The latest cycle of artificial intelligence exuberance may be cooling, with claims of transformative gains giving way to doubts about returns, strategy, and staying power. According to The Telegraph, assumptions underpinning the current AI surge—broad productivity gains, concentrated corporate winners, and unbroken improvement—now look fragile.
Business returns questioned
The article states that near-term business benefits are proving elusive. Citing research discussed by the outlet, it notes MIT reported that most AI investments—described as around 95 percent—produce no net return. In the same vein, IBM is said to have found that only one in four AI projects show a net RoI, with just 16 percent suitable for full enterprise-wide deployment.
Reliability is portrayed as a core weakness: the piece characterizes generative AI as a “linguistic trickster” that remains too flaky for automating serious functions. It also references Lloyds offering insurance for “unforeseen performance issues” in “AI-driven products and operations.” The report adds that a fear-of-missing-out dynamic has propelled adoption, but buyers increasingly expect prices to fall and may choose to wait.
Capital intensity without a moat
The Telegraph argues that massive spending on data centers and graphics chips lacks durable competitive advantage, as new features and techniques are rapidly copied—often within days—and open-source models are intentionally replicable. It adds that Chinese efforts are focusing on performance at far lower cost, and claims this shift has yet to be reflected in equity markets. The piece also asserts that OpenAI’s GPT-5 undermined the notion that spending significantly more automatically yields much better results.
External costs and uncertain demand
The report lists broader externalities, including impacts on education, creative work, news markets, and the spread of low-quality content, alongside new security holes that enterprises must address. Unlike the dotcom era, The Telegraph suggests there may be no large reservoir of unmet consumer demand ready to reignite growth after a downturn. Instead, it points to machine learning serving narrower niches—such as data analysis, prototyping, and language services—with AI skeptic Gary Marcus cited as envisaging sector revenues in the tens of billions. The piece concludes that, given AI’s history of “AI winters,” the coming one could be the coldest yet.
