In March 2024, Microsoft agreed to pay about $650 million to Inflection AI, a start-up that had raised $1.3 billion less than a year earlier from backers including Microsoft itself, Nvidia, Reid Hoffman, Bill Gates and Eric Schmidt. The deal was not, formally, an acquisition. Microsoft hired Inflection's co-founders, Mustafa Suleyman and Karén Simonyan, along with most of its roughly 70 staff, and paid a licence fee for the company's models. Mr Suleyman became chief executive of a new Microsoft AI division. Inflection, stripped of its leaders and much of its team, pivoted to selling software to businesses.

Investors were reportedly made roughly whole, receiving about 1.1 to 1.5 times their money. Nobody lost a fortune. But the shape of the transaction was striking: the people who ran the company moved to one of the most powerful jobs in the industry, the buyer secured the talent it wanted without a conventional merger review, and the start-up that had raised $1.3 billion to build a consumer AI assistant effectively ceased to exist in its original form.

Within months the template had been copied. In June 2024 Amazon hired David Luan, co-founder of Adept, and several colleagues while licensing Adept's technology. In August 2024 Google struck a licensing deal with Character.AI that brought its co-founders, Noam Shazeer and Daniel De Freitas, back to Google; the arrangement was widely reported at about $2.7 billion. In July 2025 Google paid about $2.4 billion to license technology from the AI coding start-up Windsurf and hire its chief executive, Varun Mohan, and a group of researchers, days after a planned $3 billion sale to OpenAI collapsed. And in June 2025 Meta invested about $14.3 billion for a 49% stake in Scale AI, whose founder, Alexandr Wang, left to lead a new Meta superintelligence unit.

Read more: Town reaches unicorn valuation with $108 million round four months after launch

These are the most visible examples of a much older Silicon Valley practice. Call it the founders' scheme: a transaction in which the people who control a start-up negotiate an exit that rewards them personally, through jobs, retention bonuses and new equity at the buyer, while the company itself is hollowed out and its other stakeholders take what is left. It is usually legal. It is frequently celebrated in press releases. And it is almost never challenged in public by the venture capitalists who funded it.

The waterfall problem

To understand why these deals exist, start with how start-ups are financed. Venture firms such as Andreessen Horowitz, Sequoia Capital, Accel and Benchmark typically buy preferred shares. Those shares carry a liquidation preference, most commonly 1x: in a sale, preferred holders get their money back before ordinary shareholders, including founders and employees, receive anything.

Read more: Tolan raises $30 million for AI companion focused on everyday life

That protection works well when a company sells for more than it has raised. It creates a deadlock when it sells for less. Imagine a start-up that has raised $60 million and receives a $25 million offer. Under the standard waterfall, all $25 million goes to preferred investors. Founders' and employees' common shares are worth nothing.

A founder facing that arithmetic has little reason to spend months on due diligence, customer transitions and code handovers for a deal that pays them zero. Buyers know this. So the money is restructured. Part of the purchase price is moved out of the deal and into employment: signing bonuses, retention payments, restricted stock units at the acquirer and, sometimes, a "management carve-out" plan that sets aside a slice of sale proceeds for executives before preferred holders are paid.

The result can be that the founders earn more from their new employment contracts than investors recover from the sale of the company.

A hypothetical cap table

Consider an illustrative company — the figures below are a worked example, not a real start-up. It has raised $60 million in preferred stock with a standard 1x liquidation preference. Its last funding round valued it at $500 million. An early engineer joined four years ago and has fully vested options over 0.5% of the company, which on paper were worth $2.5 million at that valuation.

A large technology company now offers $25 million. In a conventional sale, the waterfall is simple: the first $60 million belongs to preferred holders, so all $25 million goes to investors. Common shareholders, including the engineer and the founders, receive nothing.

Now restructure the same money as a licensing-and-hiring deal. The buyer pays $25 million for a non-exclusive licence to the technology, which flows to the start-up and is used to return part of investors' capital. Separately, the buyer offers the two founders and a dozen senior researchers new jobs, with retention packages that in this example total $30 million in restricted stock over four years. Because the company was never sold, there is no change-of-control event. Options for staff who are not hired can sit inside a shrinking shell and may never be exercised at value.

The total cost to the buyer is $55 million instead of $25 million, but more than half of it is paid as compensation to the people it wants. That is the core of the arrangement: the buyer pays for talent, not for the company, and the company's own shareholders are paid last.

The case that defined the rules

The leading legal example is not in California but in Delaware, where most venture-backed companies are incorporated. In In re Trados Inc. Shareholder Litigation, decided by the Court of Chancery in 2013, the translation-software company Trados was sold in 2005 to SDL for about $60 million. Preferred investors received about $52 million. A management incentive plan paid executives about $7.8 million. Holders of common stock received nothing.

A common shareholder sued, arguing that the board, dominated by venture investors, had favoured preferred holders and management over common. Vice Chancellor J. Travis Laster found that the directors had not followed a fair process: their duty, he wrote, ran to the common stockholders as the residual owners, not to the preferred. Yet he declined to award damages, concluding that the common stock had no meaningful value under any realistic outcome, so the price was fair.

The ruling has shaped every distressed sale since. It warned boards that carve-outs and preference payouts face scrutiny, and it showed founders and investors that, if the company is genuinely underwater, a well-documented process will probably survive challenge. Lawyers now routinely advise boards to form special committees, obtain fairness opinions and record why a carve-out was necessary to close a deal at all.

