Elizabeth Warren wants Congress to close the AI acquihire loophole, and you should read that as a warning to every Big Tech company buying talent without buying the company.
The AI deal machine has been running on a simple assumption: if Google, Microsoft, Amazon, Meta or Nvidia can hire the people, license the models and leave the corporate shell behind, the hardest part of merger review may never arrive. Warren's June 24 speech says that assumption is now the target.
According to Bloomberg, Warren called on Congress to revamp merger law and sharpen antitrust enforcement around AI companies, arguing that corporations shouldn't get softer treatment because the competitive threat sits inside an algorithm. That's the right fight. Antitrust law is supposed to care about market power, not the costume a deal wears.
The numbers explain why this has become urgent. Bloomberg's tally put Big Tech's spending on these AI talent and licensing structures at more than $40 billion since March 2024. Microsoft paid a reported $650 million to license Inflection AI's technology after hiring Mustafa Suleyman and much of his team. Google paid $2.7 billion for a Character.AI licensing deal while bringing Noam Shazeer and other staff into DeepMind, as the Financial Times reported. Meta's $14.3 billion investment in Scale AI brought Alexandr Wang into Mark Zuckerberg's new superintelligence effort. You don't need to pretend these are ordinary recruiting moves.
Also read: SignalFire data shows engineering hiring is more resilient than almost anyone expected, The AI world has a new obsession and it isn't chatbots, Qualcomm bets its future on a 250-core data center chip and a $3.9 billion software acquisition
The loophole is the point
A normal acquisition gives regulators a clean object to review. There is a buyer, a target, a price and a filing under the Hart-Scott-Rodino Act if the deal crosses the threshold. These AI transactions are messier by design. The buyer gets the founders, the senior researchers and access to the model or data pipeline. The startup remains alive, but often without the people who made it a threat in the first place.
Frankly, that distinction is too cute. If a dominant company can remove a rival from the frontier model race by taking its core team and locking up its technology through a license, the competitive effect can look very close to a takeover. The legal form shouldn't be allowed to do all the work.
Warren has been pushing this line for months. In February, she joined Senators Ron Wyden and Richard Blumenthal in asking the FTC and Justice Department to scrutinize AI deals involving Nvidia, Meta and Google. Their letter argued that reverse acquihires could let dominant firms dodge merger law while weakening smaller rivals. That wasn't a stray complaint. It was a map of where this fight is heading.
You can already see why regulators are interested. The Wall Street Journal reported in 2024 that the FTC was examining Microsoft's Inflection deal to determine whether it was structured to avoid review. The UK Competition and Markets Authority looked at the same transaction and cleared it, but only after finding that Microsoft's hiring of Inflection staff and related arrangements could amount to a relevant merger situation under UK rules. That phrase matters because it treats the substance of the move seriously, even when the paperwork doesn't look like a classic acquisition.
AI makes the old test look thin
The harder issue is that AI markets don't always give regulators the old signs of consolidation. A startup may have little revenue, few customers and no public market share worth measuring. Its value may sit in twenty researchers, a training stack, a model family and a product that hasn't yet found the business model. If you wait until those numbers look mature, the company may already be gone.
That is why Warren's algorithm argument has force. A pricing algorithm can help a company coordinate behavior. A ranking algorithm can decide which seller gets seen. A frontier model can become the layer other businesses build on. If those systems become tools for excluding rivals, steering customers or locking up distribution, you don't need a new moral category for them. You need antitrust law that can see what is happening.
There is a real risk of overreach, and regulators should be honest about it. Not every AI partnership is a disguised merger. Startups sometimes need cloud credits, model access or distribution because training frontier models is brutally expensive. Character.AI's interim chief executive told the Financial Times that building large language models had become insanely expensive, which is why the company shifted toward consumer products after Google's deal. That is a fact, not a talking point.
But cost pressure cannot become a permission slip for incumbents to buy the field one lab at a time. If the only realistic exit for a promising AI team is to be absorbed through a licensing and hiring package, founders will build for the giants instead of against them. Your market can still have hundreds of startups on paper while the important people and assets keep ending up in the same five companies.
Congress doesn't need to ban these deals outright. It does need to make them reviewable when the economic reality looks like control, suppression of a rival or removal of a future competitor. The AI industry has moved faster than the filing rules written for cleaner transactions. Warren is saying the rules have to catch up.
The companies will argue that they are hiring talent, not buying rivals. Sometimes that will be true. The question now is whether regulators and lawmakers are willing to look past the label when the effect is obvious.
