Build vs Buy: The Economics of Acqui-hires

The CODEW | Build vs Buy Series, Part 8 (Final)

When Microsoft paid $650 million for Inflection AI in March 2024, it wasn't buying Inflection's chatbot. It was buying Mustafa Suleyman and roughly 70 researchers, and it had them working inside Microsoft within weeks — a fraction of the six to twelve months and $5–10 million in recruiting fees it would normally cost to hire 20 senior engineers individually, let alone an entire proven team. 

Big Tech companies spent more than $40 billion on deals like this in 2024 and 2025 combined, more than all prior acqui-hire activity in history put together. This series closes where it started: acquisition is often the fastest form of building. Acqui-hires are that logic taken to its purest extreme — buying a company not for what it makes, but for who works there.

Photo by cottonbro studio from Pexels: 

How the Money Actually Flows

The headline number in an acqui-hire press release rarely describes what any single person actually receives. A typical deal prices talent at $1–5 million per engineer in Silicon Valley, structured as an asset purchase where the acquirer buys the IP, the team, and sometimes the technology. But liquidation preferences mean investors get paid first out of the company-sale portion — founders and common shareholders often walk away with far less than the topline figure implies. 

The real value for the people joining shows up somewhere else entirely: new-hire packages. Signing bonuses, RSU grants on a four-year vesting schedule with a one-year cliff, and cash bridges designed to cover whatever unvested startup equity someone is leaving behind. A senior ML engineer might land $1.5–3 million in RSUs at the acquiring company — money that has nothing to do with the sale price of the company they came from.

For the rarest tier of talent — proven frontier AI researchers, of which there are reportedly fewer than 40,000 globally — the numbers detach from any per-engineer benchmark entirely. Individual compensation packages now reach $10–20 million a year, with Meta reportedly offering some researchers packages worth up to $300 million over four years. When training a single frontier model costs upward of $100 million, a proven team that already knows how to do it efficiently is, in a strange way, a bargain — even at $30–90 million per engineer.

The Structure Built to Dodge Antitrust Review

A specific hybrid structure has become the industry default: the acquirer pays a large licensing fee for the target's technology — non-exclusive, so the startup nominally keeps its own IP — while separately hiring the founders and core team on lucrative individual packages. 

Microsoft's Inflection deal ($620 million for a model license, plus $30 million to waive legal claims), Google's $2.7 billion Character.AI transaction, and Google's roughly $2.4 billion Windsurf deal all followed this template. None of them were structured as a formal acquisition of the company, which meant none of them initially triggered Hart-Scott-Rodino premerger antitrust notification. 

Regulators have since caught on: Senators Warren, Wyden, and Blumenthal have formally urged the FTC and DOJ to review the Nvidia, Meta, and Google versions of this structure as de facto mergers, the acting head of the DOJ's Antitrust Division has called the pattern a red flag designed to sidestep merger review, and the FTC chair has said the agency is examining whether these deals should trigger HSR reporting. As of mid-2026, every template pioneered this way has still been copied within months by another buyer, regulatory attention notwithstanding.

Who's Left Behind

The structure that makes acqui-hires fast for acquirers makes them brutal for everyone not explicitly invited along. MIT Sloan research analyzing 4,000 acquisitions found that 33% of acqui-hired employees leave within the first year — nearly triple the 12% attrition rate of a traditional hire — and roughly half have departed by the four-year vesting cliff. The legacy company is typically wound down, its product discontinued, its non-selected employees left holding equity in a shell with no buyer and no path to liquidity. Reverse acqui-hires make this starker still: because the company itself is never formally acquired, the startup that gets "acquihired" by having its top talent hired away is often left an empty shell, its remaining employees and any product roadmap abandoned in place.

Why Companies Pay This Much

Three forces converge to make these numbers rational rather than reckless: a hard talent shortage, with fewer than 40,000 people globally who have real frontier-model experience; a startup failure rate that keeps refilling the supply of teams willing to be acquired, since roughly 90% of startups fail and most venture-backed companies never return investor capital; and a strategic motive that isn't always about the buyer's own roadmap — acquiring a team can be as much about denying that talent to a competitor as it is about using it. That denial motive is precisely what worries regulators and legal scholars: talent hoarded defensively, rather than deployed productively, is a cost the market doesn't easily see.

The Takeaway

Across all eight parts of this series, one theme holds: when a market is moving too fast for a normal build timeline, acquisition compresses years into weeks. Acqui-hires are the most literal version of that trade — they don't buy a product roadmap, a customer base, or even a company in any meaningful sense. They buy the only thing that was ever actually scarce: a small number of people who already know how to build the next model, wrapped in whatever corporate structure makes the transaction fastest and cleanest to close. The press release will always describe it as an acquisition. The economics make it obvious it never was one.

Previously in the series: Can Internal R&D Still Beat M&A? This concludes the Build vs Buy series — read the full series from Part 1.