- Institutional investors — pension funds, insurers, sovereign wealth funds — are demanding more detailed reporting from private equity managers including Apollo, Blackstone, and Brookfield on AI-related holdings after discovering that bets on data centers, semiconductors, power generation, and AI software are spread across infrastructure, credit, real estate, and PE funds simultaneously, making aggregate concentration impossible to assess without custom analysis.
- One Canadian pension fund declined a “potentially lucrative” data center co-investment specifically to avoid AI concentration; Florida’s $306.8B State Board of Administration is now auditing its aggregate AI exposure across all fund relationships; and Ontario Teachers’ ($303.2B AUM) is actively quantifying its direct exposure to what its CIO calls the “AI complex” — a term now in active use at major allocators.
- Circular financing structures are a growing concern: Amazon has invested in OpenAI, which uses Amazon Web Services compute; Microsoft and Nvidia back Anthropic, which relies on their infrastructure — creating cross-exposure loops where a single AI sector downturn ripples through counterparties simultaneously, and where investment decisions may be conflicted by commercial relationships between investor and investee.
- Semiconductor financing is flagged as a specific structural risk: chip loans structured with 5-year repayment lives (matching semiconductor useful life) are being packaged into infrastructure platforms, but critics argue chips are fundamentally not infrastructure assets — and residual-value backstops that assume ongoing chip demand in excess of supply may be systematically mispricing technology obsolescence risk.
What Happened?
Bloomberg interviewed more than a dozen executives across private equity firms, pension funds, insurers, and sovereign wealth funds to document a growing LP revolt — not a withdrawal of capital, but a demand for transparency that the industry is struggling to satisfy. The core problem: AI-related bets are not classified consistently. Data centers appear in infrastructure, real estate, or private equity funds depending on the manager; semiconductor financing appears in credit or infrastructure; power generation deals are categorized under energy. A single LP invested across multiple strategies at a single manager may have AI exposure that is, in aggregate, far larger than any single fund’s allocation would suggest — and the LP has no systematic way to know without custom analysis that most managers are not currently providing.
Why It Matters?
The ILPA trade group representative’s comparison to Covid is apt: at the pandemic’s onset, investors scrambled to identify sector exposure across portfolios that had not been built with pandemic risk in mind. AI concentration is a structural analog — portfolios were not built with a single-technology-theme risk framework in mind, and standard fund classifications are not revealing where that risk lives. The circular financing problem is arguably more acute: if Amazon, Microsoft, and Nvidia are all simultaneously investors in and customers of each other’s AI infrastructure, a demand disappointment from one triggers losses that flow directly back to the others. The chip financing risk is the most technically specific: a 5-year semiconductor loan that relies on residual value assumes technology doesn’t obsolete the chip — an assumption that is historically dangerous in the semiconductor industry and that current competitive dynamics (Chinese firms, rapid AI model changes) make particularly fragile.
What’s Next?
ILPA and major pension consultants are likely to push for standardized AI exposure reporting — possibly including a common taxonomy for categorizing AI-related holdings across fund strategies. Individual managers who build that transparency proactively (disclosing aggregate AI exposure on a portfolio-company level across all funds) may gain a fundraising advantage as LPs increasingly treat transparency as a due diligence requirement. The broader question — whether the AI infrastructure buildout is a dot-com-era misallocation or a genuinely demand-justified investment cycle — remains open. But the concentration of LP capital in a single thematic bet, channeled through fragmented fund structures with limited cross-fund visibility, is itself a systemic risk regardless of the underlying thesis’s ultimate correctness.
Source: Bloomberg
