The most expensive meal in Silicon Valley: He thought Zuckerberg didn't understand AI, so he turned to Google, which directly pushed out OpenAI.

Stock trading analysis is all about the Golden Kylin Analyst Reports—authoritative, professional, timely, comprehensive—helping you uncover potential thematic opportunities!

(Source: NetEase Smart)

Produced by | NetEase Smart

Author | Xiao Xiao

Editor | Wang Fengzhi

Just one dinner meeting was enough for Hassabis to completely dismiss Zuckerberg’s higher bid: he doesn’t understand AI.

In 2013, DeepMind, then mired in funding struggles, received takeover offers from both Google and Facebook. To win over this talented team, Mark Zuckerberg didn’t hold back and offered a higher premium. But after dinner with Zuckerberg, Demis Hassabis keenly sensed that his enthusiasm for 3D printing and virtual reality was no different from his interest in AI. This unfocused, trend-chasing attitude led Hassabis to conclude that Zuckerberg didn’t truly grasp AI’s unique potential.

He decisively refused the huge check and instead threw a near-maximum pressure acquisition proposal at Google.

At the negotiation table, Hassabis not only demanded that Google’s valuation of each scientist—based on talent acquisition—be doubled from $10 million, but also inserted three ironclad rules: the company must be based in London, technology must be de-militarized, and an independent ethics committee must be established. Although these terms significantly weakened Google’s control over the acquired assets, because Google believed Hassabis represented the future of AI strategy, they ultimately accepted all conditions, and in January 2014, spent $650 million to acquire DeepMind.

This dramatic compromise not only kept Google’s spark alive in the AI race but also inadvertently triggered a huge “butterfly effect” in tech history. At the moment the deal closed, Elon Musk, feeling the threat of monopoly, decided to team up with Sam Altman to create OpenAI to counter Google. A decade later, the AI power struggle was born.

01 A walk at a birthday party changed Hassabis’s mind

The story begins at a birthday party.

In June 2013, at a castle in Tarrytown, New York, actress Talulah Riley was hosting her husband Elon Musk’s birthday party, and Hassabis was among the invited guests.

Musk was an early investor in DeepMind. The company had been founded just three years earlier, focused on AGI. Hassabis wanted to create an “ Manhattan Project” for intelligence—gather the smartest minds to develop an agent capable of playing multiple Atari games. At the time, this sounded like a fantasy.

Google co-founder Larry Page also attended the party. He took Hassabis for a walk in the castle courtyard and, in a tense whisper due to vocal cord issues, said something that left a deep impression on Hassabis:

“Why not use what I’ve already created?”

Page’s logic was straightforward. Hassabis was a scientist aiming for AGI. Why cling to DeepMind as an independent company? Even if he dedicated his best years, he might not recreate Google. But if the true mission was building AGI, all of Google’s resources could be at his disposal.

This conversation struck Hassabis’s core. As head of this three-year-old company, he was tired of fundraising. “I want to go to Google. I want to get a lot of computers, and then I want to solve intelligence,” he said. That walk in the castle marked the start of one of the most critical acquisition negotiations in tech history.

02 Poker face and bluffing at Google’s negotiation table

In fall 2013, Hassabis and co-founder Mustafa Suleyman flew to Google’s headquarters. To keep it secret, they were placed in a conference room across from the main building. Google’s M&A team assembled internal AI experts to evaluate DeepMind’s strength. Hassabis showcased their recent progress on Atari agents.

But one detail surprised Google: Hassabis and Suleyman showed no interest in discussing price. Suleyman later explained they were afraid that mentioning money upfront would make Google think they just wanted a quick cash-out. “It would seem like we’re just here to take the money and run.”

Instead, they asked only two questions: what research budget could Google provide? And how would safety be ensured?

Suleyman was especially serious about safety. He proposed that if Google acquired DeepMind, an independent oversight committee should be established to protect the technology. The committee should include scientists, philosophers, and other influential figures, with the final say on AI’s societal use. “In successful scenarios, we can’t let Google’s founders use AGI solely for their own purposes,” he said.

To pressure Google on this issue, Suleyman drew on his poker experience.

“We told them we’re Europe’s best-funded startup, still pre-revenue. We have support from billionaires like Peter Thiel, Zhou Kaixuan, Musk,” Suleyman recalled. “Of course, those people weren’t really backing us, which made us feel uneasy as negotiators. But in poker, you’re not just playing cards—you’re reading other players and betting based on their psychology.”

