Raghuram Rajan questions the massive spending on artificial intelligence as mainstream businesses remain slow to adopt the technology. He said, "The capability race may not yield lasting advantages for first movers." While companies build powerful models, Rajan notes that many firms still struggle to use them for simple everyday tasks.
Raghuram Rajan points out the slow adoption of artificial intelligence by mainstream businesses despite rapid advancements. He questions heavy spending by AI companies on sophisticated models that few users can fully utilize. The capability race may not yield lasting advantages for first movers, as competitors can catch up quickly. Rajan illustrates the gap between AI's promise and utility with a personal banking example.
Former Reserve Bank of India governor Raghuram Rajan sees a possible upside to a slowdown in data centre expansion, questioning whether the billions spent on developing more powerful artificial intelligence (AI) models will give companies a lasting competitive edge.
Writing in Project Syndicate, Rajan said AI adoption among mainstream businesses remains limited, with its practical capabilities still relatively modest despite rapid technological advances.
He pointed to the gap between AI's potential and its everyday use. While some early adopters are deploying bots to negotiate contracts and match invoices with deliveries, most businesses have yet to embrace the technology on a large scale.
Rajan also questioned whether the race to build increasingly sophisticated models was justified when relatively few users could fully utilise their capabilities.
One widely held belief in the industry is that the first AI laboratory to develop a self-improving model could establish an almost unbeatable lead as its capabilities advance exponentially.
Rajan, however, argued that competitors would likely catch up within a month or two. With leadership in AI development already changing hands frequently, he questioned whether any company could maintain such an advantage for long.
To illustrate the gap between AI's promise and its practical usefulness, Rajan cited his own banking experience.
While AI models can now solve variations of complex mathematical problems, he noted that his bank still has not developed a reliable chatbot capable of resolving a funds-transfer issue without human intervention.
Rajan said concerns about China catching up were also fuelling the rush to invest in AI. Chinese models have consistently trailed their leading US counterparts by only a few months, a gap he believes is unlikely to widen significantly unless Washington takes more effective steps to restrict China's access to advanced chips.
He also raised questions about trust among AI laboratory leaders, suggesting that some may believe they alone are capable of responsibly managing the powerful technologies being developed.
At the same time, the enormous spending on computing infrastructure is increasing pressure on AI companies to start generating revenue.
Rajan pointed to Anthropic's joint venture with Ode, which seeks to bring AI into non-technology businesses by placing Anthropic engineers within these companies. Such efforts could soon give banks far more capable chatbots, potentially threatening customer-service jobs.
For businesses, Rajan said, the choice comes down to two approaches: use AI to replace workers or redesign jobs and train employees to work alongside the technology.
The first could deliver quicker financial returns but also cause considerable social harm. The second would require more investment, time and planning, but could help protect jobs and maintain stability.
He argued that companies adopting AI must take greater responsibility for what happens to their workers. Governments could also encourage a more measured transition by rewarding businesses that invest in employee training rather than simply cutting jobs to boost short-term profits.
Among the possible measures Rajan suggested were taxes on corporate token use and credits for employee training, although he noted that few governments had introduced such policies.
His broader concern is that society is not yet prepared for AI to be deployed on a massive scale.
Against this backdrop, Rajan pointed to growing calls in the US to pause approvals for new data centres, driven partly by local community concerns and wider fears about the consequences of unchecked AI expansion.
A slowdown in data centre construction could have several effects. By limiting the computing capacity available to commercialise AI, it could force investors to reconsider how much they spend on developing expensive new models.
Higher computing costs could also make replacing workers with AI less financially attractive. Meanwhile, a slower pace of expansion would give chip and memory supplies time to catch up with demand and allow regulatory frameworks to develop alongside the technology.
Rajan stressed that the objective was not to halt AI development or its use, but to give businesses and governments more time to introduce it responsibly.
Such a pause could also allow regulators to examine emerging risks, including potential threats to the stability and integrity of the financial system, rather than leaving critical decisions largely in the hands of AI company founders.
As for competition with China, Rajan argued that the risks remained manageable, particularly given the limited prospect of any company or country establishing a permanent technological lead.
A collective slowdown, he suggested, could even create room for long-overdue discussions between Washington and Beijing on how to harness AI for humanity's benefit while reducing the risk of societies becoming unintended victims of the technology they are developing.