Home
+
Published : 13 Nov 2025, 11:39 PM
In the 11 years since Google acquired artificial-intelligence startup DeepMind, founder Demis Hassabis has become the search giant’s top AI executive, won a Nobel Prize, and earned millions of dollars.
For Alphabet, Google’s parent company, the benefits have been slower to materialise.
In the scramble for AI dominance, Google risks losing the supremacy it has long enjoyed in the search industry. It was jolted when OpenAI, the creator of ChatGPT, showed in 2022 that AI could deliver information more seamlessly than a traditional search engine.
OpenAI did so, no less, using technology first developed by Google researchers. In its rush to catch up, Google suffered a series of embarrassments — from false information in an ad for its AI, to nonsensical AI-generated renderings of Nazi soldiers as Black, the result of how its systems were tuned to depict the wide diversity of human beings.
With Hassabis now in greater control of Alphabet’s AI efforts, Google has rolled out an increasing array of AI offerings. That includes an improved chatbot and AI model, now known as Gemini, that enhances its search engine and tops some industry performance rankings. A new AI photo editor, Nano Banana, attracted 13 million first-time users to the Gemini app in four days this September, allowing users to generate or edit images using photos, drawings or text prompts. The advances helped Alphabet’s share price reach an all-time high this month.
Still, some investors question why Alphabet hasn’t emerged as the undisputed AI leader given its inherent strengths. “It’s the classic team with all the talent that doesn’t win the national championship,” said Gene Munster, managing partner at Deepwater Asset Management. The Minneapolis investment firm this year sold around $14 million of Alphabet stock, concerned about its prospects, despite recent strides, in the fast-changing AI marketplace, Munster said.
In a statement, a Google spokesperson defended the company’s “leadership in AI and relentless focus on world-class research and scientific innovation.” “Anyone familiar with the AI industry,” the statement added, “recognises the importance of this work to our technical and commercial success.”
Making money for Google has often taken a backseat to higher-minded pursuits, people who’ve worked with Hassabis say. The London-born scientist methodically targeted a Nobel as one of DeepMind’s business objectives, according to three people familiar with the effort. Since joining Google, insiders say, he has worked against some efforts that could possibly have meant new revenue for Alphabet or helped it gain better footing in the AI race.
Well before ChatGPT’s rise to prominence, for instance, Hassabis declined an offer from OpenAI around 2019 to create a joint venture related to advanced artificial intelligence, four people familiar with the episode said. Later, four other people said, staffers were surprised when Hassabis didn’t prioritise a DeepMind project to apply AI to financial trading. The project included exploratory talks with BlackRock, the world’s largest asset manager, but eventually disbanded.
The OpenAI talks and the years-long trading effort haven’t been previously reported. Google and OpenAI didn’t reply to questions about their discussions. Google declined to comment on the project involving BlackRock, saying its teams pursue “a myriad of projects all the time and constantly evaluate and prioritise for the highest impact.”
Alphabet declined to make Hassabis available for an interview.
In public, Hassabis often avoids talk of money. Most industry leaders speak of the market applications of AI even as they also discuss its implications for humankind. Hassabis frames his quest in philosophical and cosmic terms, sometimes veering outright into science fiction.
“What I’ve always dreamt about,” Hassabis said in a 2022 podcast interview , is “the ultimate flourishing of humanity to travel the stars and find those aliens, if they are there.” With further advances, he added, AI could lead to “a world of radical abundance and curing diseases and solving many of the big challenges we have.”
Alphabet’s leadership has praised Hassabis. After the Nobel win, Chief Executive Sundar Pichai on X, the social media site, wrote that the laurel was “just the beginning of whats to come.”
And some colleagues brush off the idea that Hassabis, as chief executive of Google DeepMind, is too concerned with the theoretical. By making him the company’s top AI executive, Alphabet is betting more on his savvy as a scientist and entrepreneur than on any conventional talent as a salesman. His focus on the big picture, they say, complements Google’s foundational goal of organising the world’s information and helps it continue aiming high.
