Last fall, OpenAI’s CEO, Sam Altman, approached Microsoft’s CEO, Satya Nadella, asking if the tech giant would consider investing billions more into OpenAI. Microsoft had already poured $13 billion into the startup, and Nadella was initially open to continuing the funding. However, after Altman was briefly ousted by OpenAI’s board last November, Nadella and Microsoft reevaluated their stance, according to four people familiar with the situation, who spoke on condition of anonymity.

Over the following months, despite OpenAI, which expects a $5 billion loss this year, repeatedly requesting more funding and computing resources, Microsoft remained firm in its decision not to offer additional investment. Altman, who once described OpenAI’s collaboration with Microsoft as “the best bromance in tech,” has since seen that relationship become strained. Financial pressures, concerns about OpenAI’s stability, and disagreements between employees of the two companies have put significant stress on their five-year partnership, according to interviews with 19 people familiar with the situation.

This tension highlights a major challenge for A.I. startups: their dependence on tech giants for both funding and the computing power needed to develop advanced systems, as the large companies control the necessary cloud infrastructure.

The Microsoft-OpenAI relationship illustrates this dynamic well. OpenAI, the creator of ChatGPT, entered an exclusive deal to buy computing power from Microsoft when it received its substantial investment. In a statement on Thursday, Altman expressed appreciation for the partnership, acknowledging Microsoft’s significant early investment and the vast computing resources that had been essential to OpenAI’s breakthroughs. He emphasized their continued commitment to shared goals for the future.

However, over the past year, OpenAI has been seeking to renegotiate its deal with Microsoft to secure more computing power and alleviate its financial burdens. Meanwhile, Microsoft has grown wary of becoming overly reliant on OpenAI’s technology. Nadella privately expressed his concerns about Altman’s firing, as it shocked him, according to five sources familiar with his thoughts.

Since then, Microsoft has begun to diversify its investments. “We have continued to invest in OpenAI at many points in our partnership,” said Microsoft’s chief technology officer, Kevin Scott, in a recent interview, acknowledging the company’s significant role in funding OpenAI.

Nevertheless, in March, Microsoft made a strategic move, spending at least $650 million to hire most of the staff from Inflection, an OpenAI competitor. Inflection’s former CEO, Mustafa Suleyman, now leads a Microsoft team developing new A.I. technologies based on OpenAI’s work, with a focus on building alternatives that could eventually replace OpenAI’s offerings.

This shift in dynamics has frustrated some at OpenAI, including Altman, who are concerned about Suleyman’s presence at Microsoft. Suleyman’s team, which collaborates directly with OpenAI engineers, is seen as a potential threat. Tensions escalated when Suleyman allegedly yelled at an OpenAI employee during a video call, frustrated with delays in delivering new technology to Microsoft. Other OpenAI staff have complained about Microsoft engineers violating security protocols during software downloads.

After Microsoft stepped back from talks about further funding, OpenAI found itself in a difficult position, needing additional cash while also feeling constrained by its exclusive contract with Microsoft. Over the past year, OpenAI has repeatedly tried to negotiate a reduction in costs and sought permission to buy computing resources from other providers. In June, Microsoft made a concession, allowing OpenAI to sign a $10 billion deal with Oracle to gain access to additional computing power. This agreement allowed OpenAI to diversify its infrastructure without completely cutting ties with Microsoft.

In recent weeks, the two companies also renegotiated parts of their agreement, reducing the fees OpenAI pays for computing power, although the specifics of the deal remain unclear. While Microsoft remains satisfied with the partnership, OpenAI continues to explore alternative sources of computing power and investment.

OpenAI has also sought to diversify its investors. It has engaged with companies like Nvidia and MGX, as well as Apple, in strategic discussions aimed at boosting its long-term prospects. Although Apple has not yet invested in OpenAI, its collaboration resulted in the integration of ChatGPT into the iPhone earlier this year.

Earlier this month, OpenAI closed a $6.6 billion funding round led by Thrive Capital, with Nvidia, MGX, and others contributing. Microsoft also participated in this round, signaling its continued, albeit cautious, involvement with the startup.

While OpenAI expects its computing costs to rise dramatically—projected to reach $37.5 billion annually by 2029—the recent changes in its partnership with Microsoft may help alleviate some of these expenses. Nonetheless, OpenAI employees have expressed frustration, citing insufficient computing resources from Microsoft. They worry that if another company beats OpenAI to creating an A.I. that rivals human intelligence, Microsoft’s lack of support will be partly to blame.

Interestingly, OpenAI’s contract with Microsoft contains a clause that could allow the startup to break free. If OpenAI achieves artificial general intelligence (AGI)—a system matching human brainpower—Microsoft would lose access to its technologies. This clause, originally designed to prevent misuse of AGI, is now viewed by OpenAI executives as a potential escape route, with the company’s board having the authority to determine when AGI has been achieved.

At a recent A.I. conference in Seattle, Microsoft downplayed its focus on OpenAI, with Asha Sharma, an executive involved in Microsoft’s A.I. efforts, emphasizing the company’s commitment to offering a range of choices for its customers.