By Sebastian Mallaby
In an exclusive excerpt from his upcoming book, "The Infinity Machine: Demis Hassabis, DeepMind and the Quest for Superintelligence," journalist Sebastian Mallaby illuminates an acquisition race that broke out in 2013 over DeepMind and artificial general intelligence.
On a June weekend in 2013, actress Talulah Riley rented out a castle in Tarrytown, N.Y., to celebrate her husband Elon Musk's birthday.
"It was one of these fake American castles," Demis Hassabis remembered. The men dressed up incongruously as samurai warriors, and Riley arranged for the sumo world champion to be there, all 350 pounds of him. Musk took the wrestler on, throwing him impressively and injuring his own neck in the process.
Musk had been an early investor in Hassabis's AI company, DeepMind, which was three years into its restlessly ambitious pursuit of artificial general intelligence. This was before AI minted a new class of billionaires, before OpenAI and Anthropic existed. Hassabis had envisioned a Manhattan Project for intelligence, hiring the brightest minds in his field and inventing an agent that played multiple Atari videogames.
Google's chief executive, Larry Page, saw Hassabis at the party and proposed a walk with him. The two men strolled around the grounds of the castle, taking in the pointless folly of the battlements and arrow slits. Speaking in a strained whisper, the effect of a rare illness of the vocal cords, Page suggested that Hassabis's company-building endeavors might be similarly pointless. Hassabis was a scientist, and his goal was to create AGI. So why bother with the idea of an independent DeepMind?
"Why don't you take advantage of what I've already created?" Page said. He had recently bought other AI boutiques and was on a hunt for the next one.
"He was basically telling me, maybe you could build a company like Google, but it would take the best part of your career," Hassabis recalled. "But if my real mission was to build AGI, then why don't I use all the resources that he's accumulated? I thought that was a pretty good argument."
"I was fed up with scrambling around, trying to raise money for what I knew was the biggest thing of all time," Hassabis recalled.
"I'll go to Google. I'll get a s -- load of computers and then I'll solve intelligence."
This account is based on over 30 hours of interviews with Hassabis, conducted over three years. It also draws on dozens of conversations with DeepMind colleagues, investors and people involved in the company's acquisition.
In the fall of 2013, Hassabis and his co-founders flew out to the Google headquarters. To keep the acquisition talks secret, they were taken to a business office across the street from the main building. Google's mergers and acquisitions team had assembled a roster of in-house AI experts to assess DeepMind's prowess, and the visitors showed off the recent progress with their Atari agent.
What they did not show was any interest in negotiating the price that Google would pay for their company.
"We thought, the moment we mention money, they'll think we're trying to dash for the door," DeepMind co-founder Mustafa Suleyman explained. "It'll look like we're going to take the cash and head off into the sunset."
Instead, he and Hassabis asked about the research budget Google could provide. They also wanted to talk about safety, which they took very seriously.
If DeepMind was going to be owned by Google, it should be protected by an independent oversight board, Suleyman reasoned. The board would be stocked with scientists, philosophers and other reputable figures, who would have the last say on how AI should be deployed into society. "The basic idea was, look, we have to plan for success," Suleyman explained. "In a success scenario, we can't just have the Google founders using AGI for their own purposes."
To pressure Google on this point, Suleyman drew on his experience as a poker player.
"We told them, we are the best-funded pre-revenue startup in Europe. We've got Peter Thiel, Solina Chau, Elon Musk, all billionaires, all backing us," Suleyman remembered.
"Of course, those people didn't really have our backs -- that's what makes you feel queasy as a negotiator. But in poker, you learn to play the table, not the cards. You size up the other players and then you make your bets, based on your reading of their psychology."
Hassabis generally thought of himself as a chess player rather than a poker player. In chess, there are no hidden cards. The game is open and there is no scope for bluffing.
As it turned out, the bluffing may have been unnecessary. The top team in Mountain View was already experiencing its version of Suleyman's safety worries.
"We thought AI was like atomic energy," Patrick Pichette, the chief financial officer, recalled. "You can make bombs with it, but if you are smart you can also solve climate change with it. So we discussed all the big questions from the get-go. What if it takes off on its own and runs amok? How do we control it?"
With Google evidently willing to bankroll DeepMind's research, and with its leaders attuned to DeepMind's safety concerns, the path toward an acquisition seemed open. But Hassabis and Suleyman were taking nothing for granted. Hoping to push Google to commit to a deal, they flirted with another suitor: Mark Zuckerberg of Facebook.
Zuckerberg had been watching nervously as other tech behemoths built up their AI faculties. Belatedly, he had begun scrambling to catch up, making time on his calendar to woo individual AI researchers, even though he was busy running a company with 6,000 employees and a billion customers.
Suleyman flew out to California to meet Amin Zoufonoun, Facebook's head of corporate development. Zoufonoun welcomed Suleyman to his home, served him a murderous tumbler of whiskey, and teased his guest for wanting ice that would dilute it. Over a series of discussions, Zoufonoun proposed a way of making DeepMind's founders richer than they would be from a Google acquisition. If Facebook bought DeepMind, it would lowball the price it paid for DeepMind shares, but then fork over a vast signing bonus to the founders and their top colleagues.
Suleyman reported back to Hassabis. The money thing was interesting, but money was not their main objective. Meanwhile, Zoufonoun had brushed aside Suleyman's talk about AI governance. Over the coming years, Facebook's indifference to AI safety would come to be well-known. The DeepMind duo already sensed it.
