Zuckerberg’s Two Mistakes That Let Google Snag DeepMind

Background and the Missed Opportunity

According to Sebastian Mallaby’s book “Infinite Machine”, in early 2013 Demis Hassabis, co‑founder of DeepMind, met Google CEO Larry Page at Elon Musk’s birthday party. Hassabis, who had been running DeepMind for three years with the goal of creating an “AI Manhattan Project” capable of playing multiple Atari games, was approached by Page, who suggested that Google’s resources could accelerate the quest for artificial general intelligence.

Hassabis found the argument persuasive and, weary of fundraising, began talks not only with Google but also with Facebook’s Mark Zuckerberg, hoping to secure the best deal.

First Mistake: Indifference to AI Safety

Facebook’s corporate development lead Amin Zubair proposed a deal that would make DeepMind’s founders richer than a Google acquisition by offering a low purchase price for equity followed by large signing bonuses. Google, however, valued DeepMind primarily on its engineering talent, estimating each of the 30‑40 engineers at roughly $10 million, and offered about half of Hassabis’s valuation, which was nearly double Google’s price.

While Facebook’s financial offer was stronger, Hassabis and co‑founder Mustafa Suleyman cared more about AI safety. Suleyman demanded that any Google acquisition include an independent oversight committee of scientists, philosophers and other experts with final say over AI applications. Google’s leadership took the proposal seriously, whereas Facebook’s Amin Zubair was indifferent to the safety concerns.

Second Mistake: Failing Hassabis’s Test

When Hassabis learned that Zuckerberg was planning a dinner with Larry Page, Zuckerberg immediately invited Hassabis to a separate dinner. During the meal Hassabis subtly tested Zuckerberg’s understanding of AI by shifting the conversation from AI to virtual reality, augmented reality and 3D printing. Zuckerberg responded with equal enthusiasm for those technologies, revealing that his excitement was spread across many trends rather than focused on AI.

Hassabis concluded that Zuckerberg valued high salaries over a deep grasp of AI’s unique challenges, stating, “Facebook offers higher pay, but I want someone who truly understands why AI matters more than everything else.” Consequently, Hassabis and Suleyman leaned toward Google.

Outcome and Later Consequences

In January 2014 Google completed the acquisition of DeepMind for $650 million, a price later viewed as a bargain. The article then traces Meta’s subsequent “Metaverse” gamble: in 2021 Zuckerberg rebranded Facebook as Meta and poured resources into virtual‑reality initiatives. Since 2020 Reality Labs has lost over $84 billion, with a single‑quarter loss of $6.02 billion projected for Q4 2025. Horizon Worlds peaked at about 300 k monthly active users in early 2022 before falling below 200 k later that year.

Meta announced the shutdown of its Metaverse efforts on 15 June, while its Reality Labs budget still allocates roughly 30 % to VR and the rest to wearables and AI glasses. In 2025 Meta’s Ray‑Ban Meta smart glasses sold 7 million units, three times the combined sales of 2023‑2024, and the company reported $201 billion in revenue and a market valuation of about $1.6 trillion.

The piece ends with a rhetorical question: had Zuckerberg secured DeepMind, how different would Meta’s present be? The answer remains speculative.