You may have only recently heard the name Demis Hassabis. He was named to Time magazine's "AI Architects" list, won a Nobel Prize for his work using AI to predict protein folding, and now heads Google's artificial intelligence division. In 2014, the search giant acquired DeepMind, the company he co-founded. Since then, Hassabis has leveraged his new parent company's vast resources to pursue the development of machines that surpass human intelligence, known as Artificial General Intelligence (AGI).
Since the acquisition, his most prominent achievements have been winning scientific acclaim for Google; his AI systems defeated the world's top Go players, and he personally received a Nobel Prize in Chemistry for related technology. However, tangible breakthroughs at the product level have remained elusive. This situation may change in 2026—if his unconventional ideas can be integrated into Google's second-generation smart glasses. For Google, which was caught off guard by ChatGPT three years ago, such progress could represent a complete turnaround.
Google plans to launch AI-powered smart glasses in 2026, aiming to compete with Meta's similar products. To this end, Google will partner with Samsung Electronics to manufacture the device, with one model featuring a tiny in-lens display for functions like navigation or translation. This could help restore the reputation of the team led by Hassabis in two key ways. First, the original Google Glass, with its awkward design and poor user experience, tarnished the entire category of smart glasses for years; an AI-powered glasses product that is both stylish and functional could completely reshape the public's perception of the technology.
Second, it could also validate Hassabis's long-held belief that chatbots are not the only path to powerful and useful AI. OpenAI's ChatGPT, powered by large language models (LLMs) trained on web content and supported by computational resources costing hundreds of billions of dollars, ignited the generative AI boom. But Hassabis has consistently argued that "world models," trained on simulations and the physics of the real world, will lead to the next major leap in AI.
Meta's AI chief, Yann LeCun, held a similar view but failed to secure Mark Zuckerberg's approval and recently left the company. While Zuckerberg is doubling down on chatbots, emulating OpenAI in the pursuit of superintelligence, Google has adopted a more diversified strategy, betting on both scientists and potentially paradigm-shifting technologies.
In 2023, Sundar Pichai, CEO of Google's parent company Alphabet, merged the company's two main AI divisions and placed Hassabis in charge, surprising many who expected the role to go to California-based engineering head Jeff Dean. Hassabis refused to relocate to Silicon Valley and was reluctant to adopt the company's market-driven mindset, instead remaining dedicated to scientific exploration. Yet, operating from his base in London, he appears to have successfully navigated the deep-seated transatlantic rivalry between the two divisions that had previously hindered progress.
In August 2024, Google spent a staggering $2.7 billion to rehire Noam Shazeer. The maverick scientific genius is a co-inventor of the Transformer neural network architecture—the "T" in ChatGPT—which forms the backbone of modern AI. Shazeer had left in 2021 after Google declined to launch his chatbot project, only for OpenAI to stun the world two years later with technology sharing the same roots. Upon his return in 2024, Shazeer was appointed co-technical lead for the Gemini chatbot, reporting to Hassabis. On paper, this arrangement seemed fraught with difficulty, as Hassabis does not share Shazeer's long-standing conviction that large language models are the path to AGI.
However, Hassabis, a former chess prodigy, is accustomed to thinking several moves ahead. Using his personal charm and diplomatic skill, he has managed to bridge the differences with Shazeer. The stakes are now higher than ever. "He stepped into a super-capital race that he didn't start, but now that he's in it, he intends to win," commented Sebastian Mallaby, author of the upcoming biography of Hassabis and DeepMind, *The Infinite Machine*. "While a big part of his personality loves science, an equally big part wants to win."
Yet, he remains far from a decisive victory. ChatGPT not only boasts a significantly larger user base than Gemini but also gained a two-year head start in expanding into the enterprise market—the key arena for AI profitability. (In contrast, Gemini's enterprise edition only arrived in October 2024.)
Nevertheless, Google's latest AI model can be considered a major success. Gemini 3 has topped several performance benchmarks, even prompting Sam Altman to sound a "code red" within OpenAI, citing "temporary economic headwinds." Currently, Gemini 3 has surpassed 650 million monthly active app users, while its "AI Overviews" feature in Google Search is estimated to reach a staggering 2 billion users. Known for its cautious and measured pace in AI, Google, the "tortoise," is now catching up to the "hare."
Shazeer deserves significant credit for this comeback. Silicon Valley insiders suggest he identified and fixed a fundamental flaw in Gemini, dramatically improving its training efficiency and enabling it to surpass ChatGPT on certain benchmarks. However, Hassabis's defining moment is yet to come. Successfully implementing a world model, such as the one advocated in his "Project Astra," could trigger the next paradigm shift after chatbots, creating AI that understands the spatial movement and interrelationships of objects.
Imagine glasses that can remember where you left your keys, understand what they see in three dimensions, and predict what might happen next in your environment. Compared to Meta's Ray-Ban smart glasses, which can only describe what the camera sees without truly understanding the physics of your world, this would be a monumental leap forward. It would also signal a step-change in AI assistants.
The pressure is mounting. Google must prove its AI capabilities can generate profits beyond its advertising business, and Hassabis urgently needs to transform one of his "moonshot" projects into a viable business. His track record so far gives pause for thought: although AlphaFold, the protein-structure prediction tool used by 3 million scientists, is highly acclaimed, it has not yet led to a single drug approved by the US Food and Drug Administration (FDA).
However, if Google's new smart glasses, powered by a world model, succeed and sell well, the company could gain a crucial advantage in the race for AI's killer app. Simultaneously, it will determine Hassabis's future—whether he remains a distinguished scientist within Google or becomes the foundational leader guiding it into its next era.
