Microsoft is actively evaluating AI startup acquisitions as the company positions itself for a future with less dependence on partner OpenAI, according to people familiar with the deliberations. The potential targets are intended to deepen Microsoft’s AI engineering bench and help the company pursue its objective of constructing a next-generation AI model by next year.
This spring, Microsoft considered acquiring Cursor, a startup focused on code generation, but sources said the company stepped away amid internal concerns that such a transaction could face regulatory challenges because Microsoft already owns GitHub Copilot. Following Microsoft’s withdrawal from talks, SpaceX announced a deal with Cursor, people familiar with the matter said.
Microsoft has also entered discussions with Inception, a small startup formed by a team out of Stanford that pursues an alternative approach to training large language models, according to multiple sources. Inception was founded in mid-2024 and later attracted investment from Microsoft’s venture arm, M12, which participated in the startup’s $50 million seed round in late 2025. The conversations between Microsoft and Inception are ongoing and may not lead to a transaction, the sources cautioned. Inception declined to comment.
A competitive and expensive market
Deal-making for AI talent is occurring in a highly competitive environment. Researchers with expertise in cutting-edge AI development can command compensation packages worth tens of millions of dollars or more, and startup valuations have been pushed higher as investors pursue promising technology. Microsoft faces notable competition from other large players, including SpaceX, which acquired Elon Musk’s AI research startup xAI in February and pursued Inception and Cursor, according to people familiar with the companies’ outreach.
One person familiar with Inception said the startup recently hired a bank to help negotiate a potential sale and is seeking a valuation above $1 billion. SpaceX did not immediately respond to requests for comment. Cursor declined to comment.
Technology questions and scaling challenges
Catching up with leading AI labs is a significant technical challenge. Some of the most advanced labs are building models on the order of 10 trillion parameters, researchers say, up from roughly 1 trillion parameters three years ago. Inception’s research produces text through a diffusion technique more commonly used in image and video generation. Unlike conventional models that generate one token at a time, diffusion generates and refines multiple tokens simultaneously, which can materially increase generation speed.
But diffusion has drawbacks: it can be less predictable, and experts are uncertain whether the approach can be scaled to the mammoth models being pursued by frontier labs. Those uncertainties are part of the calculus for any potential acquirer evaluating Inception’s work.
How any deal fits into Microsoft’s broader AI efforts
Any acquisition would be integrated into the wider set of AI initiatives already underway at Microsoft, which includes teams led by DeepMind co-founder Mustafa Suleyman, according to a person familiar with the strategy. The company’s AI activity has long been intertwined with its cloud business; OpenAI’s release of ChatGPT in late 2022 positioned Microsoft as a prominent AI adopter and contributed to growth at its Azure cloud-computing unit.
Microsoft’s relationship with OpenAI began in 2019, when Microsoft invested $1 billion into the then-small research lab. Over time the partnership provided OpenAI with a steady supply of computing resources and gave Microsoft privileged access to OpenAI’s technology. Microsoft disclosed in an April 29 securities filing that it has provided $11.8 billion of the $13 billion it had pledged to OpenAI. In testimony on Wednesday, Michael Wetter, who oversees Microsoft’s corporate development, said the company had expended more than $100 billion on its OpenAI investments and on related infrastructure and hosting costs.
Despite the long-running partnership, tensions emerged as both sides chafed at contractual limits. OpenAI concluded that its resource needs exceeded what Microsoft could provide, while Microsoft was contractually constrained from developing a foundation model that could directly compete with OpenAI’s offerings, people familiar with the agreements said. The two firms amended their contract several times; a 2025 amendment allowed Microsoft to pursue artificial general intelligence, a still-theoretical form of AI. In late April, OpenAI and Microsoft reached an agreement that permits OpenAI to build some products with Microsoft’s competitors, including Amazon.
Outlook
Microsoft’s pursuit of startups such as Inception and its previous consideration of Cursor reflect a broader attempt to assemble talent and capabilities that could reduce reliance on external partners. The outcomes of these discussions remain uncertain: not all talks lead to acquisitions, regulatory scrutiny can derail deals, and competition from other bidders may push valuations and terms higher. How Microsoft integrates any acquired technology into its existing AI roadmap will be central to assessing the strategic value of potential transactions.
This article presents the ongoing situation as described by people with knowledge of the matter; discussions may change and not all potential deals will necessarily close.