“He’s always going to have the most ambitious, crazy view on how to solve something,” James Manyika, a senior vice president at Google, told Reuters in an interview. “A typical Demis response to things is, ‘Why not make it available to everybody? Why not solve everything?’”
To understand Hassabis’ role in the AI race and impact on Google’s business, Reuters interviewed more than 30 current and former DeepMind employees and investors, along with dozens more people from Alphabet and the technology industry who have worked or interacted with him. Combined, their accounts portray a visionary scientist whose relentless pursuit of advanced AI and desire for control at times have worked against Google’s profit interests.
Reuters’ look at Hassabis’ stewardship at Alphabet comes as the company confronts major challenges. Alphabet faces scrutiny from antitrust authorities in the United States and Europe. In April, a US federal judge found Google illegally dominates some of the market for online advertising technology.
Alphabet also is contending with upstarts. OpenAI last month unveiled ChatGPT Atlas, a new web browser , widely seen as a challenge to Google Chrome.
Some familiar with his Google career say Hassabis, 49 years old, has often kept his distance from his corporate parent. After Google acquired London-based DeepMind, 10 of these people told Reuters, he and his co-founders sought to isolate it from corporate interference. For several years, they said, he worked to spin the unit off from Google or at least engineer a relationship that would allow it to operate independently.
In Google’s statement to Reuters, the spokesperson said DeepMind has sought to “look for the best structure to achieve its goals – naturally that evolves over time.”
Hassabis’ push for autonomy, say several people familiar with the effort, reflected a belief that only a DeepMind detached from Google’s profit motive could ensure AI is developed responsibly. Safety is much discussed by top AI executives – including the possibility of doomsday if an AI model were to develop superhuman intelligence and then go rogue.
Like other chatbots, however, Google’s Gemini is still grappling with the safety and security risks associated with AI. As Reuters has detailed, some AI companies’ products have deceived users or created new ways for bad actors to commit fraud at scale. Google said in the September report that it deployed additional safeguards after Reuters demonstrated that Gemini could be used to produce phishing emails targeting senior citizens.
Hassabis’ desire to go it alone, some who know him say, stems from his belief that he is the scientist best poised to develop AI safely. Other top AI researchers and entrepreneurs, meanwhile, think they are the most responsible custodians of the revolutionary technology.
Rivals at times have cast Hassabis in dark scenarios in which AI upends society and gives machines control over humanity. Elon Musk, the billionaire entrepreneur and an investor in DeepMind before its acquisition by Google, expressed fears in a 2016 email to leaders at OpenAI that DeepMind wanted to build “one mind to rule the world.”
“Deepmind is causing me extreme mental stress,” wrote Musk, who by then had become an early backer of OpenAI. “If they win, it will be really bad news.”
Musk and Google didn’t respond to requests for comment on Musk’s views. Musk has since launched xAI, his own AI company . The email was among exchanges made public during ongoing litigation between Musk and OpenAI related to OpenAI’s corporate structure.
Regulatory filings in Britain show DeepMind over the five years ending in 2024 posted cumulative revenues of more than $7.8 billion. All of that turnover, however, comes from internal payments for DeepMind technology used by other Google platforms. There is no indication that the company received revenue from customers outside the Alphabet empire.
The filings also show DeepMind has deployed more than $9.6 billion of Google capital for operating expenses as Hassabis pursues his goal of artificial general intelligence, or AGI – an advanced state of AI in which computers in essence become as smart as humans. Considered years away by many computer scientists, AGI is the focus of ventures from China to Silicon Valley.
Despite his long-term approach, Hassabis’ pursuit of futuristic objectives has led to some fast breakthroughs for DeepMind.
One is AlphaFold, an AI programme that can predict protein structures. A version of the programme led Hassabis and a DeepMind colleague to be among a trio who won the Nobel Prize in Chemistry last year. Still, AlphaFold isn’t a significant source of Alphabet revenue, according to Alphabet filings.
“THE MOST IMPORTANT THING OF ALL TIME”
Born to immigrant parents, Hassabis displayed a hunger for knowledge and a competitive streak even as a boy. He became one of England’s best young chess players.
