Greg Jensen, co-CIO of Bridgewater Associates, proposed applying SIFI-style oversight to companies controlling over 5% of global or US AI compute capacity. This framework, established after the 2008 financial crisis, imposes strict capital requirements and stress testing on institutions deemed systemically important to prevent potential systemic failures.
Jensen cited projections suggesting that OpenAI and Anthropic could collectively control between 35% and 50% of global AI compute capacity within approximately two years. He argued that such concentration mirrors the risks seen in the financial sector before 2008, warning that reactive regulation after a crisis would be significantly less effective than proactive measures.
Bridgewater Associates maintains its own AI division, AIA Labs, established in 2023. The unit currently employs over 80 staff members and manages approximately $4.5 billion in assets linked to AI-driven investment strategies. The firm emphasizes that its internal processes include human oversight to ensure accountability.
Implementing these regulations presents significant challenges, as there is currently no global institutional structure for AI compute comparable to the Financial Stability Board. Furthermore, defining what constitutes 'compute'—whether it refers to chip capacity, data center size, or model training—remains a complex hurdle for potential policy development.
