Bridgewater’s Greg Jensen wants dominant compute holders regulated like systemic banks

Bridgewater managing chief investment officer Greg Jensen argues that the largest holders of AI computing capacity should face heightened oversight analogous to the rules for systemically important banks. He suggested a threshold such as control of more than 5% of US or global compute.

The analogy is about concentration and spillovers, not balance sheets. A few laboratories and cloud companies control clusters, chips, proprietary models and access for thousands of downstream businesses. A serious security failure, abrupt withdrawal of service or unsafe system released from one of them could propagate through customers much as distress at an interconnected bank spreads through finance.

Bank regulation offers several possible tools—enhanced disclosure, stress tests, capital buffers and resolution plans—but none maps cleanly onto compute. GPUs are not deposits, measuring effective capacity across different chips is difficult, and obligations based on share could entrench incumbents by making scale expensive for challengers. Oversight would need to specify the risk it is reducing rather than simply borrowing a prestigious regulatory label.

Jensen also said Bridgewater now has only a “very small position” in the AI build-out trade. What was an extraordinary investment two years ago is, in his view, largely reflected in prices. That does not mean AI demand is ending; it separates a technological thesis from the price paid for exposure to it.

The interview is useful because it joins two questions usually discussed apart: concentrated compute may be a public-risk problem even after the obvious financial trade becomes less attractive.