In a bombshell Reuters exclusive published Thursday morning, five people familiar with the matter revealed that Microsoft is actively shopping for artificial-intelligence startups as the company prepares for a future without its once-indispensable partner, OpenAI. The news marks a watershed moment in the history of AI — a public acknowledgment that one of the most consequential corporate partnerships of the 21st century is quietly unraveling.
For years, Microsoft and OpenAI were synonymous. When OpenAI launched ChatGPT in late 2022, it was Microsoft’s $1 billion 2019 bet that made the moment possible. As ChatGPT swept the world, Microsoft rode that wave into its AI renaissance, integrating GPT models into Bing, Office, and Azure, and watching its market cap swell in kind. It was a perfect symbiosis. Today, that relationship looks very different.
The Deal That Built an Era
Microsoft’s investment in OpenAI didn’t stop at $1 billion. By April 2026, the company had pumped $11.8 billion of a promised $13 billion into the startup — an arrangement that gave Microsoft exclusive rights to OpenAI’s technology and gave OpenAI guaranteed access to Microsoft’s Azure computing infrastructure. In total, including infrastructure and hosting costs, Microsoft has spent over $100 billion on the partnership, according to testimony given in court this week.
But that exclusivity carried a price. Microsoft was contractually barred from building its own foundation model that could compete with OpenAI’s offerings. As OpenAI grew from research lab to trillion-dollar darling, and its needs ballooned beyond what Microsoft could supply, the terms grew suffocating on both sides.
The contract has been loosened repeatedly since its inception. A key amendment in late 2025 finally allowed Microsoft to develop artificial general intelligence — the still-theoretical form of AI that can perform complex tasks better than a human. Then, in late April 2026, OpenAI and Microsoft restructured the deal again, eliminating exclusivity entirely. OpenAI is now free to partner with Microsoft’s rivals, including Amazon. Microsoft’s IP license on OpenAI technology was converted to non-exclusive status through 2032. Microsoft stopped paying OpenAI a share of its revenue. The partnership became a business arrangement rather than a lifeline — and that changes everything.
Courting the Startup Market
Into that vacuum, Microsoft has begun moving aggressively. The company looked hard at Cursor, the AI-powered code-generation startup whose annualized revenue has rocketed from zero to $2 billion in three years. Microsoft ultimately backed away — not from lack of interest, but from internal fears that owning Cursor alongside GitHub Copilot would draw antitrust scrutiny. Days later, SpaceX swooped in and announced a deal for Cursor instead.
That SpaceX is competing with Microsoft for elite AI startups would have seemed absurd two years ago. But SpaceX, which acquired Elon Musk’s AI research startup xAI in February 2026, is now a serious player in the AI infrastructure and talent game — and according to sources, SpaceX also courted Inception, the same Stanford-born startup that Microsoft is currently in active discussions to acquire.
Inception, founded in mid-2024, has attracted attention for an unusual reason: its large language models use a technique called diffusion, more commonly associated with image and video generation. Rather than generating one token at a time as traditional language models do, diffusion models generate and refine multiple tokens simultaneously, dramatically boosting output speed. Microsoft’s M12 venture fund participated in Inception’s $50 million seed round in late 2025. Now, the company may be looking to bring the startup fully in-house, though discussions are ongoing and may not result in a deal. Inception has reportedly hired a bank to help negotiate — and is asking for more than $1 billion.
The Race to 10 Trillion Parameters
The urgency behind Microsoft’s search is driven by a sobering reality: the frontier of AI has moved staggeringly fast. The most advanced AI labs are now building models with approximately 10 trillion parameters — a measure of their complexity and capability. Three years ago, 1 trillion parameters represented the cutting edge. The exponential gap makes catching up feel almost existential for a company that, just a few years ago, was at the vanguard of the technology.
This is why Microsoft’s AI strategy under Mustafa Suleyman, the DeepMind co-founder brought in to build the company’s internal AI capabilities, has taken on new urgency. The company has stated publicly that it intends to have its own frontier-grade AI model in production “some time” in 2026. That goal may have seemed comfortable when OpenAI was a close partner. Now that the relationship has fundamentally changed, it’s a race against competitors who have years of head start.
OpenAI is not sitting still either. The company has restructured internally, shed what it internally calls “side quests,” and is now aggressively pursuing enterprise and coding markets with backing from Amazon, Google, and Oracle — companies that have collectively indicated willingness to fund tens of billions of dollars in infrastructure. OpenAI’s strategic flexibility has increased enormously, even as Microsoft has lost the guaranteed exclusivity that once defined the deal.
What This Means for the Industry
The slow-motion separation of Microsoft and OpenAI is more than a business story. It’s a reorganization of the entire power structure of the AI industry.
When the partnership was at its zenith, it effectively created a duopoly of trust at the frontier: OpenAI supplied the intelligence, Microsoft supplied the compute and distribution. Every enterprise customer choosing Azure was, indirectly, choosing OpenAI. That linkage is now severed. OpenAI can sell through AWS. Microsoft can build its own models. Enterprise buyers now face a genuinely more fragmented marketplace — and arguably a more competitive one.
For startups like Inception, the disruption creates an extraordinary moment of leverage. When Microsoft, SpaceX, and the broader big-tech ecosystem are all competing for the same handful of elite AI teams, valuations become untethered from traditional metrics. A 2024-founded startup with an unproven technique is commanding a billion-dollar price tag — not because diffusion language models are proven at scale, but because Microsoft and its rivals are desperate for differentiated talent and technology.
Diffusion-based text generation carries real uncertainty. AI researchers have noted that the technique can be unpredictable, and it’s unclear whether it can scale to the mammoth multi-trillion-parameter models that frontier research increasingly demands. But in a market defined by strategic hedging and arms-race logic, uncertainty is almost beside the point. What matters is not having access to the next generation of AI talent — that is the real existential risk.
The End of an Era, the Start of Something Stranger
For those who have watched the AI industry up close, there is something almost poignant about this moment. The Microsoft-OpenAI partnership built the current AI age. ChatGPT wouldn’t have reached hundreds of millions of users without Azure’s infrastructure. Microsoft wouldn’t have reclaimed its reputation as a cutting-edge innovator without GPT-4. Both parties transformed each other.
Now they are, slowly but unmistakably, becoming rivals.
The Reuters report paints a picture of a company — Microsoft — that is no longer content to be a distribution channel for someone else’s intelligence. After spending more than $100 billion, watching OpenAI’s valuation soar past $300 billion, and seeing its own contract terms loosened year after year, Microsoft appears to have decided that the only durable strategy is independence.
Whether Inception’s diffusion models will get Microsoft where it wants to go is an open question. But the direction is now clear: Microsoft is done waiting at the frontier, and it intends to plant its own flag there. The era of the Microsoft-OpenAI partnership that defined AI’s breakout years is ending — and whatever comes next will be built by companies that have learned, sometimes painfully, that in AI, dependency is a liability.
