October PMI – The Private Equity S&P, Plus AI’s railroad problem

The S&P 500 Moment for Private Equity 

The S&P 500 is the cornerstone benchmark for U.S. public equities—simple, standardized, and universally cited. As we’ve covered in prior editions, private markets continue to expand. While it’s hard to pin down an exact figure, data from McKinsey, Apollo, and other institutional sources suggest that private markets now represent roughly 10% of global capital markets—up from less than 5% a decade ago—and could rise to 15–20% by 2030 as institutional and individual allocations accelerate. Naturally, investors need a way to measure performance across this maturing asset class. 

Enter the S&P Private Equity 50. 

Launched last month by S&P Dow Jones Indices, the S&P Private Equity 50 brings the rigor and transparency of public-market benchmarking into private markets. Each index tracks the performance of 50 of the largest private equity funds from a given vintage year across North America and Europe, providing a representative snapshot of institutional private-equity portfolios. 

The index blends two components: the net asset values (NAVs) of the 50 constituent funds and a cash component that represents uncalled capital and distributions. That cash compounds under two different “reserve” scenarios—either the S&P 500 Total Return Index (for investors assuming public-market reinvestment) or SOFR + 2.9 bps (for a risk-free equivalent). Together, these offer a realistic view of both invested and uninvested capital, creating the first daily performance benchmark that mirrors how LP portfolios behave in practice. 

Yes, it’s priced daily—but don’t picture 50 buyout shops marking to market every morning. The daily movement comes from the cash component—uncalled capital and distributions—compounded at the reserve rate to reflect that cash isn’t expected to remain idle, while fund NAVs still update quarterly. 

For family offices and private-markets investors broadly, the implications are significant. The index introduces a way to compare private-equity performance, volatility, and deployment pacing (the rhythm of capital calls and distributions) using a standardized metric. 

AI’s Railroad Problem  

Everyone wants to talk about how powerful or intelligent AI is. Far fewer stop to ask: how profitable? The capital outlays fueling the AI infrastructure buildout are already staggering, and the revenue required to sustain them feels almost unfathomable.  

The Mismatch 

Deloitte estimates that hyperscalers and AI infrastructure providers will spend roughly $370 billion globally in 2025, a 44% increase year over year as they race to expand data center capacity. At those levels, industry revenues would need to climb to several hundreds of billions annually simply to justify the capital invested and earn a reasonable return. For perspective, the U.S. electric-power industry generated about $500 billion in revenue in 2023.  

Some projections are optimistic—Citi, for instance, forecasts AI revenues could reach $780 billion by 2030—but the magnitude of that requirement still underscores the imbalance between the speed of spending and the pace of monetization 

Depreciation Compression 

Investors often model compute infrastructure and data centers with a 10-year depreciation curve. Yet the advances in compute power and datacenter design may indicate a much shorter true useful life—3 to 5 years, perhaps less—as each generation of GPUs, cooling systems, and rack architectures quickly makes the last obsolete. Recalculate breakeven under those assumptions, and the required revenue target doubles or triples. 

Echoes of 2000: Circular Deals & Self-Funding 

In the fiber-optic and telecom boom, equipment providers sold to service operators, then financed those same customers—loaning them money or taking equity stakes—to keep orders flowing. When external demand lagged, the illusion unraveled and many investors were wiped out. 

Today, similar patterns are emerging: 

• Nvidia plans to invest up to $100 billion in OpenAI, combining chip supply with capital infusion—effectively funding its own customer. 

• AMD struck a multi-year supply deal with OpenAI that includes a warrant option allowing OpenAI to acquire up to 10 % of AMD shares—blurring the line between vendor and investor.  

• Nvidia also committed $5 billion to Intel, tying infrastructure and chip-architecture bets together 

• Microsoft has invested more than $13 billion in OpenAI, with revenue-sharing and preferential Azure access arrangements that effectively intertwine their economics. (AP News) 

These are not proof of structural weakness, but when growth depends heavily on internal capital recycling, it can be a late-cycle indicator. 

Lessons from Railroads and Fiber 

The parallel to past infrastructure manias is striking. Railroads transformed the U.S. economy, but their overbuilding in the 1870s triggered repeated financial panics. Fiber-optic networks did the same a century later—critical infrastructure whose long-term value survived even as most of its original investors did not. Both became essential to progress, but the builders rarely captured the ultimate returns. 

The Cleanup Trade  

History suggests that the greatest returns often come not during the frenzy but in the cleanup. In the aftermath of railroad or fiber bubbles, value was created by those who acquired infrastructure, rights-of-way, or distressed assets when speculation faded. 

Finding value in such an over-bought environment is a challenge, though private markets do uniquely provide niche opportunities. The categories we like to dive into are upstream theses (picks and shovels) and downstream theses (ride the wave of cheap intelligence). The most obvious of the upstream plays is energy infrastructure, which still has some rationally priced deals.