Hassabis wasn’t used to this approach. He later said he saw himself as a chess player, not a poker player. Chess has no hidden cards; it’s all open, with no bluffing. Turns out, Suleyman’s move might have been a waste, because Google’s top team in Mountain View was also concerned about safety.

Then-Google CFO Patrick Pichette later said they saw AI as atomic energy—capable of making bombs but also solving climate change. From the start, they discussed big issues: what if it gets out of control? How to control it?

It seemed both sides could talk about safety. But Hassabis and Suleyman didn’t take anything for granted. To push Google to decide, they also contacted another bidder: Facebook’s Mark Zuckerberg.

03 Zuckerberg invites dinner, Hassabis tests his resolve with a question

While negotiating with Google, DeepMind didn’t put all eggs in one basket. Suleyman even flew to meet Facebook executives.

At that time, CEO Mark Zuckerberg was feeling anxious. Seeing other tech giants building AI teams, he scrambled to catch up by recruiting talent himself. Despite managing 6,000 employees and a billion users, he still made time to dine with individual AI researchers.

To show sincerity, Facebook’s head of corporate development, Amin Zoufonoun, hosted Suleyman at his home, pouring him a shot of whiskey and even offering a more generous bid than Google: though they would lowball on DeepMind’s share price, they promised hefty signing bonuses for founders and key staff—more than what Google offered.

Suleyman reported back to Hassabis. The money was tempting, but not their main goal. More importantly, Zoufonoun showed no interest in AI governance issues. Facebook’s attitude toward AI safety was clear to outsiders, and Hassabis and Suleyman sensed it early.

Zoufonoun returned to Zuckerberg with the news: DeepMind had a strong team of AI scientists, and if Facebook didn’t buy it, Google would.

Hassabis then went to the West Coast for a lunch with Larry Page. Zuckerberg heard about his visit and invited him to dinner.

At Zuckerberg’s Palo Alto home, Hassabis conducted a subtle test. They discussed AI’s potential, and Zuckerberg appeared very excited. But when Hassabis shifted the conversation to virtual reality, 3D printing, and other hot tech topics, Zuckerberg showed the same enthusiasm.

“Tell me everything I want to know,” Hassabis later said. “Facebook offered more money, but I wanted someone who truly understood why AI is more important than all these other things.”

In the end, Hassabis sent a reply to Page: “Let’s keep moving forward!”

Rejected, Zuckerberg’s competitive instinct was triggered. He doubled down on recruiting individual researchers and even invited Yann LeCun, then at NYU, for another dinner.

Zuckerberg asked how to get LeCun to join Facebook. LeCun said he wouldn’t leave New York or quit his teaching job. He thought those conditions would make Zuckerberg give up.

But the next day, Zuckerberg accepted all terms. LeCun’s reply was simply: “Where do I sign?”

04 Christmas-time talent hunt nearly tore DeepMind apart

In early December 2013, DeepMind staff attended the world’s largest machine learning conference. Negotiations with Google had entered the contract review stage, and they were reviewing documents during the conference. Meanwhile, Zuckerberg and LeCun rented a banquet hall in a hotel and announced the creation of a new AI lab in Manhattan.

At the conference, Hassabis met LeCun. “You’re not planning to poach all my people, right?” Hassabis asked.

“I just signed the papers, and that’s basically what I’m here to do,” LeCun recalled.

Soon after, LeCun offered high salaries to poach key DeepMind researchers, including his former student Koray Kavukcuoglu.

Suleyman later said he thought DeepMind might really be doomed. Hassabis quickly stabilized the situation and privately told core staff that Google was about to acquire the company. The stock options they had previously dismissed might soon be worth a fortune. He urged Google to close the deal quickly.

By late December, Google’s team flew a Gulfstream jet to London’s DeepMind office. Jeff Dean, Google’s legendary engineering leader, requested to review Atari system code to verify DeepMind’s technology.

“That was a crossing of the Rubicon moment,” Hassabis recalled. “The world’s biggest, best company was watching all your research. If you didn’t close the deal afterward, you’d be destroyed.”

Dean gave a thumbs-up on the code. Only one question remained: the price.

At that time, DeepMind had no revenue; its only asset was its people. Google’s M&A team had a mature valuation method for talent acquisitions. Lead negotiator Don Harrison estimated DeepMind had about 30-40 core scientists—not engineers, but scientists. Valued at $10 million each, the total price was roughly set. Harrison, a Canadian lawyer who had handled dozens of deals, rarely faced resistance.

But this time, Hassabis and Suleyman pushed back hard, nearly doubling the offer. Harrison later said everyone in the conference room felt uncomfortable, even Dean thought the price might be too high.

