Big Tech isn't just hiring AI talent anymore. It's buying entire teams to get it.

Between March 2024 and January 2026, Microsoft, Google, Meta and Amazon put more than $20B into a new kind of deal: license the startup's technology, hire its founders and core team, and skip the slow path of a full acquisition. The startup keeps existing on paper, usually with a new CEO and a fraction of its staff. No shares change hands, so there is no merger to file.

Microsoft wrote it first. In March 2024 it hired Inflection AI co-founders Mustafa Suleyman and Karén Simonyan along with most of the company, and paid Inflection $620M to license its models plus another $30M for Inflection's agreement not to sue over the hiring. Less than a year earlier, Inflection had raised about $1.3B at a $4B valuation.

Amazon ran the play twice that summer: Adept in June, taking its founders and roughly a third of the staff, then Covariant in August, taking three founders and about a quarter of the team. Google paid $2.7B to Character.AI in August 2024 and $2.4B to Windsurf in July 2025. Meta went biggest among the four, putting $14.3B into Scale AI for a 49% non-voting stake and bringing Alexandr Wang in to lead its superintelligence effort.

Google did it again this January. DeepMind signed a licensing deal with voice startup Hume AI and brought over CEO Alan Cowen and about seven engineers to work on Gemini's voice features. Terms weren't disclosed. And the largest deal of the pattern didn't come from the Big Four at all: Nvidia paid a reported $20B in December 2025 to license Groq's inference chip technology and hire founder Jonathan Ross.

The size gets the headlines. The pace matters more. Four of the largest AI companies in the world independently landed on the same playbook inside two years, and each new deal got copied within months of the last one. That isn't opportunism. It's an admission that buying is now faster than building.

The same companies have mostly stopped doing traditional acquisitions. MicroVentures counts the Big Five at a decade low of seven deals in 2024, 14 in 2025 and 12 through May 2026. Capability is still changing hands. It just moves through license agreements and offer letters instead of merger filings.

In a market where model capability compounds and the talent pool is this thin, losing six months to organic hiring costs more than paying a premium to skip it. In a lot of these deals the product is beside the point. The deal is a shortcut past the hiring pipeline entirely, at a price these companies can absorb without blinking.

The math on individual hires explains why. Levels.fyi puts median total compensation for a software engineer at Thinking Machines Lab at about $746K. Meta was reported to have offered one of that lab's co-founders a package worth roughly $1.5B over six years, a figure Meta publicly disputed. When single researchers are priced like that, $650M for a working team with trained models looks cheap. There is a quieter motive too: a license fee can be booked as a research expense rather than a capitalized purchase, which makes the structure attractive to the buyer's finance team, not only its lawyers.

These deals are generous to founders and senior researchers. Everyone else on the cap table has a wider range of outcomes. Inflection's investors were reportedly paid back between 1.1 and 1.5 times their money. Covariant's reported price came in under its $625M valuation from 2023. Character.AI used its $2.7B to buy out investors and give remaining employees one-time cash-outs.

The companies left behind can survive, but they have to rebuild. Nvidia showed a new inference processor built on Groq's licensed technology less than three months after the deal. Groq spent that time re-staffing, and confirmed a $650M raise in June. Hume is still selling its models to other AI firms and says it is on track for $100M in revenue this year.

On September 9, the New York Times reported that the Justice Department is investigating whether Nvidia structured its Groq deal to avoid antitrust review, and has sent Nvidia a formal request for information. Fines are possible; forcing an unwind is considered unlikely. In February, Senators Warren, Wyden and Blumenthal urged the DOJ and FTC to treat these structures as acquisitions, and the FTC said in January it would take a closer look at deals like these.

The gap they are probing is simple. Merger review under Hart-Scott-Rodino kicks in above $119.5M, but it covers equity and asset transfers. A license plus a round of job offers sits outside the plain text. Expect that gap to narrow, and expect future deals to take longer, cost more in legal work, or convert into clean acquisitions.

If you're building in this space right now, understand what you actually look like to a potential acquirer: not a competitor to be neutralized, but a hiring problem someone else is willing to pay billions to solve in one move. Five things to do with that:

The market has told you what it values. Build a company worth more than the team inside it, or price the team like the asset it is.

Small and mid-size labs with open roles right now. If you're competing for the same people, these are the numbers you're bidding against.

Grouped by location. Tap a role to open the official listing.

Base salary as posted on each company's own job board, checked September 23, 2026. Equity and bonus not included. Listings close fast, so confirm on the posting before you apply or benchmark against it.

Reads and videos from the last 21 days, picked for early-stage founders.

Nothing here is legal or investment advice.