An older playbook

Acqui-hires long predate the AI boom. In 2009 Facebook bought FriendFeed for a reported $47.5 million, largely to recruit its founders, among them Bret Taylor, who went on to become Facebook's chief technology officer, later chairman of Twitter and OpenAI, and co-founder of Sierra. FriendFeed's backer, Benchmark, made a respectable return; the product itself faded.

In 2010 Facebook bought Drop.io, a file-sharing service, and shut it down; its founder, Sam Lessin, became a Facebook product executive. Google, Yahoo and Twitter ran dozens of similar deals in the early 2010s, many of which closed products within weeks while founders joined as directors or vice-presidents.

Those deals were often defensible: a small team with modest funding and a product that was not working found a home, and investors recovered something. The problem grew as round sizes swelled. When a start-up has raised tens or hundreds of millions of dollars, the gap between what investors put in and what a talent deal pays out becomes large, and the question of who receives the money matters a great deal more.

The 2022 to 2024 correction made that gap visible. As interest rates rose from near zero to above 5% and software valuations fell sharply, hundreds of companies that raised at 2021 prices could not raise again. Olive AI, a health-care automation start-up once valued at about $4 billion, shut down in 2023 and sold its business lines in pieces. Many smaller companies quietly sold teams to larger rivals. Founders announced they were "joining" a new employer. Rank-and-file staff learned their options were worthless.

The deal tracker

The largest recent arrangements share a structure: a licence to the technology, a transfer of key staff, and a remaining company that continues in smaller form. Values are as publicly reported.

In each case the start-up survived as a legal entity, which is what separates these deals from a conventional acquisition.

The reverse acqui-hire and the regulators

The AI-era deals added a new element: regulation. A conventional acquisition above the thresholds of the Hart-Scott-Rodino Act must be notified to the Federal Trade Commission and the Department of Justice Antitrust Division before it closes. Hiring people and licensing technology generally does not trigger the same filing.

Regulators noticed. In 2024 the FTC opened an inquiry into whether Microsoft's Inflection deal was structured to avoid merger review. The UK's Competition and Markets Authority examined it as a merger and cleared it in September 2024, concluding it qualified as one but did not substantially lessen competition. The FTC also reportedly examined Amazon's arrangement with Adept, and US senators wrote to regulators asking for scrutiny of the wider pattern.

For founders, the reverse acqui-hire offers something even better than a traditional soft landing: a headline. "Joins Microsoft to lead consumer AI" reads as a triumph. The company left behind, its remaining employees and its investors' long-term upside become footnotes.

The Windsurf episode showed the cost. After Google hired Mr Mohan and a group of senior staff, many of the roughly 250 remaining employees were initially left at a company without its leaders. Within days the AI developer Cognition agreed to acquire what was left, and said it would accelerate vesting so that all staff shared in the outcome. The rescue was widely praised precisely because it was unusual.

Why the venture firms stay quiet

The investors with the most legal standing to object are the venture firms on the board. They almost never do so in public. Their silence has three causes.

The first is reputation. Venture capital is an access business: the best founders choose their investors. A firm known for suing founders, or for blocking a founder's soft landing, risks being labelled hostile, and losing the next competitive deal to a rival. Firms such as Founders Fund, Greylock, Kleiner Perkins and General Catalyst market themselves on being on the founder's side. Greylock, Kleiner Perkins and General Catalyst were among Windsurf's backers; none publicly criticised the Google arrangement.

The second is the power law. In a typical venture portfolio, one or two companies produce most of the returns. A firm that backed an early Stripe, Airbnb or Facebook does not need to recover a few million dollars from a failed bet; it needs partners to spend their time on the winners. Litigation over a write-off is expensive, slow and distracting.

The third is reporting. A bankruptcy or a public dispute forces difficult explanations to limited partners, the pension funds, university endowments and foundations whose money venture firms invest. A quiet sale described as an "acquisition" is easier to report than a failure, even when the economics are similar.

Venture firms also have their own interests in the outcome. Inflection's investors included Microsoft, the buyer. Large firms frequently sit on both sides of the ecosystem, backing start-ups that later sell to companies in which they also hold stakes or with which they want to keep good relations. Publicly attacking a big buyer can cost future exits.

Who pays

When the founders' scheme works, it has clear winners: founders who secure senior jobs and compensation, buyers that acquire talent without a full merger, and venture firms that avoid an ugly write-off. The costs fall elsewhere.

Early employees are the most exposed. They accept below-market salaries in exchange for stock options that sit at the bottom of the waterfall. In a carve-out, bonus pools rarely extend beyond a small group of senior staff. Those not hired by the buyer lose both their jobs and their equity.

Limited partners bear a slower cost. If founders know they can land softly whatever happens, they have less reason to manage cash carefully, and less fear of failure. Across thousands of companies, that weakens the discipline that venture capital is supposed to impose.

Markets lose information too. When failures are announced as acquisitions, it becomes harder for employees, customers and future investors to judge which companies actually succeeded.

What could change

Several remedies are discussed in legal and investor circles. Boards could be required to disclose side employment and retention arrangements to all shareholders when a company is sold. Carve-out plans could be extended to cover more employees rather than only executives. Investors could negotiate clauses giving common holders a minimum share of proceeds in talent deals. Regulators could treat large hire-and-license arrangements as mergers for notification purposes, as the CMA effectively did with Inflection.

None of these would end acqui-hires, which can be a sensible outcome for small teams. They would make it harder to describe a collapse as a triumph, and harder for founders to collect the rewards of a sale while others absorb the loss.

Until then, the founders' scheme is likely to remain one of the technology industry's most reliable exits: privatising the upside for those at the top, and quietly socialising the failure for everyone else.