The Failed Spin-Out: DeepMind's Battle for Independence
In early 2017, Demis Hassabis, co-founder of DeepMind, embarked on a grueling journey to transform the AI research lab into a semi-independent entity under Alphabet. After months of transatlantic flights between London and Mountain View, coupled with exhaustive legal negotiations and persistent jet lag, he believed a breakthrough was imminent. On January 9, 2017, Hassabis met with Sundar Pichai at Google's headquarters, with Mustafa Suleyman joining remotely from Asilomar. The discussion initially seemed promising, with Pichai appearing warm and open to finalizing details through David Drummond.
A Sudden Reversal and Strategic Objections
The next day, however, Drummond delivered a stark message: Hassabis had completely misread the situation. Pichai was firmly against the "Alphabetisation" of DeepMind. As detailed in Sebastian Mallaby's new book, The Infinity Machine, Pichai's objection was rooted in strategy. The Alphabet "bet" structure, designed for moonshot projects like autonomous cars and life-extension science, was unsuitable for artificial intelligence. AI was not a peripheral venture; it was destined to be central to Google Search, Google Cloud, and virtually every other aspect of the company's operations. Spinning out DeepMind, even partially, would mean relinquishing control over this critical engine, a risk Pichai was unwilling to take.
Misplaced Assumptions and Founder Absence
Hassabis and Suleyman had assumed that Larry Page's support would secure the deal, but this proved futile. Page arrived two hours late to a decisive meeting, and Sergey Brin was even later. With the founders effectively disengaged, Pichai emerged as the key decision-maker. Despite DeepMind's preparation of a $1 billion walk-away plan, it remained unused, highlighting the power dynamics at play.
The Negotiation Timeline and Failed Proposals
The negotiations had begun in late 2015 as a genuine effort to establish a novel governance structure: a 3-3-3 board comprising three representatives from DeepMind, three from Alphabet, and three independents. Hassabis and Suleyman had even secured commitments for $5 billion in outside capital, including a $1 billion pledge from Reid Hoffman, and proposed a legal framework termed a "global interest company." None of these proposals gained traction. By mid-2017, Suleyman announced the spin-out to DeepMind employees at an off-site in the Scottish Highlands, only for Google to return the proposal heavily marked with red lines, signaling its rejection.
Hassabis's Evolution: From Idealist to Realist
Reflecting on the saga years later, as CEO of the enlarged Google DeepMind—which now includes Google Brain and related teams—Hassabis arrived at a blunt conclusion. He stated that governance structures, such as independent boards or safety charters, often fail because they pursue their own agendas or draw lines in inappropriate places. "Safety isn't about governance structures," Hassabis emphasized. Instead, he advocated for earning trust within Google by being actively involved in decision-making processes, rather than negotiating from an external position. He described this shift as moving from an idealist to a realist, while striving to maintain core values intact.
