Good for ↑
Nvidia's reported $3B co-investment in an Ohio AI campus locks in hyperscale power demand on the PJM grid. The market is long Nvidia and data center REITs; it's underweighting the power generation and transmission companies that are the physical constraint. Long US large-cap independent power producers with PJM exposure.
Source: EconoTimes ↗
The obvious read: Nvidia is doubling down on the AI infrastructure buildout, validating the capex supercycle. That's already priced — Nvidia is up massively and XLK led sectors +2.0% this week. The non-obvious read is two steps further down the chain. An AI campus of this scale in Ohio — reportedly for OpenAI — will require hundreds of megawatts of baseload power, potentially exceeding 1 GW at full build-out. Ohio sits squarely in PJM Interconnection territory. PJM's interconnection queue is already overwhelmed: as of 2025, PJM had over 3,000 projects queued representing ~400 GW of requested capacity, with average wait times exceeding 4 years. A single high-profile, well-capitalized project backed by Nvidia, SoftBank, and OpenAI has the political and financial muscle to jump the queue — but it cannot conjure generation capacity that doesn't exist. This creates a structural squeeze for existing dispatchable generators in PJM — specifically gas-fired combined cycle plants and nuclear operators that can sign long-term power purchase agreements (PPAs) with data center operators. These are the assets that hyperscalers MUST contract with because they provide 24/7 carbon-free or reliable baseload power. US large-cap independent power producers (IPPs) with significant PJM footprint — think companies with gas and nuclear generation in Ohio, Pennsylvania, and the mid-Atlantic — are the direct beneficiaries. They hold the physical asset that cannot be replicated quickly. XLU led sectors this week at +2.7%, confirming the market is beginning to rotate into power names, but the IPP sub-sector (not regulated utilities) has more operating leverage to PPA repricing than regulated utilities whose rates are capped by state commissions. The third-order insight: Nvidia co-investing — rather than just buying chips — signals that the GPU maker is beginning to behave like a vertically integrated AI infrastructure company. This is the Microsoft/Azure or Amazon/AWS playbook applied to AI compute. When hyperscalers vertically integrate into physical infrastructure, they create captive demand for the underlying commodity inputs (power, land, water cooling). That captive demand is the most durable form of pricing power for IPPs — it removes volume risk entirely.
The market consensus is that this Nvidia investment is bullish for Nvidia itself, for data center REITs, and for AI semiconductor supply chains. Power infrastructure is viewed as a supporting actor — a cost center for hyperscalers — and the utility/IPP trade is considered 'already crowded' after the 2024-2025 AI power theme.
Why that may be wrong
Category confusion: the market is buying regulated utilities (XLU constituents) as the AI power proxy, but regulated utilities cannot capture the margin upside from PPA repricing — their returns are capped by state public utility commissions. IPPs operating in PJM's capacity and energy markets are the correct vehicle for the power scarcity trade, but they are being priced in the same bucket as regulated utilities. This is a structural mispricing: the assets with the most upside (merchant generators, unregulated nuclear and gas) are being valued on the same multiple as assets with capped returns.
The thesis breaks if the Ohio campus is powered by a large-scale on-site generation solution — for example, if SoftBank/Nvidia co-develop dedicated SMR (small modular reactor) capacity or a captive gas plant that bypasses the PJM market entirely. Observable signal: an announcement of a dedicated behind-the-meter generation project at the campus site, which would eliminate the PJM grid dependency and remove the scarcity premium for existing IPPs.
The strongest bear case is that hyperscalers, having learned from Northern Virginia's power constraints, are now deliberately building campuses in regions with surplus renewable capacity (Texas ERCOT, Midwest wind corridors) and will use on-site solar + battery storage to minimize grid dependency. If the Ohio campus is designed as a largely self-powered facility with behind-the-meter generation, existing PJM IPPs see no PPA demand uplift. The specific assumption most likely to be wrong is that the campus will rely on grid power rather than captive generation — given Nvidia and SoftBank's capital, a dedicated power solution is financially feasible. The data point that would flip conviction entirely: a construction permit or EPC contract filing for an on-site power plant at the Ohio campus location.
The closest historical rhyme is the 2000-2007 telecom/data center buildout in Northern Virginia (Loudoun County, the 'Data Center Alley'). Dominion Energy and AES signed long-term power contracts with early data center operators at rates that seemed expensive at the time. Within five years, those contracts were repriced at 2-3x as demand outstripped local generation capacity and transmission constraints emerged. IPPs with existing capacity in that corridor saw capacity market prices in PJM's MAAC zone surge. The lesson: when a well-capitalized anchor tenant locks in a geography, existing power generators in that geography get a scarcity premium that the market prices with a 12-24 month lag. A second analogy: when Amazon announced HQ2 in Northern Virginia (2018), local commercial real estate repriced within 6 months but the power infrastructure story took 18 months to fully emerge in utility earnings calls. We are likely at the 'HQ2 announcement' stage for Ohio power right now.
Nvidia's reported $3B investment in an Ohio AI campus confirms a massive, concentrated power demand node is being built in PJM territory, requiring hundreds of megawatts to over 1 GW of reliable baseload power at full build-out.
PJM's interconnection queue is already severely backlogged (400+ GW queued, 4+ year average wait times). New generation cannot be built fast enough to serve this campus, forcing the operator to contract with EXISTING dispatchable generators in the PJM footprint — specifically gas combined-cycle and nuclear plants in Ohio, Pennsylvania, and the mid-Atlantic. This drives PPA prices higher for existing capacity holders.
US large-cap independent power producers with unregulated merchant generation in PJM (gas and nuclear) are the direct scarcity beneficiaries. Unlike regulated utilities, their revenues are not rate-capped, so PPA repricing flows directly to earnings. As more hyperscale campuses anchor to Ohio and adjacent PJM zones, capacity market auction prices in those zones rise, creating a multi-year earnings tailwind for IPPs that the market is currently pricing as a regulated utility story rather than a merchant power story. The Nvidia co-investment structure — equity rather than just a supply contract — signals that AI infrastructure is entering a phase where physical power constraints are the binding limit on compute expansion, making PJM-footprint IPPs a structural long for 12-24 months.
No comments yet.
Sign in to comment
Earlier connections in the same sector that share a theme with this one.