Microsoft feared reliance on OpenAI, Musk and Altman trial testimony reveals

The legal battle between Elon Musk and Sam Altman over the past few weeks has largely revolved around the evolution of OpenAI. But it also highlights the challenges that the rapid growth of artificial intelligence has caused for OpenAI’s closest partners. microsoft.

The high-profile discovery reveals that Microsoft CEO Satya Nadella was worried about OpenAI replacing him in his company’s technology hierarchy as far back as April 2022, seven months before the launch of ChatGPT, the event that sparked the generative AI boom and made Altman a celebrity.

“I don’t want to be IBM And OpenAI becomes Microsoft,” Nadella wrote in an email to executives in April, nearly three years after Microsoft wrote OpenAI its first $1 billion check.

Nadella was referring to the early days of technology, when Microsoft was more important than IBM, the dominant computer maker at the time. In 1980, IBM agreed to distribute Microsoft’s operating system on its computers, but the software maker ended up siphoning off much of the market value.

To avoid the same fate, Microsoft needed to ensure that it not only provided Azure cloud services to OpenAI, but also stood to benefit from early intellectual property agreements with labs at the forefront of large-scale language model development.

“It’s become more core and more important to have real ownership at every layer of the stack,” Nadella testified Monday in the Musk v. Altman trial in Oakland, California.

More than four years after the IBM irony, history has vindicated Nadella’s initial concerns. OpenAI’s models and services have become so popular that the company’s valuation has ballooned to $850 billion and it has formed partnerships with Microsoft cloud rivals. google, oracle And as of February, Amazon.

“Obtaining the intellectual property rights was very important to us because essentially we were giving up the opportunity to develop independently,” Nadella said Monday, referring to the initial agreement with OpenAI.

Microsoft and OpenAI have had to change their terms and conditions multiple times. Most recently, in April, the agreement was amended to cap OpenAI’s revenue sharing payments to Microsoft and allow OpenAI to offer “any product” to any provider’s customers, including Amazon and Google.

This partnership remains critical to Microsoft’s cloud business as demand for computing resources needed to run large-scale models continues to soar. About 45% of Microsoft’s remaining commercial performance obligations are tied to OpenAI at the end of 2025, and the software giant is building data centers to keep up with demand.

“It’s better to be an investor and not take on all the execution risk!” Nadella told executives in a July 2022 email that was discovered.

Michael Wetter, Microsoft’s head of corporate development, said in court Wednesday that Microsoft will have spent more than $100 billion on OpenAI by June 2026, including investment commitments, infrastructure and hosting costs.

Kevin Scott, Microsoft’s head of technology, testified that the first supercomputer his company built using OpenAI several years ago took about six months to assemble the 10,000 graphics processing units that made up most of the data center.

“I’m extremely proud of the work we’ve done to make OpenAI possible, both in terms of our research and the products we bring to the world,” said Scott.

Nadella said at trial that OpenAI taught Microsoft how to build supercomputers suitable for AI work, and that “one of the benefits was the know-how that came with doing that.”

But while Microsoft has established itself as a leading provider of AI infrastructure, it is no longer uniquely positioned to offer OpenAI’s technology to its customers. And Microsoft has struggled to sell its AI story, which helps explain why its stock is down 16% this year while its cloud peers are trading higher across the board.

By January 2024, Nadella said, the model is becoming “more of a commodity.” Two months later, the company hired Mustafa Suleiman, co-founder of DeepMind AI Lab, now owned by Google. Mr. Nadella appointed Mr. Suleiman as CEO of the new AI division, with responsibility for Bing and other products.

With Suleiman at the helm of AI, Microsoft focused on building its own models that could compete with OpenAI and other leaders in the space such as Anthropic and Google. Last August, the company announced it had begun testing an AI model it developed that could potentially improve the functionality of its consumer assistant Copilot.

The market is moving at an accelerated pace as Microsoft tries to gain a foothold in the modeling world. In March, Microsoft reorganized Copilot’s leadership team, selecting recent hires and former employees. snap Executive Jacob Andreou will be responsible for the consumer and commercial Copilot experience. As part of the reform, Suleiman will pursue model development.

Since 2024, Microsoft has openly acknowledged that OpenAI is a competitor, partnering with other companies that build AI models, including Musk’s xAI, and releasing those tools to developers through Azure. Microsoft announced last year that it was leveraging Anthropic’s AI models for specific use cases, and shortly after agreed to invest $5 billion in the OpenAI rival.

Microsoft credits AI for much of its continued growth in Azure. And the balance sheet includes profitable equity investments. But unlike ChatGPT, none of the software products that incorporate AI have become blockbusters.

This left the company with a decentralized strategy of engaging with many model developers while also building its own parts of the stack, such as models and chips.

“We have decided to relinquish some of our early rights to give us the flexibility to allow OpenAI to continue to scale and thrive,” Nadella said Monday. “For me, our guiding principles as a company are to be a good partner and a good platform company. So regardless of what’s going on on the side, we’ll continue to keep that front and center.”

Featured: Microsoft CEO Satya Nadella testifies in OpenAI lawsuit