Bridgewater’s Greg Jensen Calls for Regulating AI Firms Like Systemically Important Banks
Dakin Campbell interviews Bridgewater managing CIO Greg Jensen, who argues that any firm controlling more than ~5% of U.S. or global AI compute should face heightened oversight “like we do with systemically important banks,” and that caps on compute ownership may be needed. He warns open-source models are especially hard to police because they can be RL-trained privately—Bridgewater itself finds that “extremely effective”—and projects OpenAI and Anthropic could control 35–50% of world compute within two years. On Bridgewater’s own AI: AIA Labs (founded 2023) has 80+ employees and a dedicated fund managing $4.5B; investment decisions in that stream are machine-made but risk controls, data acquisition and trade execution stay human-driven; Bridgewater rents compute rather than owning GPUs/DCs; partners include Thinking Machines for RL tooling; constantly benchmarks Anthropic vs OpenAI rather than locking in. Investing view: AI infrastructure build-out trade is largely priced in; firm holds a “very small position” and prefers disruption/adoption trades; models DC build through 2028–29 and sees funding/execution/regulation as key risks—not calling it a bubble. Caveats: interview/opinion; 5% threshold and 35–50% concentration figures are Jensen’s estimates; China-cooperation thesis is speculative.
Why it matters
High-value investor/fund intel for Dealroom: Bridgewater AIA Labs scale ($4.5B AI fund, 80+ staff), rent-vs-own compute posture, Thinking Machines RL partnership signal, and a concrete SIFI-style compute-regulation thesis that would reshape OpenAI/Anthropic/hyperscaler governance and fundraising narratives.
Executive takeaways
Jensen: treat >~5% U.S./global compute holders as systemically important institutions (bank-like oversight; possible ownership caps). • Open-source RL risk: Bridgewater finds private RL on open models extremely powerful and currently untracked—slowing frontier labs alone is “not nearly enough.” • Concentration warning: OpenAI + Anthropic could hold 35–50% of world compute in ~2 years; argues bipartisan case against monopolistic compute control. • Bridgewater AIA: $4.5B AI-managed fund; machine investment decisions with human risk/data/execution controls and kill switches; rents all compute; uses Thinking Machines + multi-model benchmarking. • Portfolio: AI build-out largely priced in; “very small position”; pivots to disruption/adoption; funding/execution/regulation flagged as 2027+ risks.
What The Information may be missing
No named AUM share of Pure Alpha vs AIA, exact cloud/RL partner commercial terms, or which specific models currently win Bridgewater’s benchmarks. Unclear how a 5% compute cap would be measured (training vs inference; owned vs reserved) or whether Jensen’s China-cooperation view has any Beijing-side corroboration. Competitors’ (Citadel, Two Sigma, Point72) AI fund structures not compared.
