§ 01 · The partnership in numbers

A $13 billion commitment became a $100 billion relationship.

The headline investment figure has always understated the real exposure. Microsoft committed $13 billion to OpenAI and, according to an April securities filing, had delivered $11.8 billion of it. But the bigger number is the one most people missed: the cost of building and running the infrastructure that OpenAI’s models depend on. That is how a $13 billion equity commitment becomes a relationship worth more than $100 billion in total spend.

For anyone who has worked on a large partnership, the pattern will be familiar. The equity cheque is the visible part. The operational commitment — capacity, engineering time, product integration, exclusivity — is where the real money and the real lock-in sit. And when the strategic interests of the two sides diverge, it is the operational commitments that make unwinding painful.

§ 02 · Why buy, and why these targets

Antitrust is now a term in the deal, not a footnote.

The most instructive detail in the reporting is the deal Microsoft didn’t do. It reportedly weighed buying Cursor, the fast-growing AI coding company, and backed away over concerns that the deal would not survive regulatory review, given that Microsoft already owns GitHub Copilot. Shortly afterwards, SpaceX — which had absorbed Elon Musk’s xAI in February — announced a tie-up with Cursor, and in June said it planned to acquire the company in a $60 billion all-stock deal.

That sequence is worth sitting with. The best-capitalised buyer in the market could not credibly bid for the most obvious strategic target, because the overlap with an existing product made the regulatory risk too high. In practice, that means antitrust exposure is now being priced into AI M&A the same way it has long been priced into telecom or pharma deals. The buyer with the most to gain is often the buyer with the hardest path to closing.

It also explains why Inception is an interesting target. A small lab working on a different technical approach — diffusion models that generate and refine many tokens at once rather than one at a time — does not overlap with anything Microsoft sells. It is a capability and talent acquisition, not a market-share acquisition. Microsoft’s venture arm, M12, had already invested in the company’s seed round, which gives it the classic corporate-venture advantage: information, relationships and a seat near the table before a process formally starts. Reuters reported that Inception had hired a bank and was looking for a price above $1 billion, with SpaceX also circling.

§ 03 · Three seats at the table

How this reads depends on where you sit.

§ 04 · The startups already inside the funnel

Microsoft’s cheapest deal-sourcing tool is a cloud credit.

There is a quieter part of this story that gets less attention. Long before Microsoft’s corporate development team looks at a startup, Azure often already knows it. Through programmes such as Microsoft for Startups, Microsoft hands early-stage companies large Azure credit grants — frequently six figures for venture-backed teams — along with access to its AI models. The credits are a customer-acquisition tool for Azure, but they also put thousands of AI startups inside Microsoft’s ecosystem, building on its infrastructure, years before any of them become acquisition targets.

From a deal perspective, that relationship cuts both ways. A startup built on Azure is easier for Microsoft to integrate, and its technical diligence is simpler. But many startups do not end up using all of what they were granted. Teams pivot, move workloads to another cloud, get acquired by a buyer on a different stack, or simply grow more slowly than the grant assumed. Credits expire on a fixed schedule regardless.

That leftover balance is increasingly something founders and CFOs look at during a transaction or a strategic change, rather than writing off. A secondary market has emerged where companies holding unused Azure credits can recover part of their value before they lapse, rather than watching them expire. Whether that is possible in a given case depends on the specific programme terms, so it is worth reading Microsoft’s own programme FAQs carefully — but for a startup preparing for an exit or a stack change, it belongs on the same checklist as cloud commitments, software contracts and other liabilities that change hands in a deal.

§ 05 · The bottom line

The era of renting frontier AI is ending.

Microsoft’s shift is not a dramatic break. It is what happens when a partnership designed for one phase of a market meets a different phase. In 2019, buying a frontier lab was neither possible nor necessary; partnering was the efficient route. In 2026, with researchers commanding enormous pay packages, startup valuations rising fast and the leading labs building ever-larger models, owning capability has become the safer bet — even at prices that would have looked absurd three years ago.

For the rest of the market, three things follow. AI startups with differentiated technology and no product overlap with the giants will see more bidders, not fewer. Founders whose products compete directly with a platform should plan for a buyer universe shaped by regulators as much as by strategy. And everyone on the sell side should treat the details — cloud commitments, credit balances, key-person retention, regulatory pathways — as part of the valuation, not as clean-up work for after signing.