Besides price, Hassabis also proposed three conditions: he wanted to stay in London, DeepMind’s tech must not be used for military purposes, and an ethics and safety review board (comprising founders and external experts) should be established to limit Google’s control.

“This was a big issue for me,” Harrison recalled. “I had to pitch this to the board—it’s not just about the price, but the structure. We’re paying a lot for this asset, but control is being weakened.”

In the end, Google swallowed these concerns because of Hassabis. “If we weren’t absolutely convinced that Demis represented our AI future, we wouldn’t agree to this structure,” Harrison said afterward.

By late January 2014, Google acquired DeepMind for $650 million. By today’s standards, a bargain. But for Hassabis, the real reward was that over the next decade, Google invested hundreds of billions into DeepMind’s research. His childhood dream of AGI finally entered the fast lane.

05 An acquisition that created a formidable rival

After the deal, DeepMind became Google’s AI engine. But the most interesting part was the side effect.

Elon Musk, an early investor in DeepMind, watched Google acquire it and felt uneasy. He worried about Google monopolizing AI, so he teamed up with Sam Altman to create OpenAI.

In November 2022, OpenAI launched ChatGPT, catching Google off guard. Throughout 2023 and 2024, doubts grew whether Google could keep up. Hassabis admitted the core technology—the Transformer architecture supporting large language models—was originally developed by Google researchers. The real issue was their slow pace in commercialization and scaling.

“That’s where OpenAI and others excel,” he said.

In 2023, Google made a key move: merging Google Brain and DeepMind under Hassabis’s leadership. They had to return to startup roots—be more agile, faster, and launch products quickly.

By March 2025, Gemini 2.5 was released. In November, Gemini 3.0 launched, impressing the industry. Alphabet’s stock hit its best performance since 2009. Media said Google finally regained its AI mojo.

By late 2025, Hassabis told Fortune that the competition was fierce. Many industry veterans said it was the most intense environment they’d seen in decades—perhaps ever.

Hassabis now discusses strategy almost daily with Google CEO Sundar Pichai, talking about the future of technology and what Google needs. He compares DeepMind to Google’s engine room—like a nuclear power plant powering Search, YouTube, Chrome.

He says they’ve been building this backbone for years—models and infrastructure—so these innovations can be delivered at lightning speed.

Hassabis divides his day into two parts: meetings in the office during the day, then returning home to family. At 10 p.m., he starts his second shift—working until 4 a.m.—believing that’s his prime time for creative and research thinking.

This intense interdisciplinary collaboration kept DeepMind from being bogged down by corporate bureaucracy. Over ten years, Google invested billions, resulting in groundbreaking scientific achievements. The most famous is AlphaFold, solving the 50-year-old protein structure problem, earning Hassabis a Nobel Prize, and spawning startups like Isomorphic, aiming to use AI to cure all human diseases.

Currently, Isomorphic has partnered with pharma giants like Eli Lilly and Novartis, launching about 17 drug projects, aiming to bring the first anti-cancer drugs into clinical trials by 2026.

06 AI bubble, glasses, and general assistants

What does Hassabis think about the current AI bubble?

He says some parts “might be in a bubble,” others not. He compares it to the late 1990s internet bubble. “In the end, the internet became critical, and some epoch-making companies emerged. It was almost inevitable. Once everyone realizes how transformative a technology is, they get overly excited. Then comes a correction, and the real things survive.”

He criticizes early-stage funding rounds with valuations in the hundreds of billions before products are even developed as “unsustainable in the long run.”

Looking ahead, Hassabis is very specific. By the end of 2026, AI might be able to build autonomous agents capable of handling entire tasks. He’s excited about AI-powered glasses, believing the technology has just reached the point where it can truly become viable—AI could be a killer app for glasses.

He also envisions a “general assistant”—an AI helper on computers, phones, browsers, and eventually glasses. It would understand different contexts—car, office—and connect conversations across scenes, helping with chores, making recommendations, enriching life.

When asked about his boldest future prediction, Hassabis says if everything goes well, in 10 to 15 years, we’ll enter a new golden age of discovery. “Human health will undergo revolutionary change; medicine will no longer look like today. We’ll use AI to solve big scientific problems—new materials, nuclear fusion, solar energy, battery optimization—and somehow address the energy crisis. Then, we’ll enter a world of abundant resources, exploring the galaxy.”

“I think that’s our destiny.”

Looking back at that $650 million deal from over a decade ago, it seems ridiculously cheap today. It not only kept Google’s spark alive in the AI race but also freed a group of talented minds focused on intelligence from mundane fundraising, propelling them into the super-fast lane toward AGI.