Demis Hassabis on the Paradox Slowing AI: Why Progress May Be Self-Limiting

Google DeepMind founder Demis Hassabis told Semafor that AI’s rapid advance is being constrained not only by critics but by practical, industry-wide limits. Shortages of critical components, a pullback from open research, and the full commercialization of AI systems are slowing how fast models can scale—constraints that may act as de facto guardrails as the field evolves.

Scaling Limits And Supply-Chain Pressures

Hassabis highlighted shortages of key hardware—most notably high-bandwidth memory (HBM)—as a major bottleneck. With more companies building accelerated-compute data centers, competition for semiconductors, memory and energy has intensified. "Right now, it’s memory chips, but it’ll probably be something else tomorrow," he said, underlining the shifting nature of resource constraints.

From Open Research To Commercial Secrecy

Another important friction is the retreat from open, public research. Firms increasingly keep model architectures, training data and optimization techniques private for competitive reasons. That reduces the cross-pollination of ideas that helped transform AI from a niche field into a fast-moving area of discovery. "There’s so much commercial pressure, so some of those things can’t be shared quite as openly anymore, which is a shame on the one hand, but it’s understandable," Hassabis observed.

Commercialization Shifts The Priorities

The industry has entered an era of full commercialization. Massive investment in data centers is now often about serving large user bases with models already built, not only inventing new approaches. "We’re now also in the era of full commercialization of these systems, so you’ve also got to balance serving with training," Hassabis said. That shift can slow fundamental research and reallocate talent and resources toward productization and reliability.

Public Pushback And Environmental Concerns

Public resistance to AI infrastructure is growing across political lines. Local communities are organizing against data-center construction; some political candidates run on anti-AI platforms blaming tech firms for energy-price spikes; climate advocates point to emissions and resource use. To reduce local friction, Microsoft recently launched an initiative to limit water consumption and shield communities from potential power-cost surges tied to data centers.

Using AI To Build Public Good

Hassabis argued that demonstrating AI’s tangible benefits in science and climate technologies could help blunt some opposition. Examples include AI-accelerated drug discovery in biotech, materials discovery for carbon capture and better batteries, and DeepMind’s own work applying AI to nuclear fusion research. "One of the only ways to tackle climate in today’s fragmented political world is to come up with some new technologies," he said.

The Paradox Of Progress

Hassabis warned of a paradox: the commercial success of generative AI could actually lengthen the timeline to the next breakthroughs. Heavy commercialization, tighter secrecy and resource competition may make it harder to reproduce the collaborative, open era that produced many early advances. "There were only a few dozen researchers working on AI properly and we all knew each other. We were, in most cases, really good friends," he recalled.

What This Means

For Hassabis, a slower pace is not necessarily bad. It may give society more time to debate ethical, commercial and philosophical questions before the field reaches more transformative milestones such as artificial general intelligence (AGI). He urged the industry to invest more in public-good applications of AI to demonstrate value and reduce political and social backlash.

“Look, it may be a good thing that it’s not as fast... We don’t have a lot of time to sort out before we get to [Artificial General Intelligence],” — Demis Hassabis

Overall, the combination of supply-chain limits, commercial incentives, growing public scrutiny and environmental concerns may slow AI’s headline pace of progress—even as companies like Google double down on data-center expansion and product integration.