AGI is achievable by approximately 2030 and requires solving a small number of remaining conceptual hurdles, primarily continual learning, long-term planning, and memory, through the development of 'agents'.
The ultimate purpose of AGI is to serve as a tool to accelerate scientific discovery, tackling grand challenges in biology (AlphaFold, Virtual Cell), physics (fusion), and materials science.
Classical computing, through advanced neural networks, is sufficient to effectively model and solve complex natural and quantum phenomena like protein folding, challenging the necessity of quantum computers for many such tasks.
A dual strategy of developing massive, proprietary frontier models (like Gemini) while simultaneously open-sourcing highly efficient, distilled edge models (like Gemma) is optimal for driving both research and commercial platform adoption.
The societal impact of AGI will be immense, equivalent to ten times the industrial revolution at ten times the speed, necessitating the creation of an international oversight body for AI safety modeled on the IAEA.
2010
Co-founded DeepMind with a two-step mission: solve intelligence (build AGI), then use it to solve everything else. Established an initial 20-year timeline for achieving AGI, targeting approximately 2030.
Post-2016 (AlphaGo)
Following AlphaGo's victory, Hassabis identified it as the milestone proving DeepMind's algorithms were general enough to tackle major scientific problems, leading to the initiation of the AlphaFold project.
Post-AlphaFold II
After solving the protein folding problem, DeepMind spun out Isomorphic Labs under Hassabis's leadership to focus exclusively on applying AI to the entire drug discovery pipeline.
c. 2023-2024
Following the consolidation of Google Brain and DeepMind, Hassabis notes an acceleration in model performance due to unified compute resources, leading to the development of the Gemini series of models.
Future (1-2 years)
Hassabis anticipates a significant shift in AI capabilities, predicting systems will become more 'agentic' and autonomous, which will raise critical technical challenges around control and alignment.
▶AI for Science (AI4Science)Feb 2026
Hassabis frames AI not just as a commercial technology but as a fundamental tool for scientific discovery, analogous to how mathematics is the language of physics. This is exemplified by flagship projects like AlphaFold for protein folding, the 'Virtual Cell' for biology simulation, 'Weather Next' for meteorology, and collaborations in nuclear fusion and materials science.
Investors should view Google DeepMind not just as a product engine for Google, but as a high-risk, high-reward R&D lab whose breakthroughs in fundamental science could create entirely new, defensible markets, as seen with the creation of Isomorphic Labs.
▶The Path to AGI: Agents and Unsolved Problems
Hassabis has a clear vision for achieving AGI, which he believes is only one or two major breakthroughs away. He consistently identifies the key missing capabilities as continual learning, long-term hierarchical planning, and memory, and asserts that developing active, problem-solving 'agents' is the necessary path forward, building on concepts from reinforcement learning pioneered in systems like AlphaGo.
Analysts tracking the race to AGI should monitor progress in 'agentic AI' and continual learning as key indicators. Hassabis's focus suggests that labs demonstrating superior performance in these specific areas, rather than just LLM chatbot capabilities, may have a more direct path to AGI.
▶The AGI Revolution: A 100x Event
Hassabis quantifies the societal impact of AGI as being 10 times greater than the industrial revolution and occurring at 10 times the speed, effectively compressing a century of change into a decade. He believes that while AI is overhyped in the near-term, its long-term revolutionary impact is still significantly underappreciated.
This '10x at 10x speed' framework implies that traditional economic and market forecasting models are inadequate for capturing the potential disruption. The primary bottleneck he identifies is access to large-scale compute, making the semiconductor and energy sectors critical enablers and investment areas for this thesis.
▶Strategic Commercialization and Openness
Hassabis employs a dual-pronged strategy for commercialization and research. He advocates for keeping frontier models proprietary while strategically open-sourcing smaller, highly efficient 'edge' models like Gemma to foster developer ecosystems and secure platforms like Android. This is complemented by spinning out specialized companies like Isomorphic Labs to tackle specific, high-value verticals like drug discovery.
This strategy suggests a belief that the primary long-term value is in owning the frontier research and its direct application in specialized domains, while commoditizing smaller models to build a defensive moat around Google's ecosystem. Competitors relying solely on open-source or closed-source strategies may be at a disadvantage.