Zoufonoun reported back to Zuckerberg. DeepMind had a strong roster of AI scientists -- and if Facebook didn't buy the company, they would end up in the arms of Google.
Hassabis came out to the West Coast to have lunch with Larry Page, still the strongest suitor. Zuckerberg got wind of his visit and invited him to dinner.
Arriving at Zuckerberg's Palo Alto home, Hassabis administered a subtle test on him. The two men discussed the potential of AI, and Zuckerberg expressed appropriate excitement. But then, as the dinner continued, Hassabis brought up other hot technologies: virtual reality, augmented reality, 3-D printing. Zuckerberg sounded equally excited about all of them.
"That told me what I needed to know," Hassabis said later. "Facebook offered more money, but I wanted somebody who really understood why AI would be bigger than all these other things."
After the dinner, Hassabis got back to Larry Page. "Let's go further," he told him.
Spurned, Zuckerberg's competitive instincts kicked in ferociously. He redoubled his wooing of individual researchers and invited Yann LeCun, the deep learning pioneer based at New York University, over for another recruitment dinner.
What would it take for LeCun to join Facebook, Zuckerberg demanded?
If he couldn't buy DeepMind and acquire a ready-made team, Zuckerberg wanted a famous professor to assemble an AI squad for him. Armed with a virtually unlimited war chest, the professor would pick off the top scientists at less well financed labs -- starting, presumably, with DeepMind.
LeCun said he wasn't going to leave New York or quit his professorship. He assumed that these conditions were deal breakers. The next day, Zuckerberg accepted them.
"Where do I sign?" LeCun responded.
In early December, the DeepMind leadership showed up at the world's biggest machine-learning conference. By now the talks with Google had advanced and they were reviewing rough drafts of the acquisition documents between scientific meetings. Meanwhile, Zuckerberg and LeCun rented a hotel ballroom and announced the creation of a new AI lab in Manhattan.
At the conference, Hassabis bumped into LeCun.
"You're not going to poach all my guys, are you?" Hassabis asked.
"I had just signed on basically to do that," LeCun remembered.
Two weeks later, just before Christmas, LeCun phoned Koray Kavukcuoglu, one of his former students and a key contributor at DeepMind, and offered a huge pay raise to come over to Facebook.
"That was the moment," Suleyman said later, "I thought DeepMind might really fail."
Hassabis scrambled to fight back. He let Kavukcuoglu in on the secret: DeepMind was on the verge of selling itself to Google. The stock options that the DeepMind scientists had mentally written off might soon be worth a fortune. Kavukcuoglu agreed to sit tight for the moment.
Hassabis urged Google to close the acquisition as rapidly as possible.
In late December, a Google team flew into London on a Gulfstream jet and arrived in the offices of DeepMind.
The visitors were shown into a conference room and treated to another series of demos. Google's legendary engineering leader, Jeff Dean, asked to inspect the code that powered the Atari system. Demos were all very well, but Dean knew they could be faked. He wanted to look under the hood to make sure there was a real engine.
"It was a crossing of the Rubicon moment," Hassabis remembered. "The biggest, best company in the world gets to see all your research. If you don't do the deal after that, you'll be crushed. It was high stakes for us."
Dean gave the code a thumbs-up. But what would Google pay for it?
DeepMind had no revenues; its main asset was its people. Google's acquisition specialists had a standard way of valuing "acquihire" transactions of this kind. "We had a price-per-engineer model," Don Harrison, the chief Google negotiator, said later.
Harrison figured that DeepMind had perhaps 30 or 40 technical stars. They were not engineers; they were scientists. Back of the envelope, each one might be worth about $10 million. A tough Canadian lawyer who had helped take Google public, Harrison had hammered down the details of dozens of deals. He seldom met resistance.
On this occasion, however, Hassabis and Suleyman pushed back aggressively. They proposed a valuation for DeepMind that was roughly twice as high as Harrison's.
"Everyone had upset stomachs," Harrison said later. Jeff Dean agreed that DeepMind's target might be excessive.
There were other conditions for a sale. Hassabis was determined to remain in London, and he insisted that the uses of DeepMind's technology should be restricted. Military applications would be banned. An ethics and safety review board -- a committee that would include the DeepMind founders and some external grandees -- should be set up to dilute Google's power over the technology.
"For me, this was a huge problem," Harrison remembered. "I was in front of our board of directors selling a deal that wasn't just about the price. It involved a structure that reduced our control over an asset that we were spending a great deal of money on."
In the end, Google swallowed these concerns because of Hassabis. "There is no way we would've agreed to the structure without being absolutely convinced that Demis represented the future of our AI strategy," Harrison said later.
At the end of January 2014, Google bought DeepMind for $650 million, a bargain by today's standards. But the real payoff for Hassabis came over the next decade, as Google poured billions into DeepMind's research. The quest for superintelligence, which Hassabis had harbored since his teenage years, would soon go into overdrive.
From "The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence," by Sebastian Mallaby, to be published on March 31, 2026, by Penguin Press, an imprint of Penguin Publishing Group, a division of Penguin Random House LLC. Copyright (c) 2026 by Sebastian Mallaby.
(END) Dow Jones Newswires
March 25, 2026 12:00 ET (16:00 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments