SoftBank’s AI Power Play: Bypassing the Grid to Build the Next Compute Paradigm

The math here is a first-principles case for a paradigm shift. The exponential demand driver is clear: artificial intelligence is consuming power at a rate that is fundamentally reshaping the energy infrastructure. The numbers tell the story of a tripling curve. Today, data centers use 40 gigawatts of electricity. By 2035, that demand is projected to triple to 106 gigawatts. This isn't just growth; it's a step change in the scale of compute.

This shift is defined by hyperscale operations. The average planned data center will draw well over 100 megawatts, a massive jump from the current landscape where only 10% of facilities exceed 50 megawatts. Nearly a quarter of the new sites will be larger than 500 megawatts, with a few exceeding 1 gigawatt. This concentration of power at a single site creates a new kind of grid load, one that traditional utilities are struggling to accommodate.

The result is a tangible "power crunch." In regions like the PJM Interconnection, which covers much of the Midwest and East Coast, the grid's authority to manage new connections is under review. The independent monitor has argued that PJM has the authority to create a load queue for large data center projects, essentially forcing them to wait for grid upgrades. This isn't theoretical. In July 2024, a voltage fluctuation in Virginia caused 60 data centers to disconnect simultaneously, triggering a 1,500-megawatt surplus and forcing emergency grid adjustments. The instability is real, and it's pushing companies to act.

The market is responding with direct contracting and project delays. As one report notes, companies are delaying projects, contracting power directly from private producers, and installing inefficient backup generators. This is the validation SoftBank's move bypasses. By building its own power generation and transmission infrastructure, the company is not just securing energy-it's building the fundamental rails for the next compute paradigm. In the race to capture the exponential adoption of AI, the infrastructure layer is the true strategic asset.

The Infrastructure Layer: Building the Rails for the AI Paradigm

This project is a textbook example of vertical integration for a technological S-curve. SoftBank is not just buying power; it is building the entire energy and compute stack. The structure is a 10 gigawatt public-private partnership where SB Energy, a SoftBank Group company, is planning to build 10 gigawatts (GW) of new power generation-including 9.2 GW of natural gas generation that will connect to the local grid. This power is paired with a new 10 GW data center development at the Portsmouth Site. The co-location is the key innovation. By placing the massive power plant and the data center on the same DOE land, the project provides a guaranteed, large-scale, dispatchable energy source directly to the compute load. This is a direct solution to the "power crunch," bypassing the grid bottlenecks that are already forcing companies to delay projects or install inefficient backup generators.

The scale is staggering. A 9.2 GW natural gas plant would be the largest in the U.S., capable of powering millions of homes. The data center component, at 10 GW, aligns with the new paradigm of hyperscale facilities. This isn't a speculative bet on future demand; it's a pre-emptive build-out of the fundamental rails. The partnership with the U.S. Department of Energy and AEP Ohio further de-risks the transmission side, with SB Energy investing $4.2 billion with AEP Ohio to upgrade and build new transmission lines at no cost to the public.

Crucially, this isn't a blank check. SoftBank has already secured a pre-emptive customer base through its existing partnership with OpenAI. The company is a partner in the Stargate project with the AI giant, and they are in the process of building a "proof of concept" data center at GM's former Lordstown automotive assembly plant. This alignment with a major AI player provides a clear anchor tenant and validates the technical and commercial thesis. In the race for AI dominance, the company that controls the infrastructure layer-power, cooling, and compute-controls the adoption rate. SoftBank's Ohio bet is a strategic move to own that layer.

Environmental & Competitive Counterpoints: The Gas vs. Renewables Debate

The project's sustainability trade-off is stark. The 9.2 gigawatt natural gas plant would emit roughly 15 million metric tons of carbon dioxide per year. When methane leaks from the supply chain are factored in, the climate impact could be even larger. This is a direct investment in fossil fuel infrastructure at a time of global climate urgency. Critics point to a more lateral solution: wasted renewable energy. As one industry perspective notes, there is plenty of wasted energy around that could solve the AI problem. Purpose-built data centers co-located with existing wind or solar sites could utilize this surplus power, turning a waste stream into a compute resource. This approach offers a path to sustainable AI growth without locking in decades of high emissions.

SoftBank's project counters with a key advantage: dispatchability. Unlike intermittent solar and wind, natural gas can provide a reliable, on-demand power source. This is critical for the 24/7 operation of hyperscale data centers. In the near term, this reliability is a major selling point. Yet the long-term viability of this bet hinges entirely on regulatory and carbon policy shifts. The project's $33 billion price tag already reflects soaring construction costs, and it faces a multi-year permitting and build-out process. If carbon pricing or stricter emissions regulations tighten in the coming decade, the economic case for a massive new gas plant could unravel quickly.

