Man Group: The AI Bubble — Hidden Risks and Opportunities

Core Thesis

Man Group's February 2026 analysis argues that while AI technology itself is transformative, "the inflated financial architecture supporting it may be unsustainable." The report identifies a bubble driven by leverage rather than genuine adoption, following historical patterns seen in railroads, fiber optics, and the dot-com era.

Key Risk Mechanisms

Closed-Loop Financing Problem The report describes a recursive financing cycle among hyperscalers (Microsoft, NVIDIA, Amazon, Meta, Google, OpenAI, Anthropic) where these firms simultaneously act as suppliers, customers, and investors. This creates "circular and divorced from the market" demand signals, generating reflexive demand risk where one firm's slowdown cascades across the cluster.

Asset-Liability Duration Mismatch A critical tension exists between lenders and equity investors. Private credit providers finance data centers assuming "seven-to-15-year useful lives" with stable cash flows, yet GPU chips face "approximately one year" effective economic lifecycles. This masks "a ticking time bomb in credit markets."

The Capex-Revenue Chasm Despite $200 billion in infrastructure spending, actual AI revenue generation lags dramatically. The report notes that token costs fall "more than 70% per year," requiring "more than 225%" annual demand increases just to maintain economics.

Inference Economics Problem

Current AI workloads concentrate in "role-playing and conversational entertainment" (difficult to monetize) and "coding assistance" (highly price-sensitive). This creates persistent margin pressure as "token prices are falling faster than inference demand can rise."

Likely Unwinding Scenarios

1. Capex Compression: A 20-30% GPU order cut by one hyperscaler would trigger "widespread economic cascade" 2. Data Center Overbuild: Low utilization mirrors the "telecom fibre bust" 3. Recursive Loop Breakdown: Microsoft's reduced OpenAI commitment signals this fracturing has already begun 4. Private Credit Crisis: Expected in "year three" (2027-2028) when "collateral creditors believed was worth 70-80 cents on the dollar may be worth 20-30 cents"

Winners vs. Losers

Survivors solve fundamental cost problems (chip designers, software optimization firms) or create efficient distribution infrastructure.

Losers include data center financiers and power infrastructure builders who assume "current power and space demands will persist or grow linearly" — described as "a fallacy."

Systemic Contagion

Risk is "metastasising through the economy" into utilities, insurers, pension funds, and retail interval funds holding data center REITS and infrastructure credit, most of whom don't recognize their exposure to "GPU cycles."

Conclusion

The technology survives; the financial architecture likely doesn't. The report emphasizes that "risk is increasingly migrating away from tech company balance sheets" into less sophisticated investors, creating conditions for "opaque, protracted, and potentially disorderly" workouts outside traditional banking oversight.