Microsoft Hunts AI Startups to Reduce Dependence on OpenAI

Microsoft is actively shopping for AI startup acquisitions as it prepares to operate more independently from OpenAI, its longtime but increasingly complicated partner. The software giant is exploring deals that would bolster its AI talent base and help it build a frontier-grade AI model by next year — an ambition that puts it in direct competition with the very lab it has bankrolled to the tune of over $100 billion in investments and infrastructure costs.

This spring, Microsoft seriously considered acquiring Cursor, the popular AI code-generation startup, but walked away after internal concerns that the deal would draw regulatory fire — Microsoft already owns GitHub Copilot, a direct competitor. Within weeks of Microsoft's retreat, Elon Musk's SpaceX swooped in and announced its own deal with Cursor, underscoring how fiercely contested the AI acquisition market has become.

Microsoft is now in active talks with Inception, a startup founded in mid-2024 by a Stanford University team that builds large language models using a technique called diffusion — the same approach typically used to generate AI images and video. Unlike standard models that produce one word-like token at a time, diffusion models generate and refine multiple tokens simultaneously, potentially delivering significant speed advantages. Microsoft's venture arm M12 already participated in Inception's $50 million seed round in late 2025. Inception has reportedly hired a bank to shop itself and is targeting a valuation above $1 billion. SpaceX was also courting the startup, making the outcome far from certain.

The broader context is a Microsoft-OpenAI relationship that has grown strained despite its enormous financial scale. Microsoft has disbursed $11.8 billion of a promised $13 billion to OpenAI, and a late-2025 contract amendment finally allowed Microsoft to pursue artificial general intelligence research on its own. An April 2026 deal then gave OpenAI the freedom to partner with Microsoft rivals including Amazon. Any new acquisitions would layer onto internal AI work already underway, including teams led by DeepMind co-founder Mustafa Suleyman. The frontier Microsoft is chasing is formidable: leading labs are now building models with roughly 10 trillion parameters, up from about 1 trillion just three years ago.