New AI deal: Buy everything but the company

Erin Griffith and Cade Metz

In 2022, Noam Shazeer and Daniel De Freitas left their jobs developing artificial intelligence at Google. They said the tech giant moved too slowly. So they created Character.AI, a chatbot start-up, and raised nearly $200 million. Last week, Shazeer and De Freitas announced that they were returning to Google. They had struck a deal to rejoin its A.I. research arm, along with roughly 20 percent of Character.AI’s employees, and provide their start-up’s technology, they said. But even though Google was getting all that, it was not buying Character.AI.

Instead, Google agreed to pay $3 billion to license the technology, two people with knowledge of the deal said. About $2.5 billion of that sum will then be used to buy out Character.AI’s shareholders, including Shazeer, who owns 30 percent to 40 percent of the company and stands to net $750 million to $1 billion, the people said. What remains of Character.AI will continue operating without its founders and investors.

The deal was one of several unusual transactions that have recently emerged in Silicon Valley. While big tech companies typically buy start-ups outright, they have turned to a more complicated deal structure for young A.I. companies. It involves licensing the technology and hiring the top employees — effectively swallowing the start-up and its main assets — without becoming the owner of the firm. These transactions are being driven by the big tech companies’ desire to sidestep regulatory scrutiny while trying to get ahead in A.I., said three people who have been involved in such agreements. Google, Amazon, Meta, Apple and Microsoft are under a magnifying glass from agencies like the Federal Trade Commission over whether they are squashing competition, including by buying start-ups.

“Large tech firms may clearly be trying to avoid regulatory scrutiny by not directly acquiring the targeted firms,” said Justin Johnson, a business economist who focuses on antitrust at Cornell University. But “these deals do indeed start to look a lot like regular acquisitions.” In a statement, Google said it was “thrilled” that Shazeer was returning alongside some of his colleagues and declined to comment on antitrust scrutiny. On Monday, a federal judge issued a landmark ruling that found Google had violated antitrust law by abusing a monopoly in online search. A Character.AI spokeswoman declined to comment beyond the announcement of the Google deal. The Information earlier reported on the deal’s details. Since the A.I. boom took off in late 2022, it has transformed tech deals. Investors initially raced to pour money into A.I. start-ups at high valuations. That led to an unusually frenzied pace, with start-ups such as Anthropic raising large sums frequently and agreeing to various funding conditions, such as using chips and cloud computing services from the companies that invested in them.

That excitement cooled as it became clear that some high-profile A.I. start-ups would not succeed, creating an opportunity for big tech companies to swoop in with nontraditional deals. Microsoft kicked off the trend in March when it agreed to pay the A.I. start-up Inflection more than $650 million to license its technology and hire almost all of its employees, including its founder, Mustafa Suleyman. Suleyman, an A.I. veteran, now leads Microsoft’s consumer A.I. business. In June, Amazon inked a similar deal with the A.I. start-up Adept, bringing on many of its employees, including its founder, David Luan. Amazon paid Adept at least $330 million to license its technology, with much of the money going toward paying back the $414 million that the start-up had raised from investors, three people with knowledge of the transaction said. Amazon also offered a $100 million retention bonus to Adept employees who joined, the people said.

Regulators are watching. The F.T.C. is working on a broad study of A.I. deals between start-ups and Microsoft, Amazon and Google, the agency said in January. It is also investigating whether Microsoft should have notified regulators about the Inflection deal, which would have subjected the arrangement to more immediate scrutiny, a person with knowledge of the matter said.

On Thursday, Britain’s antitrust regulator said it was investigating an investment deal that Amazon had made with Anthropic. Silicon Valley has embraced the unusual deals because start-up founders can continue working on their technology with the resources of a large company, without worrying about making money on their own.

The transactions can also provide a fast return for investors. Investors in Character.AI, which was privately valued at $1 billion, made a two-and-a-half-times return from the Google licensing deal after just two years. With the Adept and Inflection deals, most investors got their money back, people familiar with the transactions said.

Yet the transactions have also left behind orphaned corporate entities, stranding remaining employees at start-ups where the founders and investors have moved on. Those employees do not get to partake in the financial spoils of these deals.

That has caused consternation among some tech investors and entrepreneurs. “If you build a company and you take on money from investors, every person involved deserves to be rewarded,” said Sebastian Thrun, an A.I. researcher and serial entrepreneur known for founding Google’s self-driving car project. “This is why Silicon Valley emerged. If you water things down, it will be hard for the ecosystem to survive.” Matt Turck, an investor at the venture firm FirstMark Capital, said he hoped these types of deals would not continue because they created “a messy structure that breaks down the alignment” of founders, employees and investors.

More of these deals may surface. Many A.I. start-ups have raised huge sums on wildly ambitious goals, and large acquirers remain eager to pay for the best talent, ideas and products. At the same time, some of the start-ups are struggling to make money and to compete with the bigger players, so they may be more willing to entertain deal talks. “Founders and investors realise that not every high-profile A.I. start-up with great founders is going to be the next OpenAI or Google,” Turck said.