In recent years, the US technology industry invented a new kind of mergers-and-acquisitions transaction. Traditionally there were two main sorts of M&A transaction:

1. The acquirer buys the target company, in a merger or stock purchase; the shareholders of the target company get something (cash, acquirer stock) in exchange for their shares of the target, and the acquirer ends up owning the target corporate entity.
2. The acquirer buys the assets of the company, in an asset purchase; the target company gets something (cash, acquirer stock) in exchange for its business. If the target is selling all of its assets, it will be left as an empty shell, and will probably distribute the cash or stock to its shareholders and liquidate.

These are both pretty normal transactions, and many rules are set up to treat them similarly. So, for instance, Delaware corporate law requires a shareholder vote for a merger, or for a sale of “all or substantially all” of a company’s assets. US antitrust law requires a “Hart-Scott-Rodino premerger notification report” — where the acquirer tells antitrust regulators about the deal and gives them the opportunity to object — for any acquisition of more than $133.9 million worth of stock or assets. You might get different accounting or tax treatment for different deal structures, but many of the big-picture effects are the same however you structure the deal.

The antitrust part is important. The big tech companies are very big, there is a lot of political and regulatory worry that they are monopolists, and there is at least some winner-takes-all flavor to the AI race. Meta or Alphabet or Nvidia might really want to do a $20 billion AI acquisition here or there, but that might be risky from an antitrust perspective. So they’re in the market for ways to do acquisitions without, technically, doing acquisitions.

Many early-stage artificial intelligence startups consist mostly of (1) people and (2) ideas. The startup has some AI researchers, and they’ve come up with some model or harness or chip design or robot or whatever, and it’s pretty promising. But the startup doesn’t have, like, a giant portfolio of data-center real estate, or a robot assembly line, or a bunch of customer contracts. 

If a big tech company wants to acquire a startup like that, it could buy it in a merger, or buy all of its assets. But the new structure is:

1. The acquirer hires the target’s employees, or most of them, or at least the people who are critical to its valuable ideas.
2. The acquirer gets a waiver, from the target’s board of directors, allowing it to do all that hiring. (California doesn’t enforce non-compete agreements, but if you are going to, like, have the chief executive officer of the target email all of the target’s employees from her work email saying “hey come with me to the acquirer; we’ll match your salary,” then it’s probably good to get the target’s board’s permission.)
3. The acquirer licenses the target’s intellectual property: The target enters into a nonexclusive license allowing the acquirer to use the target’s ideas, patents, chip designs, etc., for its own purposes.
4. In exchange for the waiver (Step 2) and license (Step 3), the acquirer pays the target a big pile of cash.
5. The target is now more or less an empty shell. It has lost most of its employees, or at least the ones who know about the stuff that made it valuable. It still technically owns its intellectual property — it gave the acquirer a non-exclusive license — but everyone who knows anything about that intellectual property has decamped for the acquirer, so it can’t do anything useful with it. It’s just a collection of miscellany, plus the big pile of cash that it got from the acquirer.
6. So the target pays out the cash to its employees. Then maybe it shuts down, or maybe it keeps going in some less ambitious form.

This structure is called a “reverse acqui-hire.”^[1]

This structure has some disadvantages, compared to a traditional merger. It is not especially tax-efficient; the target pays tax on the licensing fee.^[2] Because it leaves the target as an independent company, and gives the acquirer only a nonexclusive license, it is not quite as good as a merger from a competitive perspective: What if your biggest competitor manages to strike its own licensing deal with the stub target company?

But it has some pretty important advantages. For one thing, it probably doesn’t require a vote of the target’s shareholders. (Just the board of directors.) This allows it to happen more quickly than a merger. It might also allow the reverse acqui-hire to happen over the objections of a majority of shareholders, though that’s probably not a big factor. (The big shareholders of an early-stage AI startup will probably be mostly (1) venture capitalists who are represented on the board and (2) co-founders and other employees who have some leverage over getting acqui-hired.) But in theory, a reverse acqui-hire that is good for the employees and bad for the outside shareholders might be doable, whereas a merger that is good for employees and bad for outside shareholders would be harder to pull off.^[3]

The main advantage, though, is probably antitrust. If it’s not a stock or asset acquisition, it doesn’t require an HSR filing, which means you can close the deal before the antitrust regulators can do anything to stop it. I once explained this deal structure with a thought experiment: “Imagine if mergers and acquisitions were illegal.” The reverse aqui-hire is a way to do a merger without the legal trouble of antitrust review. 

