SoftBank plans $33 billion AI data center powered by natural gas

SoftBank plans a 10 GW AI data center on federal land in Ohio, backed by roughly $33B of natural gas-fired power and total data-center costs of $30–$40B including chips and equipment. SB Energy has sourced turbines totalling 9.2 GW to be delivered starting within a year and brought online by decade-end, with an additional 800 MW planned. The project is tied to broader U.S.-Japan trade discussions (cited as part of a $550B trade deal) and will be built at a former DOE uranium enrichment site.

Analysis

This announcement is a demand shock for long-lead industrial equipment, fuel transport, and AI hardware markets that will play out over several years rather than quarters. Expect order backlogs, longer lead times, and pricing power for large turbine and compressor OEMs — the incremental revenue accrues unevenly across suppliers depending on installed base compatibility and aftermarket service capture.

Midstream players and regional transmission owners will see structurally higher utilization and pricing optionality as pipeline and grid upgrades become gating constraints; congestion will create localized tolling economics that boost margins for owners of right-of-way and interconnect capacity. Conversely, vertically integrated utilities with weak merchant exposure or constrained balance sheets may be forced into dilutive financing if they must co-invest in upgrades or face stranded-asset politics from local ESG opposition.

Key risks are political and executional: federal/state permitting, interagency security reviews given the strategic nature of AI infrastructure, and a multi-year construction window that amplifies inflation and component lead-time risk. A meaningful pivot in US climate policy or a regulatory decision limiting fossil-fired generation on federal land would be a binary downside catalyst; conversely, public-private funding or tariff waivers could accelerate procurement and backlogs within 12–24 months.

Key Decisions for Investors

• Long GE (GE) — buy on confirmation of large OEM supply agreements or on a pullback >5% from today; horizon 12–24 months. Rationale: turbine and service aftermarket capture. Target +25–40% if backlog recognition accelerates; downside is order cancellations or warranty/technical issues.
• Long midstream pipeline owners (KMI, WMB) — accumulate over 6–18 months via cash-producing bonds or stock exposure. Risk/reward: 6–10% yield plus 15–30% upside as incremental capacity uses existing rights-of-way; tail risks are permitting delays and local opposition.
• Long AI hardware exposure via call-spread on NVDA (NVDA) — 9–15 month call spread to limit premium decay while retaining upside from sustained hyper-scale GPU demand. Reward: asymmetric upside if hyperscale procurement continues; risk: valuation shock or demand re-weighting to custom ASICs.
• Pair trade: long GE or KMI, short select pure-play renewable developers (NEE/NextEra favored shorts only tactically) — 12–36 month horizon. Thesis: short-term grid build and merchant gas demand favor equipment and pipeline owners over developers whose near-term returns suffer from higher financing costs and curtailment risk; manage with 1:1 notional and strict stop-losses for policy shocks.