Markets were up this week, with the S&P 500/Nasdaq 100 +1%+3%. Strong perfomance was driven primarily by continued strong corporate earnings and easing fears about Middle East conflict escalation. WTI crude fell ~5% on the week to $95/bbl. Our NPM Private Market Tracker*, which shows the average price performance of the 50 largest names in our internal Tape D® data, is +25% YTD vs. SP500/Nasdaq 100 +7%/+13%. (Bloomberg; NPM)

AI and the Coming Power Supercycle

For most of the AI era to date, data centers have been viewed primarily as real estate and computer hardware. Increasingly, however, the limiting factor for next generation AI infrastructure is electricity, not chips or land. The market has shifted from a competition for GPUs to a competition for megawatts.

US power demand was relatively flat for about 15 years following the 2008 financial crisis, primarily due to efficiency gains. This has now changed. The scale of projected electricity demand growth from AI is extraordinary. U.S. data centers consumed about 4% of total U.S. power in 2023, about the equivalent of Pakistan’s entire annual consumption. The Lawrence Berkeley National Laboratory report, the federal government’s reference forecast, projects data centers will consume 325-580 TWh by 2028, or 7%-12% of U.S. electricity. A modern AI campus can require more electricity than a steel mill or aluminum smelter, and these loads are appearing much faster and in far greater geographic concentration than utilities were designed to handle.

As a result, “speed to power” has become a critical competitive advantage in the data center industry. Today, many AI infrastructure firms are constrained because they have customer demand but lack deliverable electricity. Early movers that already control land with transmission access, utility relationships, and generation are dramatically better positioned than operators waiting in utility interconnection queues. (Bloomberg; NPM)

In many regions, the needed new transmission infrastructure will not arrive until the early 2030s. This delay is forcing new projects toward hybrid power architectures and “behind the meter” generation. The best positioned data center operators today are generally those with either existing power capacity or the balance sheets to effectively become power developers themselves. (Bloomberg; NPM)

CoreWeave is currently the strongest pure-play AI infrastructure company in terms of contracted power relative to deployed infrastructure. The company currently has more than 3.1 GW of contracted power capacity, with plans to scale toward 5+ GW by 2030. Active power capacity at the end of 2025 was roughly 850 MW, implying that CoreWeave already controls substantially more future power than it has currently energized. (Bloomberg; Company Reports)

Crusoe is also well positioned because it is vertically integrating energy and compute more directly than many peers. Crusoe’s Abilene, Tx facility is scaling toward 1.2 GW at a single site, while the company claims it has >1.6 GW under operation or construction and a development pipeline >10 GW. One of Crusoe’s strengths is that the company was originally built around energy infrastructure and stranded gas monetization before it pivoted into AI cloud infrastructure. The company is heavily focused on colocating compute with power supply and increasingly appears willing to develop dedicated generation solutions alongside data centers. (Bloomberg; Company Reports)

Nscale’s announced power ambitions are gigantic. The company recently announced plans tied to the Monarch Compute Campus in WV that could scale toward 8 GW of capacity, with Microsoft signing an LOI for up to 1.35 GW. Nscale also has renewable-heavy Nordic deployments, including large hydropower-backed facilities in Norway (Stargate Norway) and Portugal. However, Nscale to our knowledge has not disclosed a firm “currently contracted power” figure. (Bloomberg; Company Reports)

While being one of the oldest neo-data center players, Lambda is a relatively recent entrant to the hyperscale AI game. Unlike CoreWeave or Crusoe, Lambda historically focused more on AI cloud services and GPU infrastructure, rather than massive hyperscale campus development. Lambda has important strategic positioning with AI startups and enterprise customers, but the company’s disclosure on contracted power is more limited. (Bloomberg; Company Reports)

Growth in “Behind the Meter”

The “short power” dynamic is driving explosive growth in “behind the meter” (“BTM”) generation. Historically, most data centers simply connected to the utility grid. That model is breaking down because many grids are maxed out and expansion timelines are too slow relative to AI deployment. In Texas, for example, data center load is projected to exceed 40 GW by 2028 (Bloom Energy) vs. ~8 GW today. ERCOT estimates over half of the state’s electricity demand growth through 2031 (about 32 GW) will come from data centers.

With grid interconnection lagging, a meaningful share of new capacity to meet demand is being procured behind the meter. The first phase of OpenAI’s Stargate campus in Abilene, for example, is being powered by 360 MW of BTM gas turbines. These projects involve a mix of dedicated natural gas plants, reciprocating engines, microgrids, batteries, and hybrid generation systems located directly on the data center campus. We believe that over time the norm will become data center campuses fueled by a hybrid of both grid energy and BTM assets. In constrained regions, BTM penetration could be the majority. (Bloomberg; NPM)

One of the most underappreciated bottlenecks is backup generation equipment. Every hyperscale campus requires extensive diesel backup systems because uptime requirements are effectively absolute. Lead times for large diesel generators, transformers, and electrical equipment have stretched dramatically, in some cases extending several years. The shortage has become severe enough that active secondary markets, effectively reassigning assets from oilfield and aerospace applications to AI infrastructure, are emerging for industrial generation assets. Heavy-duty turbine OEMs are now quoting six-year lead times, with prices reportedly up nearly triple vs. 2019 levels. (Bloomberg; NPM)

What Is the Role of Nuclear and Renewables?

In the short term we expect natural gas to be the main swing generation fuel. Over the longer term, however, nuclear power may become the strategic backbone of AI infrastructure. Nuclear’s appeal is straightforward: it provides carbon-free baseload electricity with extremely high reliability and massive continuous output. (Bloomberg; NPM)

The AI industry increasingly recognizes that long-duration compute clusters may ultimately require dedicated nuclear generation. Microsoft’s 9/24 20-year PPA with Constellation Energy to restart Three Mile Island Unit 1 — now renamed the “Crane Clean Energy Center,” an 835 MW carbon-free facility expected online in 2027 — signaled that hyperscalers are now willing to directly underwrite nuclear projects to secure future electricity supply. (Bloomberg; NPM)

Small modular reactors (“SMR”s) are also attractive because they could theoretically be colocated with AI campuses and scaled incrementally with compute demand. While SMRs still face licensing and commercialization hurdles, they align closely with the operational profiles of hyperscale AI facilities. Existing nuclear operators, uranium suppliers, and reactor developers could therefore become major long-term beneficiaries of the AI supercycle. (Bloomberg; NPM)

Renewables will also play a significant role, although primarily within hybrid systems rather than as standalone power sources. Hyperscalers remain among the world’s largest purchasers of renewable energy through PPAs and dedicated solar and wind projects. However, there is increasing realism that intermittent renewables alone cannot support frontier AI workloads without massive storage or backup systems. As a result, the emerging architecture is likely to combine renewable generation with batteries, gas peakers, nuclear baseload, and intelligent load management. (Bloomberg; NPM)

The chart below shows performance by industry/sector within the top 100 names we track in the private secondary market. (NPM)