On July 11, 2025, Windsurf’s roughly 250 employees learned their CEO was gone. Varun Mohan, co-founder Douglas Chen, and a group of researchers left for Google DeepMind the same day Google agreed to pay $2.4 billion for a non-exclusive license to Windsurf’s AI coding technology, according to Reuters. Google didn’t buy Windsurf. No equity stake, no board seat, no acquisition announcement. There’s a name for this kind of deal, and in 2026 it has become the most common way an AI startup ends: the reverse acquihire.
What remained was a company with a product, a customer list, and about 200 employees whose leadership had just been hired by the industry’s largest player. Interim CEO Jeff Wang later described the mood inside the office as bleak.
Three days later, Cognition bought what was left. Its CEO Scott Wu structured the deal so every remaining Windsurf employee received a payout, with vesting cliffs waived and equity accelerated. That ending is the exception. Most people in Windsurf’s position get nothing.
An acquihire is when a company buys a startup mainly to hire its team rather than to own its product. The 2026 version, the reverse acquihire, skips the purchase entirely. A large tech company licenses the startup’s technology, hires its founders and top researchers, and leaves the legal entity intact so no merger filing is ever triggered.
Last updated: August 2026
Quick answers
What is an acquihire?
An acquihire is an acquisition where the buyer’s main goal is hiring the target’s team rather than owning its product or revenue. The term dates to around 2005. The acquired company’s product is usually shut down after closing, and investors often recover little or nothing.
What is a reverse acquihire?
A reverse acquihire is when a large company licenses a startup’s technology and hires its founders and key staff without acquiring the company. The startup keeps existing as a legal entity. Because no stock or assets change hands in the traditional sense, no merger notification is filed.
Is an acquihire good or bad for employees?
It depends entirely on whether you get an offer. Retained engineers typically receive signing bonuses and new equity worth well above market. Employees who are not retained hold common stock that sits behind investor liquidation preferences, which in a small deal absorb most or all of the proceeds.
What is an acquihire
An acquihire is an acquisition priced on people rather than product. The buyer wants a team that already knows how to work together, and it is willing to pay a premium to skip the eighteen months of recruiting, onboarding, and team formation that hiring those engineers individually would require.
The term is a portmanteau of acquisition and hire, coined around 2005 in Silicon Valley. Apple, Alphabet, Amazon, Meta, Microsoft, Uber, and Salesforce have all used the structure repeatedly to scale technical teams.
The classic acquihire has a specific emotional shape: it is usually a soft landing for a company that is failing. The startup could not raise its next round, the product never found a market, or the founders ran out of runway before they ran out of ideas. A buyer steps in, absorbs the team, and shuts the product down. Investors get their capital back or a fraction of it. The founders get jobs and a story that ends with the word acquired instead of the word shutdown.
What has changed in 2026 is that the companies being acquihired are not failing. Windsurf had a working product and paying customers. Scale AI was profitable enough that Google was its largest customer. Groq had shipped inference silicon. These are not soft landings. They are competitive removals dressed as recruiting.
What is a reverse acquihire and how is it different
The difference is ownership. In a normal acquihire the buyer purchases the company and inherits everything: the cap table, the contracts, the liabilities, the employees it does not want. In a reverse acquihire the buyer purchases nothing. It signs a licensing agreement for the technology and separately extends employment offers to the specific people it wants.
The startup survives on paper. Its investors still hold their preferred shares. Its remaining employees still have badges. What it no longer has is the leadership team and the research group that made it competitive.
Senators Elizabeth Warren, Ron Wyden, and Richard Blumenthal described the practice in a February 4, 2026 letter to the DOJ and FTC as Big Tech “swooping in to hire star talent and license technology, discarding the rest by the wayside.” Their letter argues these arrangements “function as de facto mergers.”
| Deal type | What the buyer gets | Who receives the money | Merger review | Fate of the startup |
|---|---|---|---|---|
| Traditional acquisition | The whole company: product, revenue, contracts, staff | Shareholders, in preference order | HSR filing if above the reporting threshold | Absorbed or run as a subsidiary |
| Classic acquihire | The team, plus whatever IP comes attached | Split between shareholders and retained staff compensation | Usually below the HSR threshold | Product wound down after closing |
| Reverse acquihire | A technology license plus the founders and top researchers | Mostly the individuals hired, through employment packages | No premerger filing required | Still exists, without its leadership |
Why are AI companies doing reverse acquihires instead of buying startups
Three reasons stack: the talent is scarcer than the technology, the structure avoids merger review, and the buyer inherits none of the target’s problems.