This is all pretty controversial, for more or less the reasons that you’d expect. If you think of a reverse acqui-hire as a deal structure that was invented to avoid antitrust scrutiny, then it will obviously annoy people who want more antitrust scrutiny. Bloomberg News reported earlier this year:

The US Federal Trade Commission is taking a closer look at moves by major tech companies to hire away the employees of a startup instead of buying the venture outright, the agency’s Chairman Andrew Ferguson said, addressing a growing practice seen as a way to avoid antitrust review.

“We are beginning to examine these acqui-hires to make sure they are not an attempt to get around” the agency’s merger review process, Ferguson said ... in an interview on Bloomberg Television.

And the New York Times reported last month:

The Justice Department is investigating whether Nvidia tried to skirt antitrust scrutiny of a deal it struck with the artificial intelligence chipmaker Groq last year, according to two people with knowledge of the inquiry.

Groq described the deal at the time as a “nonexclusive licensing agreement” that gave Nvidia access to Groq’s chips, which are custom-built for running A.I. As part of the arrangement, Groq’s chief executive, Jonathan Ross, and its chief operating officer, Sunny Madra, joined Nvidia. ...

Government officials have considered whether these deals should receive more scrutiny while they grapple with the way a small number of companies have consolidated power over A.I.

Conceptually, if reverse acqui-hires are another sort of M&A transaction, then they should be covered by antitrust law. There are some line-drawing problems there — just hiring someone probably shouldn’t trigger antitrust review — but you can see why the regulators are interested.

Similarly, though, if reverse acqui-hires are another sort of M&A transaction, then they should be treated that way under corporate law. A merger, or a sale of all of a company’s assets, triggers certain corporate requirements: Directors have a fiduciary duty to maximize value in a sale,^[4] and the shareholders have to vote to approve the deal. If a reverse acqui-hire effectively means selling all of a company’s assets for cash — because its assets are basically its people and ideas — then shouldn’t that trigger the same requirements?

The Financial Times reports:

Nvidia’s $20bn licensing deal with Groq fleeced employees who were left out of the transaction, according to a lawsuit filed by two engineers who owned shares in the start-up. 

The legal complaint filed on Friday in a Delaware corporate law court alleges that Groq’s board improperly sold the company’s core assets and top employees to the $5tn tech titan in 2025, consigning those left behind to a hollowed-out shell.

The transaction was framed by Nvidia at the time as a “non-exclusive” licensing arrangement, with Groq remaining as an independent and functioning company.

But, according to the lawsuit, the Groq board was riven with conflicts of interest and failed to observe its legal obligation to get the best price and structure for all shareholders, not allowing some of them to vote on the transaction.

Here is the complaint, whose basic point is “oh come on this is a merger”:

Groq’s board of directors (the “Board”) sold the Company to NVIDIA without the stockholder vote Delaware law requires and without any process designed to test or maximize the value of what NVIDIA bought. A majority of the Board was conflicted. It handed NVIDIA Groq’s valuable technology and the engineers who built Groq, took billions of dollars in benefits for itself, senior management, and affiliated funds that it did not share with Groq’s other stockholders, and then squeezed out those stockholders at a lowball price it set itself.  …

But a merger that avoids antitrust rules:

A stock or asset acquisition was unattractive because, given NVIDIA’s market share, a Groq merger would have faced serious antitrust scrutiny. …

Because the buyer acquires neither voting securities nor assets, the transaction falls outside the federal antitrust regulators’ premerger notification regime. ... The structure has become increasingly common in recent years, promoted by venture investors and savvy transactional attorneys. …

NVIDIA had reason to prize that avoidance. A prior attempt to acquire Arm Holdings from SoftBank drew sustained multi-jurisdictional antitrust scrutiny and NVIDIA ultimately abandoned the transaction without closing. An acquirer that has watched a headline acquisition of a semiconductor licensor collapse under regulatory pressure has a powerful institutional incentive to design its next acquisition so that no filing is required.