At the age of 11, after losing an eight-hour match in Liechtenstein, he had an insight that would later drive his search for AI. Watching the tournament’s players, he thought: “What if all this incredible collective mental effort being expended could instead somehow be channelled into something beyond games?” Hassabis recounted the moment in “Game Changer,” a 2019 book about DeepMind’s research and chess.
He grew enthralled by the big questions posed by physicists. As a teen, he said in a 2018 interview, he wondered if a “final theory of everything” could be found to describe the workings of the universe. His love of games led him to his first job in tech – working, at age 16, for Bullfrog Productions, an English video game company.
A game Hassabis helped design, “Theme Park,” became a hit in 1994, just before he entered Cambridge University. By 1999, The Guardian newspaper reported, Theme Park generated sales of $80 million for Bullfrog, which was acquired by a bigger company. Hassabis told an alumni magazine in 2014 that a local gaming executive lent him a Porsche, which he drove around Cambridge and drag-raced nearby.
After earning a computer science degree, Hassabis completed a doctorate in neuroscience. In London, he later met Shane Legg, another post-doctoral researcher also focused on AI. Together, they co-founded the startup that would become DeepMind. Alphabet declined to make Legg available for an interview.
Their initial efforts to raise capital were complicated by a challenge confronting Hassabis still: Many investors lack the patience to embrace distant concepts. “What’s your product?” they would ask, he recalled in a company-sponsored documentary about DeepMind last year. “It’s like so prosaic a question.” His answer: “The most important thing of all time.”
Through Legg, the pair was invited to a conference in 2010 where they met Peter Thiel, the Silicon Valley venture capitalist. To compete with others vying for Thiel’s time, Hassabis ditched his DeepMind pitch during a brief introduction and instead appealed to Thiel’s interest in chess to secure a second meeting. “Dynamic tension,” he told Thiel, according to Hassabis’ account in the 2019 book, hinged on “the exquisite balance of the bishop and knight.”
Thiel agreed to talk further. In early 2011 his firm invested more than $2 million in DeepMind. Thiel, through a spokesperson, declined to comment.
Hassabis began hiring researchers, touting the venture more like a science project than a business. The co-founders asked potential hires if they wanted to “build brains,” according to two early recruits.
Hassabis himself continued attending university lectures, often bringing staff with him. He later told employees, and said publicly, he was a “scientist first” and “entrepreneur second,” according to two people who heard variations on his remarks.
For much of its ramp-up, DeepMind had no product or revenue. It worked with a 10-year plan to develop an AI “agent,” or programme, with the thinking power of a rodent, five people familiar with the matter said.
Hassabis wooed further financing to stay afloat. That included more investments from Thiel’s Founders Fund and a 2013 investment of about $8 million by Musk, who has said he provided the backing as a way to monitor AI progress.
That year, DeepMind had its first breakthrough, presenting research about “Deep Q-Networks,” a type of early AI that could outplay humans in some Atari games. DeepMind didn’t commercialise the technology.
But its work garnered attention. In January 2014, Google made its move.
Despite an initial fear of losing independence, Hassabis and his partners agreed to sell. The money would help DeepMind survive while giving them time and resources to focus on science, they reasoned, said several people familiar with the decision.
Google has never disclosed the value of the deal, but three of the people said it totalled around $650 million.
As a founder with a 21 percent stake in DeepMind, according to a 2014 regulatory filing, Hassabis earned a handsome payout. In the 2024 documentary, he praised the transaction: “We were able to not only join Google but run independently, in London,” he said, “not deal with products, do pure research.”
“MAYBE NOTHING”
At Google, Hassabis created a DeepMind division that would apply the startup’s research toward some of the parent company’s ventures. The effort used AI, for instance, to make data centre operations more efficient. It later enhanced software to prolong mobile battery life for Android phones.
But it spawned no new businesses. DeepMind remained focused on the long-term goal of AGI, and soon, Hassabis was rethinking his unit’s role.
Google restructured in 2015, creating the broader holding company now known as Alphabet. Hassabis wanted to make DeepMind a standalone unit, according to three people familiar with the effort. Google’s autonomous car unit, now known as Waymo, would soon do just that.
