SoftBank’s Son Says Ohio Data Center to Be $500 Billion Project

SoftBank plans an AI-focused data-center campus in Ohio targeting 10 gigawatts of capacity and claiming to channel $500 billion into a single site; the first phase is expected to be ~800 MW, cost $30–$40 billion and finish in early 2028. The complex would be powered by roughly $33 billion of natural gas-fired electricity and sit on a former U.S. Energy Department uranium enrichment site. The project elevates SoftBank's Stargate ambitions but raises local backlash and policy concerns over water, power commitments and regulatory/political scrutiny ahead of U.S. elections.

Analysis

Concentrating massive compute capacity into a single campus creates asymmetric winners along the physical stack: owners/operators of hyperscale real estate and short-lead electrical equipment (transformers, switchgear, chillers) capture monopoly pricing power during the build window, while commodity cloud providers face stickier marginal costs if they must take capacity later at a premium. The build also reshapes the local industrial ecosystem — expect multiyear demand for high-voltage interconnects, modular substation builds, and liquid-cooling OEMs; bottlenecks in those won’t be smoothed by software and thus set a hard floor on near-term pricing for critical hardware.

On energy markets, a single large load materially alters capacity-auction dynamics and spark spreads in the host grid: merchant generators with fast-start peaking fleets and owners of incremental gas supply stand to capture outsized short- to medium-term margins, while incumbents with long lead-time transmission obligations face execution risk. Water-availability constraints will push design trade-offs toward higher-capex, lower-opex cooling solutions (air vs water vs closed-loop liquids), shifting the vendor mix and lifecycle cost profile in favor of specialized cooling integrators.

Key risks are implementation and political friction. Local permitting, interconnection queue churn, and component lead times are likely to stretch timelines by 12–36 months; a protracted delay flips the cashflow outlook and creates stranded-equipment risk for early vendors. On the demand side, a faster-than-expected ramp in chip supply or efficiency gains in AI models could compress the pace of new-builds; conversely, any geopolitical export controls that constrain chip supply would reinvigorate the build rationale and spike short-term equipment scarcity.

Consensus frames this as an unalloyed hardware and utility win, but that view underweights concentrated project execution risk and local political externalities. The practical alpha is in tradeable supply-chain bottlenecks and merchant power generators that can flex into newly created load, not simply the headline semiconductor or mega-cap cloud beneficiaries. Position sizing should favor optionality and convexity to capture those asymmetric payoffs while controlling drawdown from multi-year execution delays.

Key Decisions for Investors

• Long Digital Realty (DLR) — 12–24 months. Rationale: direct exposure to hyperscale land/lease pricing and premium rents during concentrated campus builds. Trade: buy equity with a 20% stop; target +30–50% on successful site activation; expect 12–18 month realization of rent roll upside.
• Long Eaton (ETN) or ABB (ABB) — 12–18 months. Rationale: short-lead, high-margin electrical distribution and transformer vendors will see multi-quarter order-book expansion. Trade: buy shares or 12–18 month call spreads sized for 2:1 reward:risk; set a 15% stop if order-book miss is reported.
• Long merchant generators (NRG or CPN) — 6–12 months. Rationale: capture widened spark spreads and capacity market clears as incremental load bids into auctions. Trade: buy equity or 6–12 month call options; target 25–40% upside if realized peak-load margins materialize, cut to break-even if regulatory curbs on merchant pricing are introduced.
• Pair trade — long data-center real estate (DLR) / short regional incumbent utility (AEP) — 12–24 months. Rationale: REITs capture outsized lease economics quickly; regulated utilities bear long lead-time capex and political pressure. Trade: 60/40 notional weighting to long REIT; use 20% stop on short leg to limit regulatory-event gamma.
• Optioned long on NVDA or AMD — 9–18 months. Rationale: convex upside to continued AI-capacity demand that drives GPU absorption; hedge with defined-risk call spreads to limit premium decay. Trade: buy 9–18 month call spreads sized to 3–4% of portfolio, target 3:1 reward:risk if AI compute demand surprises to the upside.