AI Power & Grid

The question

The operative question for this book is not whether data centers will need more power. They will. The question is where, in the chain from fuel to turbine to transmission to ratepayer, does the binding constraint sit, and which companies have the asymmetric payoff from that constraint being priced lower than it will clear. Interconnection queues in PJM and ERCOT have lengthened from roughly two years in 2020 to five-plus years today. That is the single most important operational fact in the cohort, and it is what separates the names that will compound from the names that will simply rerate.

Base rates

The base rate for utility-sector outperformance during load-growth inflections is mixed. The 1960s (nuclear buildout) delivered roughly a decade of above-market returns for regulated utilities, followed by a decade of below-market returns as rate cases lagged cost inflation. The 1990s independent power producer cycle ended in widespread bankruptcy. The 2010s renewables buildout produced a narrow set of winners (roughly 15% of the starting cohort) and a long tail of capital destruction. In each case, the companies that preserved capital had three things in common: regulatory compact in an accommodative state, balance sheet capacity to self-fund growth, and a technology mix that didn't require a policy environment to monetize. Our reference class is the top quartile of that 15%, not the median.

Why the consensus view is wrong (or incomplete)

The consensus has converged on two views we think are incomplete. First, that the load-growth story is best expressed through merchant power names with direct data-center PPAs. These names have already rerated 3-5x and now price in utilization and pricing assumptions that sit at the optimistic tail of plausible outcomes. Second, that regulated utilities are a defensive way to play the same theme. They can be, but only in the specific jurisdictions where constructive rate mechanisms let capex compound into rate base without regulatory lag eating the return. Most of the regulated cohort does not clear that bar. The information-efficiency gap we are exploiting is narrower: it is the equipment and engineering-procurement names whose backlogs extend past 2030 at pricing that the market is still modeling off pre-2024 comparable transactions.

Position construction

The book is organized into three sub-books.

Equipment and build-out primary (40%) — GEV 20%, NEE 20%. GEV is the cleanest expression of the turbine-and-grid-equipment thesis: order book extending past 2028, pricing power reasserting after a decade of margin compression, and a service tail that compounds at roughly 8% annualized. NEE sits in both sub-books — it is a regulated Florida utility with a constructive commission, and it owns the largest development pipeline in the US. We own it for the regulated floor with development optionality on top.

Merchant generation with data-center exposure (30%) — VST 10.5%, SRE 9.1%, ETR 6.6%, NRG 5.0%, TLN 2.9%. This sleeve is sized smaller than a pure-merchant-beta book would suggest, because we think the current pricing of PPA economics is at the high end of plausible. VST and TLN carry direct hyperscaler contracts; SRE and ETR are hybrids with regulated anchors and merchant upside. Weights reflect our read on contracted-cashflow visibility net of rerating risk.

Fuel and engineering tail (30%) — PWR 9.9%, CCJ 5.8%, FSLR 3.8%, OKLO 3.7%, BWXT 2.9%. PWR is the largest grid-construction contractor and the most leveraged pick-and-shovels expression. CCJ and BWXT are nuclear fuel-cycle and SMR component plays with multi-decade contracted visibility; OKLO is a smaller option on advanced reactors. FSLR is in the sleeve because utility-scale solar is the only form of incremental capacity that can be interconnected on a two-year rather than five-year timeline, and that timing optionality has value.

Asymmetric payoff

Our expected-value frame assumes three states. Base case (50%): load growth tracks 2.8% CAGR, merchant power prices stabilize at current forwards, and the equipment sleeve compounds at 12-15%. Upside (25%): load growth exceeds 3.5% CAGR and interconnection queues do not clear, which drives merchant power pricing and SMR deployment timelines; book delivers 25%+. Downside (25%): hyperscaler capex is cut 30%, load-growth forecasts revise to 1.5%, and the merchant sleeve draws down 40-50% while the regulated and equipment names protect to single-digit losses. Probability-weighted, we see a mid-teens expected return with a margin of safety anchored in the regulated-rate-base and equipment-backlog floor.

Three things that would change our mind

First, a sustained inflection in data-center power-per-query efficiency that breaks the load-growth trajectory before 2028. Second, a meaningful policy shift on interconnection reform that clears the queue materially faster than our base case; this would compress the scarcity premium in merchant power and the order-book premium in equipment. Third, a credible signal that hyperscalers are shifting toward behind-the-meter generation at scale, which would reroute the economics around the regulated grid entirely.

What we are explicitly NOT betting on

We are not betting on a specific nuclear technology winning; OKLO and BWXT are sized as options, not core positions. We are not betting on natural gas prices rising; the merchant sleeve is sized to work at flat gas. We are not betting on a federal policy tailwind; the regulated names chosen sit in state compacts we think clear without federal help. And we are not betting on the grid being rebuilt; we are betting on incremental capacity being added at prices the market is still modeling conservatively.