But Alphabet executives, including Ruth Porat, now the company’s president and chief investment officer, opposed it, these people said. Artificial intelligence was too strategic, top executives argued. And unlike Waymo, poised to have a broad range of consumer and corporate clients, DeepMind had only one customer – Google itself.
A spinoff was out of the question, they told Hassabis. Google declined to make Porat available for an interview.
In late 2015, a group including Musk and rival AI researchers founded OpenAI. Several had worked at DeepMind and replicated some of their ex-employer’s plans there, former DeepMind employees told Reuters. Because OpenAI was originally founded as a non-profit, some of these people said, various DeepMind defectors believed it was a safer place to develop AI than within the confines of for-profit Alphabet.
An OpenAI spokesperson in a phone interview disputed the assertion that it had replicated DeepMind plans.
Hassabis shrugged off early efforts by competitors. One person familiar with his views said Hassabis told staff DeepMind’s technology was superior.
As OpenAI launched, DeepMind began discussions with asset-management giant BlackRock about a partnership to apply AI to financial trading, said three people familiar with the matter. In December 2015, one of these people said, a BlackRock team met with Hassabis and other DeepMind executives in London to discuss a venture.
Hassabis made clear his ambitions went higher than the pursuit of profits, this person said. According to notes from the gathering, read by this person to Reuters, Hassabis described DeepMind as “an Apollo programme” whose mission was, first, to “solve” intelligence, then to solve “everything else.” Compared with that “lofty vision,” the person said, talk of markets afterwards sounded “grubby.”
Still, the two teams agreed to keep talking. They envisioned a project called “DeepTick” that would apply AI to financial trading, according to notes from a follow-up encounter.
Hassabis also pursued other projects, including one that soon sparked major buzz. DeepMind taught AlphaGo, an AI programme, to excel at Go, an ancient Chinese game whose complexity surpasses even that of Hassabis’ beloved chess. DeepMind trained AlphaGo by showing it examples of humans playing Go, then having it play itself many millions of times.
At a highly publicised five-day event in March 2016 in Seoul, AlphaGo went up against Go superstar Lee Sedol of South Korea. AlphaGo made one move so perplexing onlookers thought it a mistake. But the manoeuvre proved ingenious. AlphaGo won four of five games.
The triumph led to praise from Alphabet, but also to tensions.
To drum up excitement, Hassabis had promised a $1 million purse to the victor. But he hadn’t secured Google’s approval, said one person familiar with the internal finances. Google wired DeepMind the money, which was eventually given to non-profits. The sum was a rounding error for the Silicon Valley giant. But the unauthorised wager troubled some members of Alphabet’s finance staff because they had to scramble to send the funds, the person said.
Some investors and researchers questioned whether DeepMind’s game-centric work would ever pay off. “They produced exquisite toys,” said one person who has done business with Hassabis. “What does it mean to be a master at Go? Maybe a lot, maybe nothing.”
Hassabis continued pushing for autonomy. At a retreat in Scotland in 2017, three people familiar with the event told Reuters, he and other DeepMind leaders told staff they planned to break away from Google. To aid that quest, people aware of his efforts said, Hassabis sought more ways for DeepMind to make its mark.
Hassabis wanted to see how AI might enhance other sectors, like medicine. He told staff to target problems so big that solving them could clinch a Nobel, three people with knowledge of the effort told Reuters.
The Google spokesperson said in a statement that “a Nobel has never been an end in itself, but instead, the result of pursuing world class research. We are extremely proud of the Nobel Prize and believe it is a reflection of the groundbreaking work we continue to drive on the frontier of AI.”
For half a century, scientists had struggled to understand how amino acids twist and fold into three-dimensional proteins. Within five years, DeepMind developed a programme that can predict the process. It released the technology to millions of scientists for free.
The research prompted Alphabet to launch Isomorphic Labs, a DeepMind sister company, also led by Hassabis. It’s now working to develop drugs quickly using AI. Given the slow pace of pharmaceutical research and development, the venture is too young to have had commercial success. But its potential, Hassabis’ allies say, exemplifies his ability to glean business opportunities from even slow-moving scientific research.
