosed cap. Sources indicate this revenue share covers all enterprise income generated through third-party cloud channels like Amazon and Alphabet. Concurrently, Microsoft is no longer required to pay OpenAI a 20% share from its own cloud sales of OpenAI models.
Under the previous agreement, the revenue-sharing mechanism would terminate immediately if OpenAI announced AGI achievement.
The core amendments reflect both companies' willingness to make key concessions, removing restrictions that could have significantly hindered OpenAI's plans for a public listing. As one of OpenAI's largest shareholders, with a stake valued at least $135 billion, Microsoft stands to gain substantial financial returns from a successful OpenAI IPO.
The new agreement also signals a shift in the deeply integrated partnership, initially formed to give Microsoft a dominant position in AI. OpenAI is now engaging commercially with multiple competitors of Microsoft simultaneously.
Amazon CEO Andy Jassy described the new Microsoft-OpenAI pact as "intriguing" in a social media post on Monday. He announced that Amazon Web Services (AWS) will soon offer OpenAI's large models to customers and launch jointly developed enterprise automation AI agent products.
OpenAI forecasts revenue of approximately $284 billion by 2030. At the 20% share rate, this could result in a payment exceeding $56 billion to Microsoft, significantly higher than this year's estimated $6 billion. However, the final amount remains uncertain due to the cap. The agreement also reportedly eliminates OpenAI's previous option to defer revenue share payments, a cash-flow optimization mechanism.
One core clause remains unchanged: Microsoft retains perpetual intellectual property licensing rights if OpenAI acquires other software or application companies. This IP-sharing obligation previously obstructed OpenAI's attempted acquisition of AI code provider Windsurf last year.
This revision marks the third amendment to the partnership terms within a year and represents a significant turning point in the collaboration that produced ChatGPT and reshaped the global AI landscape.
The partnership began in 2019 when Microsoft, aiming to catch up with Amazon in cloud computing, invested $1 billion in OpenAI's newly formed for-profit entity. Elon Musk, an early donor to the non-profit OpenAI research lab founded in 2015, has alleged that both companies violated the original agreement since the for-profit arm's creation; his lawsuit against them is scheduled for trial this week.
Following ChatGPT's viral success in early 2023, which ignited a global AI model race, Microsoft invested an additional $10 billion in OpenAI.
In return, the companies signed a complex revenue and technology-sharing agreement later that year, deepening their ties. Microsoft gained preferential access to advanced AI technology and the right to resell it to enterprise customers. It secured its position as OpenAI's exclusive cloud provider and accepted a unique profit-participation structure giving early investors like Microsoft priority on initial profit distributions.
However, tensions soon emerged. OpenAI's demand for data center servers surged beyond Microsoft's computational capacity delivery, compounded by disagreements over various 2023 agreement terms. Renegotiations began in the fall of 2024, focusing on converting the profit-sharing mechanism to a standard equity structure.
Between 2024 and 2025, Microsoft gradually eased restrictions, allowing OpenAI to rent cloud servers from Amazon, Alphabet, and Oracle, and to build its own data centers.
Last autumn, OpenAI restructured, shifting from a profit-sharing entity to a standard equity model, a move Microsoft permitted. OpenAI gained autonomy to use third-party cloud compute without prior approval. Concurrently, Microsoft began procuring Anthropic's models for its Copilot suite and reselling them on the Azure platform.
Microsoft, however, retained exclusive rights to sell OpenAI's cloud services. This clause reportedly caused dissatisfaction among OpenAI management, as competitor Anthropic operated without such constraints, rapidly gaining enterprise market share with a multi-platform strategy. Microsoft later adjusted its sales incentives, offering equal commissions for promoting both Anthropic and OpenAI models.
An unexpected dispute arose from a clause added in a fall 2025 supplemental agreement, permitting OpenAI to "co-develop products with third parties." According to an OpenAI blog post from October that year, Microsoft retained exclusive rights to sell stateless, general-purpose models via API, while non-API products could be offered across all cloud platforms.
Some Microsoft executives initially overlooked this clause, but OpenAI's management saw it as an opportunity, secretly collaborating with Amazon to develop a new product: AI agents for enterprise automation, fully hosted on AWS.
On the day the Amazon partnership was announced, a joint statement clarified that Microsoft Azure would exclusively offer stateless, general-purpose OpenAI models via API. Simultaneously, OpenAI announced a "stateful" AI product co-developed with AWS, focused on enterprise automation and custom agent development.
Stateless API services provide standardized model access, while stateful AI can retain personalized content like enterprise information, business data, and conversation history.
Sources revealed that OpenAI and AWS deliberately labeled the product "stateful," surprising Microsoft management, as it operated entirely outside the Azure ecosystem, bypassing Microsoft's exclusivity. Amazon further committed up to $50 billion in potential investment for expanded server leasing, featuring the large-scale deployment of its custom Trainium AI chips.
Many Microsoft executives questioned the announcement's compliance with existing contracts. Weeks later, Microsoft stated publicly that it believed OpenAI was "fully aware of and adhering to its legal contractual obligations."
The escalating dispute led to multiple emergency meetings between Nadella and Altman, aimed at resolving the legal risks.
A key factor enabling successful negotiations was Microsoft's decreasing reliance on OpenAI's technology. Microsoft has since integrated Anthropic's models extensively into its Copilot suite for tasks like meeting scheduling and Excel financial modeling, with Anthropic's models sometimes outperforming in specific scenarios.
Nadella has recently downplayed the importance of any single model, emphasizing practical application and deployment as the core focus.
"The industry has entered a phase of model surplus; technical capabilities are mature. The central challenge now is effective implementation," Nadella remarked in a recent podcast interview. "That's why I prioritize real-world product impact over mere model parameters."
Comments