Greg Jensen's argument is straightforward: when a small group of companies controls enough of any critical infrastructure to destabilize society, regulators eventually step in. In his view, AI compute—the specialized processing power, drawn largely from advanced chips and data centers, that trains and runs AI models—is next in line.

Jensen, co-chief investment officer of Bridgewater Associates—the world's largest hedge fund—is calling for oversight mechanisms modeled on the framework applied to systemically important financial institutions, better known as SIFIs, for any company controlling more than 5% of global or US AI compute resources.

Why the 5% threshold matters

The SIFI framework emerged from the 2008 financial crisis, when regulators recognized that certain institutions had grown so large and so interconnected that their failure could threaten the entire system. The designation triggers heightened capital requirements, stress testing, and regulatory scrutiny that ordinary firms never face. In the United States, the Dodd-Frank Act of 2010 created the Financial Stability Oversight Council specifically to identify such systemically important non-bank firms, giving the concept Jensen invokes a concrete statutory basis.

The timing of Jensen's remarks—reported around September 17-18, 2026, following an interview with The Information, a technology-focused news outlet—is not incidental. He cited projections that OpenAI and Anthropic together could command between 35% and 50% of the world's AI compute capacity within roughly two years. That degree of concentration, in his view, creates precisely the conditions that led regulators to regret their hands-off approach to financial conglomerates before 2008.

His preferred analogy is February 2020, the quiet weeks before COVID-19 became a global emergency. Jensen appears to view the AI governance debate as being at a similar inflection point: the window to build proactive structures is narrowing, and reactive regulation—the kind that arrives after a significant incident—will prove more costly and less effective.

Bridgewater's own AI footprint

Jensen's call for oversight carries an added layer of complexity given Bridgewater's own position in the AI landscape. The firm established AIA Labs in 2023, a dedicated AI investment and development unit that now employs more than 80 staff and manages approximately $4.5 billion in assets under management tied to AI-driven strategies.

Bridgewater has maintained that AIA Labs operates with deliberate human oversight built into its processes—a design choice that mirrors Jensen's broader argument: AI systems with significant influence over critical decisions should not operate without meaningful accountability structures.

What ownership caps would mean in practice

Translating Jensen's 5% threshold into actual policy is where things get complicated. The SIFI designation for banks rests on well-established global regulatory architecture, with the Financial Stability Board coordinating international standards and national regulators implementing them domestically. AI compute has no equivalent institutional structure yet, so any comparable oversight regime would have to be constructed through new legislation and cross-border coordination rather than added to an existing regulator.

Defining what counts as compute for regulatory purposes is genuinely difficult. Does it mean raw chip capacity, data center square footage, model training runs, or inference throughput? Each definition produces a different map of who controls what, and each creates different incentives for companies trying to stay under a regulatory cap.

The companies most directly in Jensen's crosshairs, OpenAI and Anthropic—two of the most prominent developers of large-scale AI systems—are both privately held and heavily backed by some of the largest technology and venture capital firms in the world. Regulatory concentration limits could complicate their current growth trajectories at a moment when both are making substantial infrastructure commitments.

Source: CryptoBriefing