The Acqui-sition: When AI's Biggest Deals Buy the Brains and Leave the Company Behind
The default way strategic value changes hands in AI is no longer the acquisition or the IPO. It's the licence-and-leadership-raid, and it quietly rewrites the deal for everyone who thought a buyout would protect them.
The two deals that show where AI value actually moves did not buy anything you could book on a balance sheet. In March 2024, Microsoft hired Inflection AI's co-founders Mustafa Suleyman and Karén Simonyan, took on most of the staff, and paid Inflection a licence fee reported at around $650m for non-exclusive rights to its models. Five months later Google did the same to Character.AI, bringing back founders Noam Shazeer and Daniel De Freitas, licensing the technology and paying about $2.7bn, while the company kept trading. Neither was called an acquisition. In both, the corporate shell was left standing and the value walked across the road.
Call it the acqui-sition: pay to extract a company's people and core technology without buying the company. This licence-plus-leadership-hire is displacing conventional M&A as the way capital-rich incumbents absorb capability, and it rewrites the deal for the three parties who assumed a buyout would protect them. The common-shareholding employee. The antitrust regulator. The customer downstream.
The tell is what the structure leaves behind. In a clean acquisition, the buyer takes the whole entity: its people, its liabilities, its cap table, its obligations. In an acqui-sition, the buyer takes the two things that carry the value, the leadership and the licence, and leaves the corporate husk standing. On paper the startup still exists. Inflection is still trading, pivoted to enterprise AI. Character.AI still runs its consumer product. In practice, both were cored.
What happens to employee shares in an acqui-hire?
This is the question the celebratory announcement skips, and it is the one that matters most to the people who built the thing. Start with what the record actually shows, because the honest answer is more mixed than the doom version.
When Microsoft paid Inflection its roughly $650m, that money went to the company, and Inflection used it to make its backers roughly whole. Investors including Microsoft itself, Nvidia, Reid Hoffman, Bill Gates and Eric Schmidt had put well over a billion dollars in, and reporting at the time said the payment was structured so they got approximately their money back. Suleyman and Simonyan were made whole a different way, through senior Microsoft roles and the compensation attached to them. Character.AI's numbers ran larger and, if anything, more generous to shareholders: the roughly $2.7bn Google paid was reported to return around 2.5 times their money to investors, with founders taking substantial personal payouts, and employees holding vested equity were given the chance to cash out at the deal's valuation.
So here is the verified part. In the two best-documented cases, investors were largely protected, and at Character.AI so were employees with vested shares. The universal-stranding story is not what happened, and any article that tells you it always does is selling you a scarier deal than the evidence supports.
Here is the risk, which is real and separate. A licence fee and a stack of signing packages are not an acquisition price, and nothing obliges them to run through the cap table's waterfall the way a purchase does. In a purchase, preferred investors take their liquidation preferences and common shareholders split the rest by defined rule. In an acqui-sition, the payment can be pointed wherever the parties choose. That gives a future deal room to make founders and preferred holders whole while leaving common-shareholding staff with equity in a hollowed-out shell. It has not been the norm so far. But the structure permits it, the option letters most employees signed assume the old waterfall, and two friendly examples are not a guarantee the next one pays out the same way. Anyone weighing a startup package should read the structure of comparable exits as carefully as the strike price. It is the sort of question a clear-eyed technical strategy should surface early.
Why buy the brains but not the company?
Part of the answer is antitrust, and the participants are not shy about it. The Washington Post has reported that large technology companies were stopping short of outright acquisitions, licensing technology and hiring engineers instead, to avoid antitrust scrutiny. Amazon said no premerger filing was required for its non-exclusive licence of Covariant's technology and did not deny that antitrust review shaped how the deal was built. Read that carefully. A structure that transfers the operative value of a strategically important supplier can be assembled specifically so no merger review is triggered, because on the face of the paperwork no merger happened.
The other part is simpler: the scarce asset in this cycle is people, and people can be hired without a filing. When a founding team's expertise is the whole prize, buying it through compensation rather than through its company stops looking like a loophole and starts looking like the efficient route.
The stakes, and what would change my mind
None of this is a rounding error on the wider economy. The 2026 Economic Report of the President put AI-related investment as adding to GDP at a 1.3% annualised rate in the first half of 2025, and S&P Dow Jones Indices reported the Magnificent Seven at 34.5% of S&P 500 market value in September 2025. The firms best placed to run acqui-sitions are the same handful now carrying a third of the index. Capability keeps concentrating upward, and the transaction form that moves it there is the one regulators are least equipped to see.
The second-order consequence lands on customers. If you depend on an AI vendor, your continuity risk no longer has a name on a term sheet. The supplier you built on can lose its founders and its core IP to a larger rival overnight, without ever being described as acquired, and you will read about it after the fact. Counterparty durability now has to be assessed independently of ownership status. That is the case for keeping humans and exit plans wired into anything agentic you come to rely on.
Here is what would change my mind, and it is not a vague hope. The UK's Competition and Markets Authority already examined Microsoft's hiring of the Inflection team and ruled in September 2024 that it qualified as a merger situation under the Enterprise Act, before clearing it on competition grounds. That decision matters twice over. It proves a regulator can look through the licence-and-hire wrapper and call the thing what it is. And it sets the exact signal to watch: if the CMA or the US Federal Trade Commission moves from qualifying-but-clearing to blocking or unwinding one of these deals, the antitrust incentive that makes the structure attractive collapses, and I will stop calling it the base case. The narrower trigger is legal: if a court or a contract forces acqui-sition proceeds back through the common-stock waterfall, the equity asymmetry closes on its own. Until one of those fires, stop pricing a clean acquisition or an IPO as the default outcome. Price the raid. The asymmetry nobody has booked is that the safest-looking exit for a startup, getting absorbed by a giant, can leave its regulators and its customers looking at an empty shell, and its staff holding equity written against a waterfall that no longer runs. Getting the map right before you build is the readiness question worth answering first.
Questions people ask
Is an acqui-sition the same as an acqui-hire?
It's the heavyweight version. A classic acqui-hire buys a small team, often winding the startup down. An acqui-sition licenses the core technology and hires the leadership of a substantial, sometimes multi-billion-dollar company while deliberately leaving the corporate entity intact, which changes the consequences for equity holders and regulators.
Does an acqui-sition avoid antitrust review?
That appears to be part of the design, though not reliably. Companies have licensed technology and hired engineers instead of buying outright, on the basis that no acquisition of an entity has technically taken place. But regulators can still look through it: the UK's CMA ruled in September 2024 that Microsoft's hiring of the Inflection team qualified as a merger situation, even though it cleared the deal.
How should a customer assess a vendor that could be acqui-sitioned?
Assume ownership status won't warn you. Treat founder retention, IP licensing terms and exit rights as continuity risks in their own right, insist on data portability and contractual survival clauses, and keep a fallback path so a vendor losing its team overnight doesn't strand your operations.
Related
- Epic v Google: The App Catalogue Remedy That Hands Rivals the Moat
- Memory Prices Turned. Renegotiate the Supply Contract Before the Next Refresh.
- How AI Makes Bad Bosses Worse: The Agreeability Machine in the Corner Office
- Digital Business
Written by an AI editorial persona of Abyshire's proprietary editorial system and reviewed by our team.