Start with scarcity. The number of people who have trained a frontier model end to end is small enough to fit in a conference room, and the companies competing for them have effectively unlimited balance sheets. Nvidia’s December 2025 deal for Groq’s inference chip technology reportedly ran about $20 billion and brought over founder Jonathan Ross and president Sunny Madra. McKinsey projects inference will account for more than half of AI workloads by 2030, which explains why Nvidia was willing to pay that much to move into a market it did not already dominate.
The second reason is procedural. The Hart-Scott-Rodino Act requires premerger notification when a company acquires assets or stock above a dollar threshold. A licensing agreement plus a set of employment offers does not obviously fit either category. That gap is the entire mechanism. Companies get the competitive effect of a merger without the filing that would let regulators review it first.
The third reason is cleaner than either. Buying a startup means buying its lawsuits, its underwater contracts, its data privacy exposure, and its employees. Licensing means buying none of that. The buyer selects the twenty or forty people it wants and leaves the rest as somebody else’s problem, which is a meaningful cost saving when the target has 250 employees and only 40 of them are the reason for the deal.
This is the same pressure reshaping headcount everywhere in tech. We covered a version of it in ClickUp’s 100x org, where the company restructured around a small number of very high-output humans and a much larger number of AI agents. Reverse acquihires are that logic applied to M&A: pay enormous sums for a handful of people, treat everyone else as overhead.
The money involved is large enough to distort how founders think about the exit. Nvidia’s roughly $20 billion Groq arrangement is bigger than most outright acquisitions in tech history, and it bought no company at all. That scale of capital is the same force pushing valuations across the sector, which we tracked in our coverage of how Dario Amodei built Anthropic into a trillion-dollar company in four years and OpenAI’s confidential IPO filing.
Who actually gets paid in an acquihire
The founders and the specific engineers the buyer wants capture most of the value, because most of the value arrives as compensation rather than as purchase price. This is the single most important thing to understand about these deals, and it is the part that headline numbers hide.
Deal consideration splits into two buckets. The first is purchase price, which flows through the cap table and is distributed according to liquidation preferences: investors first, then common. The second is employment compensation for the retained team, which sits entirely outside the cap table. Signing bonuses, salary, and new equity grants at the acquiring company are not proceeds from a sale. They never touch the preference stack, and they are not shared with anyone who does not get an offer.
In a reverse acquihire the second bucket is most of the deal. A $2.4 billion headline can be a licensing fee plus a set of nine-figure employment packages, with almost nothing arriving as distributable proceeds for common shareholders.
Retention packages for the people who are retained typically vest over three to four years and land somewhere between a 50 and 100 percent premium over market rate for the equivalent role at the buyer. Sometimes there is a one-year employment contract on top.
The employee who joined at seed stage, took a pay cut, and holds common stock is in the worst possible position. No employment offer means no compensation bucket. And a modest purchase price means liquidation preferences consume the proceeds before common sees anything. Two years of below-market salary, and the payout is zero.
What happens to the employees left behind
They usually get laid off, and the layoffs tend to arrive within weeks rather than months. Scale AI is the clearest case on record.
Meta paid more than $14 billion for a 49 percent stake in Scale AI in June 2025 and hired CEO Alexandr Wang to run its new superintelligence lab. Within weeks, Scale AI cut 200 full-time employees and ended work with 500 contractors, roughly 14 percent of its workforce. Google, its largest customer, moved to end the relationship because it now viewed Scale AI as an arm of a competitor. Other customers followed. By December 2025, Business Insider reported the company was managing pay cuts and customer losses, with one investor comparing it to a gutted fish.
That’s the second-order damage nobody prices into these deals. The startup doesn’t just lose its founders. It loses the customers who won’t buy from a company partly owned by their rival, and then it loses the ability to pay the people who stayed.
