Tensions between OpenAI and Microsoft are rising, as multiple reports have suggested in recent weeks. According to a recent report from the Information, the Artificial General Intelligence (AGI) clause in OpenAI’s agreement with Microsoft has become a source of friction, threatening what has so far been the most consequential partnership in AI.
The clause states that if OpenAI achieves AGI, Microsoft would lose access to OpenAI’s most advanced technologies. This provision was initially intended to prevent any single commercial entity from monopolising AGI and to keep control with OpenAI’s nonprofit board. However, Microsoft views continued access as crucial for its own business, particularly as it strives to keep pace with its rivals, such as Google.
According to a report by the Financial Times, Microsoft plans to continue utilising OpenAI’s technology under its existing commercial agreement, which is set to expire in 2030.
The definition of AGI is itself uncertain and vague. With leaked documents indicating it might be linked to an AI system generating $100 billion in profit, it remains a shifting target that could be influenced by either side.
Meanwhile, OpenAI is seeking to restructure its for-profit arm, the unit behind ChatGPT, with the eventual goal of going public. But Microsoft, its largest external shareholder, has resisted approving the move unless OpenAI removes a provision that allows it to cut off Microsoft’s access to its models upon achieving AGI.
Microsoft has invested approximately $13 billion in OpenAI since 2019.
However, OpenAI has since expanded its investor base. In October 2024, it closed a record-breaking $6.6 billion funding round from leading investors at a valuation of $157 billion. The round was led by Thrive Capital, which contributed $1.6 billion. It included participation from Microsoft, Nvidia, SoftBank, Khosla Ventures, Altimeter Capital, Fidelity, Tiger Global, and MGX, a fund backed by the United Arab Emirates.
Most recently, SoftBank has agreed to invest up to $40 billion at a valuation of $300 billion in OpenAI, with $10 billion expected to be syndicated to co-investors. However, the funding is contingent on OpenAI completing this restructuring.
If Microsoft does not approve the change by the agreed deadline (end of 2025), OpenAI risks losing billions in committed capital from these investors. That said, in all possibility, other investors may come to its rescue.
OpenAI Moves to Google Cloud
At the same time, OpenAI has already started exploring other options. According to a recent report, the startup has begun running its models on Google Cloud.
OpenAI’s decision to start using Google’s TPUs, rather than relying solely on NVIDIA chips through Microsoft and Oracle, marks an apparent effort to reduce costs and diversify its computing partners.
According to the report, OpenAI is renting TPUs through Google Cloud to handle inference workloads, as its user base and corresponding compute needs continue to grow. This not only threatens Azure’s AI revenue but also signals to the market that Microsoft’s infrastructure may not be the default choice for leading AI workloads.
At Google Cloud Next 2025, the company unveiled Ironwood, its seventh-generation Tensor Processing Unit (TPU), explicitly designed for inference. It’s a key part of Google’s broader AI Hypercomputer architecture.
Ironwood will be available to Google Cloud customers later this year, the tech giant said. It currently supports advanced models, including Gemini 2.5 Pro and AlphaFold.
Altman Plays it Down
In a recent podcast, OpenAI CEO Sam Altman addressed recent speculation about tensions between OpenAI and Microsoft.
He acknowledged that, like in any deep partnership, there are occasional points of friction. However, he downplayed the idea of a serious rift, describing a recent call with Microsoft CEO Satya Nadella as “super nice” and focused on planning their long-term collaboration.
However, it is no secret that OpenAI’s business is hurting Microsoft. For instance, last year, Amgen announced it would roll out Microsoft’s Copilot AI assistant to 20,000 employees, marking a high-profile validation of Microsoft’s significant investment in generative AI.
However, just over a year later, Amgen staff have primarily shifted to using OpenAI’s ChatGPT. The company expanded its use of ChatGPT earlier this year, citing improvements in the tool and positive feedback from employees who found it helpful for research and summarising complex scientific content.
“OpenAI has done a tremendous job making their product fun to use,” said Sean Bruich, Amgen’s senior vice president. While Copilot remains a valuable tool, he said, its primary strength lies in its integration with Microsoft applications, such as Outlook and Teams.
OpenAI and Microsoft are also reportedly in a standoff over the terms of OpenAI’s $3 billion acquisition of the coding startup Windsurf. Under their existing agreement, Microsoft has full access to OpenAI’s intellectual property and offers GitHub Copilot, its own AI-powered coding assistant that rivals OpenAI’s products.
OpenAI, however, is pushing to keep Windsurf’s IP out of Microsoft’s reach.
OpenAI’s revenue for 2024 was approximately $3.7 billion, with significant growth projected for 2025, reaching around $11.6 billion. Meanwhile, Microsoft recently informed shareholders that it is generating over $13 billion in annualised AI revenue.
Currently, OpenAI is obligated to share 20% of its revenue with Microsoft under an agreement that runs through 2030. This revenue-sharing arrangement is expected to decrease to 10% by the end of the decade, as OpenAI has informed investors of plans to reduce the share allocated to Microsoft and other commercial partners
Microsoft is also playing it safe. To hedge its bets, it has begun testing models from OpenAI rivals, including Anthropic, xAI, DeepSeek, and Meta, as potential alternatives to its Copilot tools, which are embedded in products such as Windows and Edge.
At the same time, the company is also internally building its in-house LLMs under the leadership of AI chief Mustafa Suleyman.
Microsoft’s Chip Efforts Are Failing
While OpenAI turns to Google’s TPUs, Microsoft is struggling to keep pace with its in-house AI chip development.
Microsoft’s upcoming Maia AI chip, internally code-named Braga, is reportedly facing a setback of at least six months, pushing its large-scale production timeline from 2025 to 2026, according to The Information, which cited three individuals familiar with the project.
When Braga eventually enters production, it is expected to significantly lag behind NVIDAI’s Blackwell chip in terms of performance. Blackwell began rolling out in late 2024 and has set a high bar for AI hardware.
Initially, Microsoft intended to deploy Braga in its data centres this year. However, unexpected design changes, limited staffing, and high employee turnover have all contributed to the delay, the report added.
Looking Ahead
Despite public optimism, the Microsoft-OpenAI partnership is at a pivotal moment. Altman’s discussions with Nadella and even US President Donald Trump, who announced a $500 billion AI infrastructure initiative involving OpenAI, suggest efforts to stabilise the relationship.
Notably, Microsoft was absent from the January White House event, where Altman, joined by SoftBank and Oracle, announced OpenAI’s Stargate data centre initiative.
OpenAI’s strategic moves with Google and others indicate a push for independence, while Microsoft seeks to protect its investment and market position.