“Demis does care about the long term,” Lila Ibrahim, DeepMind’s chief operating officer, told Reuters. “But he’s delivering on the short term.”
By this point, DeepMind’s exploration of financial trading was foundering. After its early discussions with BlackRock, three people familiar with the effort told Reuters, DeepMind decided to work on the research alone.
Joe Fenton, a product manager with a background in physics and corporate finance, managed the project, involving more than 20 researchers and engineers. According to Fenton’s LinkedIn profile, the project “ingested large quantities of noisy real-world data” to “build a product in the investment management domain.” Fenton, no longer at DeepMind, didn’t respond to requests for comment.
Former employees said the complexity of financial markets proved challenging even for AI. In simulations, one of the people said, its AI at times outperformed the market. But Hassabis’ focus was elsewhere, three of the people told Reuters.
“It fizzled out,” one said, “quietly disappeared.”
A “UNIVERSAL ASSISTANT”
Around 2019, OpenAI leaders made Hassabis an offer, say four people familiar with the episode. Over dinner in Silicon Valley, two said, an OpenAI executive proposed that if either OpenAI or DeepMind got close to AGI, it should inform the other, and the two teams should then collaborate.
OpenAI’s motivation, the people said, was to ensure that neither lab cut corners on safety. Those who know Hassabis say that safety is always a paramount concern. Hassabis nonetheless preferred to let DeepMind forge ahead on its own.
By 2020, OpenAI was getting attention, unveiling an early iteration of the technology behind ChatGPT. The programme was employing “Transformer” architecture – a system developed by employees from Google Brain, a US-based unit that researched AI separately from Hassabis’ team. Google had written about the Transformer in a paper three years earlier. But instead of commercialising the Transformer, Google made it freely available to other researchers.
That decision was not rare in an industry that uses publication as a way to advance scholarship and attract talent. In this case, however, Google’s early reluctance to release a chatbot based on the Transformer would later be considered a missed opportunity, some Alphabet leaders have said, because it enabled OpenAI to roll out its chatbot first.
DeepMind’s push for autonomy was soon overtaken by events – including a promotion for Hassabis. After ChatGPT had its big public breakthrough in 2022, CEO Pichai merged Google Brain with DeepMind, making Hassabis point person for all things AI.
Colleagues say Hassabis made clear his commitment to product launches as recently as this year, when Google rolled out its Nano Banana image editor. After its debut in August, the programme drove millions of downloads of the Gemini app and 5 billion requests for images.
At an event in Switzerland this year, Hassabis said Isomorphic Labs, the drug spinoff, would have AI-designed drugs in clinical trials by the end of 2025. It isn’t clear what specific illnesses may be targeted by the drugs, but the company has said oncology will be a focus of its research.
In a statement, an Isomorphic Labs spokesperson said the company is advancing “multiple drug candidate programmes towards the clinic,” while building its clinical teams under the direction of a newly appointed chief medical officer. It didn’t specify further about the drug candidates or any timeline.
Hassabis remains undeterred from big science and long-term ideas.
A directive he has described in public is the need for AI to solve “root node” problems – a computer science term for the foundational point of a complex system, like disease. Discovering protein structures, for example, increases understanding of amino acids and biology, which could help combat cancer and other illnesses.
One area of concentration is Hassabis’ vision for AlphaAssist, a “universal assistant,” according to five people familiar with the efforts. Hassabis has said he foresees an assistant that understands human contexts so well it can make personalised recommendations that increase both productivity and wellbeing. While some of the work has already been alluded to by Google in public as “Project Astra,” neither the name nor the full concept of AlphaAssist has been previously reported.
Existing virtual aides thus far can answer questions but have limited ability to act. DeepMind foresees an assistant that doesn’t trip up on more sophisticated tasks. Hassabis has compared the idea to characters from science fiction, three of the people said. Those include J.A.R.V.I.S., or “Just A Rather Very Intelligent System,” the futuristic assistant from Marvel’s “Iron Man” franchise.
