Analysis
Bridgewater Associates co-CIO Greg Jensen argued this week that the largest holders of AI compute warrant a form of regulatory scrutiny closer to what governs the world's biggest financial institutions than anything currently applied to tech companies, according to The Information. Jensen's specific proposal: anyone controlling more than 5% of U.S. or global compute should face scrutiny modeled on systemically important financial institutions -- a defined, numeric threshold rather than a vague call for guardrails.
The Concentration Number Driving The Argument
Jensen's underlying claim is what makes the proposal concrete rather than rhetorical: he expects OpenAI and Anthropic together to command somewhere between 35% and 50% of world compute within two years. That is, in his framing, an unusual level of concentration for a resource this foundational to sit almost entirely with a small number of private companies -- a comparison he draws directly to why banking regulators treat institutions above a certain size differently than smaller competitors, since their failure or misuse carries systemic rather than merely private consequences.
“He additionally proposed that any AI model entering the U.S.”
Borrowing From Existing Regulatory Machinery
Rather than proposing new legislation from scratch, Jensen floated caps modeled on futures-market position limits -- rules that already exist to prevent any single trader from cornering a commodity market -- alongside bank-style supervision that could run through existing statutes or a new law built on the same template. He additionally proposed that any AI model entering the U.S. ecosystem be bound by baseline regulatory standards regardless of where it was developed, extending the compute-concentration argument into a broader claim about model governance.
Why A Hedge Fund CIO's Voice Carries Different Weight
Jensen made a related point in a separate Information briefing this week: Bridgewater itself holds only a "very small position" in the AI buildout trade, a disclosure that undercuts the obvious rebuttal that a hedge fund executive calling for AI regulation is simply talking his own book. That combination -- calling for compute caps while disclosing minimal direct exposure to the trade those caps would constrain -- is part of why this lands differently than similar warnings from AI-safety researchers or rival lab executives who have more direct competitive stakes in how the rules get written.
The Competitive And Policy Backdrop
Jensen's intervention arrives amid a broader shift in how frontier labs are discussing their own pace of development. Anthropic CEO Dario Amodei's public essay calling for labs to slow capability gains drew public agreement from Sam Altman and Elon Musk earlier this month, and Google DeepMind separately launched an institute this week explicitly to widen public debate on AGI risk and governance. Jensen's compute-cap proposal is a more specific, more measurable version of the same underlying concern -- less about pausing model capability and more about preventing any handful of firms from cornering the physical resource capability depends on.
The case against taking this too far, too fast: compute concentration caps modeled on banking regulation assume compute is as fungible and measurable as capital reserves, and it isn't -- a GPU cluster's usable capacity depends on interconnect, power availability and software efficiency in ways a bank's capital ratio does not. Critics will also note that artificially capping OpenAI or Anthropic's compute share could simply shift concentration to whichever cloud provider or sovereign-backed neocloud absorbs the excess demand, without changing the underlying dynamic Jensen is worried about. No U.S. regulator has proposed implementing anything resembling Jensen's specific threshold, and Bridgewater itself has no formal regulatory authority to act on the idea -- this remains, for now, an argument rather than a policy.