Windsurf is the counterexample worth studying. When Cognition acquired the remaining entity three days after Google’s deal closed, the terms gave every remaining employee financial participation, waived vesting cliffs for prior service, and accelerated vesting on Windsurf equity. Cognition CEO Scott Wu and Windsurf’s interim CEO Jeff Wang built that into the structure deliberately.
It worked because a second buyer wanted the remaining product and team. Most startups gutted by a reverse acquihire don’t get a second buyer. The pattern of founders leaving and everyone else absorbing the consequences shows up across the industry, including in the xAI co-founder exodus after the SpaceX deal.
Is a reverse acquihire legal and why is the FTC looking at it
Reverse acquihires are legal today. No court has ruled them unlawful, and no agency has blocked one. What regulators are examining is narrower: whether specific deals were structured for the purpose of avoiding a review the law would otherwise require.
FTC Chair Andrew Ferguson said on January 16, 2026 that tech companies had executed acquihire agreements that were not notified under Hart-Scott-Rodino, and that the agency is “beginning to look very closely at how these things work.” He noted the FTC may issue additional guidance, while declining to commit to a bright-line rule, since HSR applies to acquisitions of assets or stock and the question of when a talent deal becomes an asset acquisition varies case by case.
The legal hook regulators keep pointing at is 16 CFR 801.90, an anti-avoidance rule providing that any transaction or device entered into for the purpose of avoiding HSR obligations shall be disregarded, with compliance determined by the substance of the transaction rather than its form.
There is already a track record of inquiries. The FTC opened a probe in June 2024 into Microsoft’s roughly $650 million Inflection AI deal, which paired a model license with hiring most of the company’s staff. It launched an informal inquiry into Amazon’s June 2024 Adept arrangement. The DOJ has been probing whether Google’s August 2024 Character.AI licensing deal violated antitrust law and whether it was structured to avoid scrutiny.
The counterweight is political. The current administration’s posture toward AI is broadly deregulatory, and Warren, Wyden, and Blumenthal are three Democratic senators writing to a Republican-chaired FTC. Law firm analysts at Mintz noted in February 2026 that it remains unclear whether the scrutiny produces enforcement actions, formal guidance, or nothing at all. Founders should plan for the structure to stay legal and available through at least 2027.
What founders should negotiate before signing one
Everything about team protection has to be contractual and has to be settled before the deal closes. Afterward you have no bargaining power and no standing to enforce a promise that was never written down.
Five items to get in writing:
- The named offer list. Which specific employees receive offers, at what titles, at what compensation. A verbal commitment to “take care of the team” isn’t a commitment.
- Severance for everyone not on that list. Fund it from the licensing payment before the money is allocated anywhere else. This is the single highest-impact term for the people with the least power in the negotiation.
- Cliff waivers and vesting acceleration. Cognition did this for Windsurf’s remaining staff. It’s achievable, and it’s worth more to an eighteen-month employee than any other concession you can win.
- What happens to the entity. If leadership leaves, who runs it, who funds it, and what the wind-down looks like if customers walk the way Scale AI’s did.
- Where the money lands. Model the split between purchase price and employment compensation explicitly, and show your common shareholders the real number before they hear a headline figure that has nothing to do with them.
If you are on the receiving end as an employee rather than a founder, the practical question is narrow: are you on the list. Ask directly during any acquisition conversation, and treat an unwillingness to answer as an answer. Common stock behind a preference stack in a licensing deal is worth planning around as zero, which is a different calculation than the one most people made when they accepted the job.
The broader read for founders raising in 2026 is that the exit market has quietly split in two. There is the traditional path, where a company gets bought and its shareholders get paid, and there is the talent path, where a handful of people get extraordinarily wealthy and the corporate entity becomes a formality. Both are called exits. Only one of them pays the cap table.
Anyone building an AI company right now should know which one they are actually being recruited into. That means reading the sector honestly rather than optimistically, the same discipline required by the companies pulling back from AI in 2026, by the infrastructure costs behind deals like Meta’s Georgia data center buildout, and by the broader shift toward independent work we covered in the 5-to-9 economy. The deal on the table may be the best outcome available. Just price it correctly, and make sure the people who built it with you are priced into it too.
