Opinion: Don’t Fear the A.I. Bubble Bursting
By Carl Benedikt Frey, December 5, 2025
As artificial intelligence captures widespread public and corporate attention, many express concern over the sustainability of rapid investment and innovation in the field. However, economist Carl Benedikt Frey of the University of Oxford argues in this guest essay for The New York Times that a deflation of the A.I. investment bubble could be exactly what the technology sector needs to thrive and advance meaningfully.
The Enthusiasm and the Bubble
Former Google CEO Eric Schmidt famously said, “Bubbles are great. May the bubbles continue,” reflecting a common belief that the continued influx of capital into A.I. infrastructure—building more data centers, acquiring faster chips—will unlock tremendous innovations. This mindset suggests that through sheer scale and investment, artificial intelligence will soon achieve breakthroughs, from curing diseases like cancer to surpassing geopolitical rivals in technological prowess.
Yet, Frey cautions that simply pouring money into scaling existing A.I. models is not a guaranteed path to meaningful progress. He notes that past technological advancements have often emerged not during times of abundant resources but rather under conditions of scarcity that spurred efficiency and innovation.
Innovation Thrives Under Pressure
Frey points to historical examples of innovation sparked by crises. During the 1970s energy crisis, despite long gas lines in the U.S., businesses and researchers developed technologies such as more fuel-efficient engines, improved building insulation, and early hybrid and electric vehicles. Similarly, in the early 20th century, natural disasters such as the Mississippi River flooding reduced available agricultural labor, accelerating the adoption of mechanized farming equipment in affected regions.
This phenomenon is known to economists as “directed technical change”—innovation driven by necessity, prompting a focus on doing more with less.
The Need for a New Direction in A.I.
Today’s large language models, though impressive, primarily function by predicting the next word or phrase in a sequence based on statistical patterns learned from past data. Frey explains these models lack true learning capabilities once deployed; they don’t accumulate knowledge over time or innovate independently. He warns that simply scaling up these models—training them on bigger datasets with more powerful chips—will only yield incremental improvements and increased energy consumption, without fundamentally advancing artificial intelligence.
Instead, Frey advocates for a paradigm shift: developing A.I. systems that learn continuously and efficiently, much like human brains do, thereby extracting more value from each unit of energy. Such systems would not just mimic human language but could innovate and generate novel ideas beyond historical limitations.
Embracing a Leaner, Smarter Future
As investments in A.I. start to wane, companies and researchers will be compelled to focus on energy efficiency and smarter architectures. This pressure may catalyze breakthroughs that large-scale investments alone have failed to achieve.
Frey concludes that the bursting of the A.I. bubble should not be feared. Rather, it is an opportunity to steer artificial intelligence toward greater sustainability and ingenuity—advancing technology not through relentless expansion but through smarter, more directed progress.
Carl Benedikt Frey is an economist at the University of Oxford and author of “How Progress Ends: Technology, Innovation and the Fate of Nations.”
