The Announcement, Decoded
Piketon, Ohio is not a random site selection. The Portsmouth Gaseous Diffusion Plant once enriched uranium for nuclear weapons. It has the land, the power grid infrastructure, and the federal ownership history that makes it usable for large-scale industrial redevelopment. When the DOE partners with a private entity on a former federal industrial site, the project has a profile that differs from a purely private build, different permitting pathways, different political durability, different access to public infrastructure.
What was announced on March 20: SB Energy, a subsidiary of SoftBank Group, will develop a co-located AI data center and power generation complex at the Piketon site. Reuters confirmed the project is supported by $33.3 billion in Japanese funding under the U.S.-Japan Strategic Trade and Investment Agreement. SB Energy’s planned power generation component includes 9.2 gigawatts of natural gas capacity, as part of up to 10 gigawatts of total new generation. The project is affiliated with the Stargate AI infrastructure initiative, which also involves OpenAI and Oracle.
What SoftBank CEO Masayoshi Son said: Bloomberg confirmed Son publicly stated the project could channel $500 billion into a single campus and described it as potentially the “largest construction project in the country.” Son’s track record of ambitious numeric projections, the Vision Fund, SoftBank’s various headline commitments, is relevant context for how to weight this statement. The $500 billion figure is a CEO’s articulated ceiling for what the project could become. The $33.3 billion is what is currently committed.
Construction is expected to begin in 2026 per reports citing SoftBank. The first-phase cost range and an early 2028 completion target that appeared in some coverage could not be independently corroborated and are excluded from this analysis.
The Energy Math
Ten gigawatts of data center capacity is a number that doesn’t have real-world precedent. A single nuclear reactor generates roughly 1 gigawatt. A typical large hyperscale data center consumes between 100 and 500 megawatts. The United States’ entire installed solar capacity is approximately 180 gigawatts, but that’s across hundreds of thousands of installations. A single 10-gigawatt facility would be in a category of one.
That’s worth saying plainly: if the Ohio project reaches its stated scale, it will be the largest data center ever built, by a substantial margin.
The power generation side is the more immediately concrete element. Utility Dive confirmed the 9.2-gigawatt natural gas component, SB Energy is developing that generation capacity to power the campus. NextEra and AEP are involved in the energy infrastructure layer, per cross-reference reporting. Natural gas was the deliberate choice: it offers dispatchable, on-demand power that solar and wind cannot guarantee at the reliability levels a major AI training campus requires.
That choice has implications. Nine-point-two gigawatts of new natural gas generation represents a significant carbon commitment. AI’s energy consumption is already a subject of regulatory attention in the EU and growing scrutiny in the U.S. A federally backed project making a large-scale gas infrastructure decision gives implicit policy cover to other AI infrastructure developers weighing similar tradeoffs. Environmental review and permitting for 9.2 GW of new gas capacity in Ohio will not be frictionless, the permitting timeline is a real constraint on the 2026 construction start claim.
For energy investors: the Ohio project signals federal appetite for dispatchable power in AI infrastructure contexts. NextEra and AEP’s involvement as energy infrastructure partners positions them as direct beneficiaries of this build. Broader implications for grid load growth and natural gas demand are worth modeling against other announced AI infrastructure commitments.
Who Benefits
The Stargate affiliation is the connective tissue. OpenAI and Oracle are named partners in the Stargate initiative, of which this Ohio project is a component. A project of this scale needs a sustained anchor tenant for its compute capacity, OpenAI’s model training and inference demands are among the largest in the industry, and Oracle has been building out its AI cloud infrastructure aggressively.
SoftBank benefits in multiple ways. SB Energy gets a large, federally legitimized development project. SoftBank Group gets a physical expression of its AI infrastructure thesis, the group has made substantial bets on AI companies and this project creates direct infrastructure value to support those portfolio companies’ compute needs.
The Ohio regional economy stands to benefit from construction employment and long-term operational employment, though the project’s actual job creation depends heavily on how automated the facility’s operations are at full build-out. That question connects to the broader AI-driven workforce displacement conversation, a fully AI-optimized data center at this scale may employ far fewer people per gigawatt than prior-generation facilities.
Semiconductor and networking hardware suppliers are indirect beneficiaries. A 10-gigawatt data center at full build-out would represent one of the largest single procurement opportunities in the history of the AI hardware market. NVIDIA’s networking and GPU businesses, along with memory suppliers and optical interconnect providers, would be in the addressable market for that hardware. NVIDIA’s networking division generated $11 billion last quarter, a data point that illustrates the revenue scale already flowing from AI infrastructure investment at current levels. Ohio would be additive.
Who’s Watching Nervously
Other states that have been competing for large AI infrastructure projects now face a federally backed competitor. The DOE partnership confers advantages, site infrastructure, political durability, permitting support, that private-only projects can’t replicate. States without comparable federal site assets are disadvantaged in this competition.
Private data center operators and hyperscalers running their own infrastructure are watching the competitive implication. A federally backed campus at this scale, with power generation co-located, could deliver compute at a cost structure that affects market pricing for AI training and inference. Whether it reaches that scale, and on what timeline, will determine the competitive impact.
Environmental and clean energy advocates will scrutinize the 9.2-gigawatt gas commitment. The tension between AI’s power demands and decarbonization commitments is sharpening. A major DOE-affiliated project choosing natural gas as its primary power source will generate policy and advocacy responses that could affect future federally backed AI infrastructure decisions.
Investment Implications
For AI infrastructure market participants, the Ohio announcement is a signal on multiple dimensions.
First, the federal government has moved from observer to active participant in AI infrastructure siting and funding. The DOE’s involvement here isn’t just a ribbon-cutting, it’s a policy commitment with capital behind it. That changes the risk profile of other federally supported AI infrastructure proposals.
Second, the $33.3 billion in Japanese funding represents government-to-government capital flows entering the AI infrastructure space. This is not venture capital or corporate investment, it’s strategic national capital. Other governments, including those of the EU member states, are watching whether U.S.-Japan cooperation on AI infrastructure becomes a model for AI industrial policy more broadly.
Third, reporting on Jeff Bezos’s exploration of a $100 billion AI manufacturing fund, a separate event, and Samsung’s $73 billion semiconductor commitment announced the same week together suggest that very large AI-aligned capital commitments are clustering. Whether this represents coordinated industrial policy response or coincident private-sector conviction, the pattern is observable.
What to Watch
The gap between announcement and groundbreaking is where AI infrastructure projects historically lose momentum or quietly restructure. Watch for:
**Construction contract announcements** in 2026 that confirm actual build start and identify primary contractors, this converts “expected to begin” into verifiable fact.
**Permitting progress** for 9.2 GW of new natural gas generation in Ohio, this is a regulatory process with meaningful timeline uncertainty.
**Stargate partner commitments** from OpenAI and Oracle that lock in the larger investment vision beyond SoftBank’s current commitments.
**Congressional scrutiny**, a $33 billion federally affiliated project of this profile will attract oversight attention on energy policy, national security, and economic development grounds.
TJS synthesis: The Ohio announcement is real, significant, and worth taking seriously, at the $33.3 billion level. The $500 billion ceiling is a vision, and visions have a way of compressing under the weight of permitting timelines, interest rate environments, and shifting corporate priorities. The more durable story is what the DOE’s participation signals: the U.S. government has decided that AI compute infrastructure is a national asset class. That decision, not the specific gigawatt count, is what will shape how the next generation of AI infrastructure projects gets financed, sited, and built.