The complaint also argues that, while technically the license to Nvidia was non-exclusive and Groq continued as an independent company, in practice there was nothing left:

The supposedly “non-exclusive” license grant was substantively an asset transfer. …

No rational third-party licensee will pay significant value for a non-exclusive right to technology that the world’s dominant AI chip company already holds on the same or broader terms ... and that it is updating daily with the very workforce that created the underlying intellectual property. The canvass of 36 parties produced no compelling bid [for the remainder of the company] for that reason: once Groq’s Board agreed to let NVIDIA develop the assets, whatever interest Groq retained in them ceased to have independent market value.

This all seems about right to me, but I’m not sure where it gets you. Part of the argument here is that the reverse acqui-hire is a novel form of M&A that should be treated like other forms of M&A, and should have to be fair to shareholders.

But the other part of the argument is that this deal wasn’t fair to shareholders, and I’m not sure. Groq’s shareholders — including the plaintiffs in this lawsuit — weren’t zeroed in the transaction. Nvidia paid $17 billion to Groq, in cash, for the license, and most of that money has been paid out to shareholders.^[5] Their complaint is that other shareholders — the founders and employees who went to Nvidia — got more. The headline value of the Nvidia acqui-hire was $20 billion, that is, the $17 billion paid to Groq plus $3 billion paid to employees for joining Nvidia. The lawsuit objects:

[Jonathan] Ross’s position illustrates the conflict. He was simultaneously a Groq stockholder, Groq’s Chief Executive Officer, a Groq director, the principal negotiator of the Company’s side of the License, and the individual NVIDIA most wanted to hire. He received NVIDIA equity compensation from the $3 billion pool on top of the proceeds of his Groq ownership. An ordinary common stockholder could monetize the value surrendered to NVIDIA only through the consideration allocated to her shares; she could not take a discount on those shares and be paid separately for following the technology to NVIDIA. Ross could, and did.

Yeah but that’s true in any merger where the CEO keeps his job. The value of Groq was largely in its employees, and Groq’s shareholders cleared $17 billion by selling the rights to those employees to Nvidia. Of course the employees also got paid. 

Frontier artificial intelligence labs are locked in bitter competition for the best researchers: The lab with the best researchers will build the best model and make a zillion dollars; the lab with the second-best researchers might fail. It is critical, for normal shareholder-value-creation reasons as well as more fanciful save/destroy-the-world reasons, to hire the best researchers. And so this competition largely takes the normal form: Big AI labs pay their researchers lots of money.

But that’s not the only form it can take. There seems to be a strong empirical correlation between (1) being a good AI researcher and (2) being interested in effective altruism. (You might wonder if this is a historical accident: Early AI research was an obscure subculture with a lot of overlap with EA, but now that “AI researcher” is a highly legible and lucrative career path, maybe the correlation will weaken. Or maybe not; maybe there is some inherent connection between being good at AI and having EA views.)

And so a lot of researchers at top AI labs have some “earning to give” tendencies, donating large portions of their massive paychecks to charities. (Because they are EA-oriented, many of them give to charities whose mission is to stop their employers from building superintelligent AI. Is that … effective?)  

