The potential acquisitions could help the company stock up on AI talent and deliver on its stated goal of building a cutting-edge AI model by next year, three of the people said.
This spring, Microsoft weighed acquiring code-generation startup Cursor, four people said. But Microsoft backed away due to internal concerns that such a deal would not pass regulatory scrutiny, given Microsoft’s ownership of GitHub Copilot, three of the people said.
Microsoft is in discussions with Inception, a small startup built by a Stanford University team focused on a different method of developing large language models, three people familiar with the matter said. Inception was founded in mid-2024. Microsoft’s venture fund M12 invested in Inception’s $50 million seed round in late 2025.
The discussions are ongoing and may not result in a deal, these sources said.
Inception declined to comment.
Heated market
Microsoft is eyeing deals in an increasingly heated market. AI researchers can easily command tens of millions of dollars or more in compensation. Startup valuations are soaring as investors scramble for positions in promising AI technology.
Microsoft is also facing significant competition for deals from other tech giants, notably Elon Musk’s SpaceX, two people familiar with the matter said. SpaceX, which bought Musk’s AI research startup xAI in February, announced a deal with Cursor shortly after Microsoft walked away.
Cursor declined to comment.
SpaceX also courted Inception, three people said. Inception recently hired a bank to help negotiate a deal, a person familiar with the startup said, adding that Inception is looking for a price of over $1 billion.
SpaceX did not immediately respond to a request for comment.
10 trillion parameters
Catching up to OpenAI and other labs at the frontier is a tall order. Some of the most advanced AI labs are building models of around 10 trillion parameters, a measurement of their sophistication, researchers say. That is up from about 1 trillion parameters three years ago.
Inception’s models produce text using a technique called diffusion, more commonly used to generate AI images and videos. While standard models generate one token at a time, diffusion generates and refines multiple tokens simultaneously. This method can significantly boost the model’s speed.
But diffusion can be unpredictable, and it is unclear if it can be used to produce mammoth-sized models, AI researchers say.
Any deals would add to the work underway at Microsoft, including teams led by DeepMind co-founder Mustafa Suleyman, a person familiar with the strategy said.
Microsoft and OpenAI have been partners since 2019, when Microsoft invested $1 billion into the then-unknown research lab. OpenAI’s release of ChatGPT in late 2022 anointed Microsoft as an AI pioneer while also powering growth for Microsoft’s Azure cloud-computing business. Microsoft has given $11.8 billion of its promised $13 billion to OpenAI, Microsoft said in an April 29 securities filing.
Microsoft has spent more than $100 billion on its OpenAI investments and its costs of building infrastructure and hosting, Michael Wetter, who runs the company’s corporate development, testified in court on Wednesday.
The initial deal gave Microsoft exclusive access to OpenAI’s technology and gave OpenAI a guaranteed source of computing resources to pursue research. But tensions flared between OpenAI and Microsoft over the years as both sides chafed over the contract’s restrictions.
OpenAI found that its needs outstripped what Microsoft could supply. Microsoft was also contractually barred from building a foundation model that could compete with OpenAI’s offerings, two of the people said. The two companies have loosened their contract several times over the years.
An amended deal in late 2025 allowed Microsoft to build artificial general intelligence, a still-theoretical advanced form of AI that can do complex tasks better than a human. In late April, OpenAI and Microsoft struck a deal that gives OpenAI the freedom to build some products with Microsoft’s rivals, such as Amazon.
