As the artificial intelligence industry increasingly consolidates around a handful of leading companies, the focus of AI regulation is expanding beyond model safety and data usage to encompass the core infrastructure that underpins AI development—computing power.

According to reports, Greg Jensen, co-chief investment officer at Bridgewater Associates, stated that the world's top holders of AI computing capacity should be subject to additional oversight similar to that imposed on large financial institutions. If a company controls more than 5% of U.S. or global computing resources, it could be designated as a "systemically important institution" and face stricter regulation.

During an interview, Jensen suggested that such entities could be governed under a framework comparable to that of systemically important banks, meaning once an organization holds a critical share of key resources, it would be required to adhere to extra regulatory scrutiny. Additionally, he proposed that any AI model entering the U.S. ecosystem should be bound by regulatory standards.

Jensen: AI Computing Power Is Highly Concentrated, Consider Setting a "Computing Capacity Cap"

Jensen believes the AI industry is forming a highly concentrated computing structure, noting that computing power has become an indispensable foundational resource for AI model training and operation. He said these companies could be treated as "systemically important institutions" and regulated in a manner akin to systemically important banks.

"We should bring any enterprise that holds more than a certain proportion—say, 5%—of global or U.S. computing resources into the regulatory framework."

This vision implies that the scope of AI regulation may no longer be limited solely to model developers, but could extend to companies that control substantial computing resources. Another idea Jensen put forward is more direct: regulators could consider limiting the scale of computing capacity any single company is allowed to own, preventing a few players from controlling an outsized share of computational resources. He noted that imposing a cap on how much computing power a firm can hold is not unacceptable, drawing an analogy to position limits seen in certain financial markets.