If you ran an AI lab, you might design your compensation practices around these facts. You might recruit researchers by offering to pay them a competitive headline number —  like, $20 million a year — but split into (1) $5 million paid directly to them and (2) $15 million paid directly to the charity of their choice. This compensation design has several key advantages:

1. It will appeal to EA types, and a lot of good AI researchers are EA types. You will maximize shareholder value by recruiting the best AI researchers. (Again, perhaps this link is weakening over time.)
2. If you’re the founder of an AI lab, you might be an EA type yourself, and you might prefer to hire employees who funnel their large paychecks to charity rather than to yachts. 
3. Relatedly, we often discuss the fact that the competitive salary for many AI researchers is enough to retire on immediately. You do not want to hire researchers who retire immediately. You want them to keep working. Making sure that most of their paycheck goes to charity, rather than to savings, is helpful for retention and thus maximizes shareholder value.
4. If you are paying the charitable donations directly, rather than through the employees, it is possible that (1) you can refuse to direct the donations to Stop AI Labs From Doing AI charities and/or (2) the Stop AI Labs From Doing AI charities will refuse to take donations from you, which is maybe also helpful for shareholder value.^[6]

At the Information, Cory Weinberg reports that Anthropic does in fact have more or less this compensation structure, which makes for slightly unintuitive accounting and disclosure:

When Anthropic shared financial figures recently with prospective investors in its planned initial public offering, it included a surprising glimpse of the enormous riches about to hit nonprofits and philanthropies—and the cost to shareholders.

The company reported a non-cash expense of more than $660 million for stock used to match employees’ charitable contributions in the six months from October 2025 to March 2026, investors said. That expense, which is already unusually large, will likely balloon into the billions of dollars in the coming quarters, after an IPO expected to create what could be an unprecedented wave of corporate donations.

The charges are tied to stock awards to employees that they pledge to charities that Anthropic matches with stock of its own. Anthropic executives have said the unusually generous matching program is an important recruiting tool, and they have emphasized how it underlines its public benefit mission. And it has become particularly valuable because of Anthropic’s rapidly rising private stock value. …

Anthropic excluded the expense, along with stock-based compensation and a large legal settlement, from its adjusted operating profit for the periods. Companies typically present investors with adjusted profit metrics that exclude expenses they don’t pay in cash or aren’t likely to recur.

The charitable expense that Anthropic excluded was far larger than any large publicly traded firm that excluded a similar expense from their adjusted profit metrics. “The adjustment for charitable contributions is not common at all,” said Calcbench’s Ghai.

No, right, of course, most companies do not report adjusted earnings that adjust out non-cash charitable giving, because “non-cash charitable giving” is not a big thing at most companies. But tech companies often do adjust out stock-based employee compensation,^[7] and that’s obviously what’s going on here. It’s just that the way Anthropic compensates employees is through charitable giving.

Also, here’s evidence that the AI/EA correlation might be weakening over time:

The company matched stock earmarked for charities on a 3-to-1 basis for early employees, who joined in 2024 or earlier. Employees had the option to donate up to half their stock to charity as part of the program. For employees hired more recently, the company matches donations of up to a quarter of their stock on a 1-to-1 basis.

If you joined Anthropic in 2023, there was a high probability that you cared a lot about funneling your potential AI riches to effective charities. In 2025, less. Eventually, after the IPO, they’ll probably phase it out.

Sure fine:

Citigroup Inc. is slashing the time its investment-banking analysts need to wait to advance to associate level, part of a push by the bank to attract and retain top talent in a fierce job market that’s seen buyout firms poach aggressively from Wall Street.

The move will cut the analyst program to two years from three for Citi’s junior bankers, accelerating their route to seniority at the firm and bringing the company in line with some competitors, David Friedland, co-head of North America investment banking, said in an interview.

I thought a two-year path to associate was pretty standard these days. I’m also not quite sure how this works to compete with private equity. Like the timeline is:

1. You start your analyst program at Citi in summer 2026.
2. In about January 2027, give or take, the big private equity firms start interviewing investment bank analysts to start in private equity in fall 2028.
3. You interview and get an offer.
4. What, you say no because you might get promoted to associate in 2028? 

Maybe — maybe banking is a more attractive long-term proposition than private equity these days — but there’s still that timing mismatch. You still get your private equity offer way, way, way before you get your associate promotion.

By the way, that Citi story goes on:

The move revives a discussion around junior-banker retention that arose last year, when bank executives pushed back at private equity firms for bringing forward their hiring cycles. ...

JPMorgan said it would fire any analyst who accepted an outside job offer within 18 months of joining the firm. 