The competitive landscape is also evolving. While SoftBank builds its own power rails, others are exploring alternative infrastructure. The company's own discussions with banks about borrowing money to fund investment of up to $10 billion in energy-related projects suggest a broader strategy that may include renewables. The project's fate is further clouded by its political framing. It is being touted as a centerpiece of a U.S.-Japan trade deal, but neither the regional grid operator nor Ohio regulators appear to have official plans for a plant of this scale. This blurriness around details and permitting adds a layer of execution risk that could delay or even derail the vision. For now, the Ohio bet is a high-stakes wager on the dispatchable energy of today, while the sustainability of tomorrow's AI infrastructure remains a work in progress.

Financial & Execution Risks: The Cost of Being First

The $33 billion price tag is the most immediate red flag. This is a staggering sum, nearly ten times the typical cost for a large gas plant. As one analysis notes, typical large-scale natural gas power plants cost between $1 billion and $4 billion to construct. SoftBank's proposed facility would cost nearly ten times that amount, suggesting either unprecedented scale, cutting-edge technology, or both. The sheer magnitude of this capital commitment creates immense pressure. If the project is delayed or scaled back, the financial hit would be severe. The bill for this gamble is already being felt in the broader energy sector, where construction costs have skyrocketed, embedding inflation into the final price.

Execution hurdles are equally daunting. A power plant of this scale is not a quick build. A power plant of this size is likely to take years, perhaps a decade, to complete. The timeline faces multiple risks. First, there is a known shortage of natural gas turbines, the core machinery for such a plant. Securing these components on schedule is a critical path item. Second, the permitting and regulatory process is a black box. Despite the project being pitched as a centerpiece of a trade deal, neither the regional grid operator nor regulators in Ohio were seemingly aware of plans for a 9.2-gigawatt plant. This blurriness around the official path adds a layer of execution risk that could lead to costly delays.

The project's ultimate success hinges entirely on the AI adoption curve staying on its exponential trajectory. The entire thesis is built on the forecast that data center power demand will triple to 106 gigawatts by 2035. If that adoption slows for any reason-economic downturn, technical bottlenecks, or a shift in AI compute efficiency-the return on this $33 billion investment would be severely pressured. The capital is locked in for decades, while the revenue stream depends on a future that remains uncertain. In this high-stakes bet, SoftBank is not just building a power plant; it is betting the farm on the next paradigm of computing. The cost of being first is not just financial, but existential for the project's viability.

Catalysts and Watchpoints: The Path to Exponential Returns

The Ohio bet is now a public partnership, but the real investment thesis hinges on a series of near-term milestones that will confirm its execution. The first and most critical watchpoint is the release of detailed permitting plans and interconnection agreements. The project is currently defined by a $33 billion natural gas plant with no public details on its configuration, permitting path, or target in-service date. For the plan to move from paper to power, SoftBank and its partners must file concrete plans with regulators and the grid operator. The absence of these blueprints is a major uncertainty that could delay the multi-year build-out.

Simultaneously, the financing structure must be spelled out. The company is already talking with banks about borrowing money to fund investment of up to $10 billion in energy-related projects. This broader strategic move signals a capital-intensive phase ahead. Investors need to see how the $33 billion for the Ohio plant will be funded-through equity, debt, or a mix-and what the terms are. The project's scale makes its financial viability a key catalyst.

On the commercial side, the ultimate validation will be tenant commitments. The project's power supply is only as valuable as the data center load it can attract. SoftBank's existing partnership with OpenAI provides a proof of concept, but the Ohio site needs its own anchor tenants. The first major AI data center companies to sign leases for the 10 GW of capacity will be the clearest signal that the integrated power-compute platform has commercial traction. This is the moment the infrastructure layer transitions from a speculative build-out to a revenue-generating asset.

Finally, the project's strategic context is expanding. SoftBank is not just building a plant; it is exploring ways to gain access to a large volume of Nvidia's graphics processing units, the critical hardware for AI. This move to secure compute chips alongside power underscores the company's ambition to control the entire stack. The path to exponential returns runs through these interconnected milestones: detailed plans, secured financing, and committed tenants. Until they are checked off, the Ohio bet remains a high-stakes wager on a future that is still being drawn.