In January 2014, Google agreed to buy DeepMind, a London-based artificial intelligence lab founded in 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman. The publicly reported price was roughly £400 million, or about $500 million at the time. The actual price, according to three people with direct knowledge of the deal who spoke to Reuters, was closer to $650 million — a figure Google never disclosed. The acquisition was not announced as a pivotal moment. It was announced as an acquisition.
Twelve years later, it looks like one of the most consequential purchases in the history of the technology industry.
The inside story, reported by the Wall Street Journal this week, describes a deal driven largely by Larry Page, then Google's chief executive, who Hassabis told might be the most important acquisition Google had ever made. That assessment has aged in ways that are both obvious and strange. Google did not just acquire a research lab. It acquired a decade of AGI ambition, a research culture that would produce AlphaFold and AlphaZero and Gemini, and, arguably, one of the primary catalysts for the competitive AI landscape that followed.
The catalyst is Elon Musk. Musk invested roughly $8 million in DeepMind in 2013 through Founders Fund, making him one of the earliest outside backers of a lab that was still defining itself. According to a 2016 email disclosed in litigation and reported by Reuters, Musk told OpenAI's leaders that DeepMind "wanted to build one mind to rule the world" and that it was "causing me extreme mental stress. If they win, it will be really bad news." That fear — that a Google-controlled DeepMind would give the company an insurmountable AI monopoly — is what pushed Musk to co-found OpenAI with Sam Altman, according to Fortune's reporting. DeepMind did not build a mind to rule the world. But its existence helped build OpenAI, which then helped build the competitive field that now includes Anthropic, xAI, and a dozen other labs that would not exist in the same form without the threat perception Musk articulated in that email.
The financial scale of Google's bet is now becoming legible through regulatory filings. DeepMind has deployed more than $9.6 billion of Google capital for operating expenses as it pursues AGI, according to Reuters's review of British regulatory filings. Over the five years ending in 2024, DeepMind posted cumulative revenues of more than $7.8 billion — all from internal payments from other Alphabet companies using DeepMind technology. That is a significant number, but it is an accounting fiction in the sense that it represents transfer pricing within a single corporate family, not external market validation.
The research output, by contrast, is real and measurable in ways the revenue figure is not. DeepMind's AlphaFold system solved the protein folding problem, a feat that structural biologists had considered open for decades. Its AlphaZero system taught itself chess, shogi, and Go from scratch and surpassed human world champions within 24 hours of training. More recently, Google's Gemini 3 model, marketed under the Deep Think designation, autonomously solved four open mathematical problems from the Bloom-Erdos Conjectures database, including the Erdos-1051 conjecture, which had remained open for decades. These are not benchmarking vanity metrics. They are problems that human mathematicians had worked on for years without resolution.
By the end of 2025, Google's Gemini app had surpassed 750 million monthly active users, up from approximately 350 million in March 2025. The user growth is real. The AGI is still a target.
There is a harder question underneath all of this that neither the Journal's account nor the filings fully answer: what did Google actually buy? It acquired a research institution that has published landmark results and trained some of the most capable researchers in the field. It acquired a pipeline of talent. It acquired DeepMind Health, which it later dissolved. It acquired the intellectual infrastructure for Google's most important AI products. But it also acquired an entity that has consumed more than $9.6 billion in operating capital while producing almost entirely internal value — a research engine that powers Google products without, by most public measures, producing a standalone commercial entity comparable to the investment it has required.
Hassabis, for his part, targeted a Nobel Prize as one of DeepMind's formal business objectives, according to three people familiar with the effort. That is either the most honest research ambition statement ever put on record or a useful fiction for recruiting purposes, and possibly both.
The least told part of the DeepMind story may be the 2019 offer that was declined. According to Reuters, Hassabis turned down a joint venture proposal from OpenAI around 2019 — years before ChatGPT made frontier AI a household concept. The offer came at a moment when OpenAI was still a relatively small research organization and the competitive dynamics that would define the next half-decade had not yet materialized. Declining it set a direction that kept DeepMind firmly within Alphabet's orbit rather than becoming a bridge between the two most prominent AI research organizations in the world.
What the Journal's account makes clear is that Google knew what it was buying, at least in the sense that Page was told it might be the most important acquisition in the company's history. Twelve years and $9.6 billion in operating capital later, the verdict is still out on whether that was prescience or self-fulfilling prophecy. The research results are extraordinary. The commercial structure is still internal. The competitive field that DeepMind helped create — through the fear it inspired in Musk and the response that fear produced — is now the defining arena of the technology industry.
The bet is still being settled.