What ever happened to that? JPMorgan really did say, in June 2025, that it would fire analysts if they accepted an outside job offer within 18 months of starting. That really did have the effect of delaying private equity “on-cycle recruiting”: No private equity firm wanted to incur Jamie Dimon’s wrath, so they all waited longer than they usually did to recruit associates for 2027. But only a few months longer: Private equity recruiting for 2027 kicked off in January 2026. By my math, if you are an investment banking analyst at JPMorgan, you started in June 2025, and you took a private equity offer in January 2026, then you had been at JPMorgan for about seven months. That’s a lot longer than in previous cycles, but it’s a lot less than 18 months. Did no one at JPMorgan take a private equity offer for 2027? Did they all get fired? Are they all hiding their offers? Has JPMorgan given up on this? Does Dimon know?

A lot of finance is about finding tricky ways through the rules. We talked last week about 351 exchange traded funds, which rely on creatively splicing together non-taxable transactions to make it so anyone can trade stocks without incurring taxes. Someone had to notice that possibility and splice together the pieces. People who can do that sort of noticing are in high demand at law firms and investment banks and hedge funds.

And so there is a certain amount of lore about financial firms being delighted to hire people based on their track record of finding tricks. Last year the Wall Street Journal recounted a story about Marshall Wace, the hedge fund:

[Co-founder Ian Wace] once challenged a group of prospective hires to visit the most countries in Europe, using as many forms of transportation and as cheaply as possible—within 24 hours. 

“I wanted to see to what extent someone could think pretty creatively. I wanted to see, if you exhausted them, what would happen,” Wace said. Among the notable results? One made it to the far side of Poland “on something like 30 different forms of transport,” Wace recalled. Another stayed local, traversing London’s many embassies.

That candidate is at the firm now. “He’s actually very clever,” Wace said. But, as he saw it, “it was a bit too cute.”

“It was a bit too cute,” he said, but he didn’t mean that, because he hired the guy. Exploiting technicalities in team-building challenges is a very good sign for your future in the financial industry. But you cannot take this too far. The Financial Times reports:

A former HSBC banker has been banned from working in financial services by the UK watchdog after being convicted of fraud over a “doughnutting” scam he devised to dodge almost £6,000 of train fares. …

The 53-year-old devised a “doughnutting” scheme to buy tickets for short trips at the start and end of the journey from his home in Orpington, south London, to HSBC’s office in Canary Wharf. The scam meant he avoided paying for a “hole” in the middle of his commute.

The former banker also obtained a Jobcentre Plus discount, for which only unemployed people or those seeking work are eligible, to get 50 per cent off ticket prices.

He was banned because cheating on train fare “shows he lacks honesty and integrity,” fine. It’s a little creative, but mostly it’s just grubby. Also he “was head of passive equity at HSBC Asset Management,” and I am not sure that sneaky ingenuity is what you want in a passive equity investor. 

Hedge Funds Exiting Crowded Trades Fueled Rout in French Bonds. How the French Bond Trade Backfired on Investors. Paramount’s $52 Billion Debt Saga Ends With Hair-Raising Finale. Paramount Debt Drop Spells Trouble for Borrowers. Paramount’s $52 Billion Debt Sale Shows How Higher Rates Are Biting Corporate America. Goldman’s Bid for CLO Manager Highlights Wall Street’s Fee Hunt. Blue Owl Again Caps Two BDCs After 39%, 17% Withdrawal Requests. Spending on AI Is Becoming Almost Impossible for Businesses to Budget. Jane Street Denies Market Manipulation to India Securities Court. Low-profile hedge fund smashes record for New York office rent. London Takes a Hit as Super Rich Worth $160 Billion Exit UK. Sweden’s Young Men Have Found Their Wolf of Wall Street. Sotheby’s dinosaur auction shows fossils boom is not dying out.

If you'd like to get Money Stuff in handy email form, right in your inbox, please subscribe at this link. Or you can subscribe to Money Stuff and other great Bloomberg newsletters here. Thanks!