Mark Zuckerberg offered more money than Google for DeepMind in 2014. He still lost — because at a dinner at his Palo Alto home, Demis Hassabis came away with a quiet conviction that the two men were not aligned on what AI actually was.
The dinner, as Hassabis described it in an excerpt from Sebastian Mallaby's upcoming book "The Infinity Machine" (published March 31) reported through the Wall Street Journal, was structured as a kind of informal test: when Hassabis steered the conversation toward other emerging technologies — virtual reality, augmented reality, 3D printing — Zuckerberg lit up about all of them with equal enthusiasm. "That told me what I needed to know," Hassabis said. "Facebook offered more money, but I wanted somebody who really understood why AI would be bigger than all these other things."
For Hassabis, that was the decisive contrast. As the Times of India account of Mallaby's reporting describes it, building AGI was not one of several exciting technology bets for Hassabis — it was a consequential pursuit in human history. He needed a partner who understood that distinction. Zuckerberg, treating it as just another item on a list of shiny innovations, had signaled he did not yet share that view.
Facebook's offer was financially serious. According to the Times of India (https://timesofindia.indiatimes.com/technology/tech-news/before-google-deal-mark-zuckerberg-wanted-to-buy-deepmind-ceo-demis-hassabis-shares-what-went-wrong/articleshow/129877927.cms), citing Mallaby's reporting, Facebook offered a lower share price for DeepMind but compensated the founders and key colleagues with signing bonuses large enough that the founders would have walked away richer in absolute terms than under Google's deal. Mustafa Suleyman, DeepMind's then-co-founder, had raised concerns about AI governance and safety in an early meeting with Facebook's head of corporate development, Amin Zoufonoun — concerns Zoufonoun brushed aside, as Mallaby's account describes it. And yet the dinner is what decided it.
Google acquired DeepMind in 2014 for a reported $650 million (https://www.theguardian.com/technology/2014/jan/27/google-deepmind-400-million-britain). Facebook's total package would have left the founders better compensated. And yet Hassabis chose Google.
Hassabis had structured the sale deliberately. As a reviewer with first-hand knowledge of the book wrote (https://aletteraday.substack.com/p/br-the-infinity-machine-demis-hassabis), Hassabis sold DeepMind for less money and less power than he was offered, betting that Larry Page, Google's co-founder, would protect AI research autonomy in ways Facebook would not. The governance question, in Hassabis's telling, mattered more than the financial question — and he had designed a deal that forced every bidder to show their hand on it.
Zuckerberg showed his. At a South Park Commons event in August 2024 (https://www.businessinsider.com/mark-zuckerberg-gives-props-deepmind-ceo-demis-hassabis-google-aquisition-2024-8), he addressed the failed bid directly: "I did want to buy DeepMind," he said. "Demis was good, by the way. He totally did a very good job of playing me off of Google to get a good price, which I respect." The admiration is genuine. Zuckerberg understood he had been outmaneuvered. What he seems not to have grasped, at least in Hassabis's account, was that the maneuver was not primarily about price — it was about whether Zuckerberg saw what Hassabis saw.
As the Substack reviewer put it (https://aletteraday.substack.com/p/br-the-infinity-machine-demis-hassabis): Hassabis turned down Zuckerberg because he was excited by everything, not just AI.
The dinner test was informal. It was not in the term sheet. But it was the test that mattered.