When the numbers start with “one hundred five billion,” most people stop scrolling. I did the same the first time I saw the figure attached to Nvidia’s involvement in SoftBank’s Ohio project. Then I kept reading. The scale is almost hard to picture: 4.25 gigawatts of capacity, exclusive Nvidia gear for two decades, and a site that could eventually reach 10 GW. That is not just another data center announcement. It is a statement about where the next decade of artificial intelligence infrastructure will live and how it will be paid for.
The Ohio Project That Changes The Map
The Portsmouth Site in Pike County, Ohio, already carries history. Once a nuclear enrichment facility, the land is now being reborn as one of the largest planned AI computing campuses on the planet. SoftBank’s energy arm, known as SB Energy, is leading the build. The initial phase targets 4.25 GW of capacity, with the full vision stretching toward 10 GW. If completed at that size, the campus would stand as the largest data center complex in the world.
What makes the story especially interesting is the tenant. The entire facility is expected to be leased to OpenAI. That single detail transforms the project from a generic power-hungry warehouse into a dedicated AI factory. Every rack, every cable, every cooling loop will ultimately serve one of the most demanding artificial intelligence workloads currently imaginable.
How Nvidia’s Support Actually Works
Nvidia is not writing a single check for the full construction cost. Instead the company is providing a carefully structured backstop worth up to $105 billion. The support covers defined portions of lease and power payments plus a residual-value commitment. It becomes effective in phases as buildings come online between 2028 and 2030. As OpenAI begins making lease payments and capacity fills, Nvidia’s remaining exposure shrinks.
This approach echoes structures that other large technology companies have used to keep heavy infrastructure commitments off their main balance sheets. The goal is straightforward: lower the cost of debt for the project while giving lenders greater confidence. In practical terms, Nvidia is using its balance sheet strength and long-term visibility into demand to secure critical land, power, and shell capacity for its own future hardware sales.
Nvidia is supporting the LPS infrastructure at Ports-Pike for approximately 4 GW over a 20-year term, securing a site on which Nvidia compute will be exclusively deployed.
That exclusivity matters. The campus will not simply host Nvidia chips. It will run the company’s full-stack DSX AI factory platform—CPUs, networking fabric, and infrastructure software included. Over two decades the site is expected to cycle through multiple generations of Nvidia systems. Each generation could represent roughly 1.5 million GPUs and somewhere between $150 billion and $200 billion in revenue for the chipmaker.
Power Generation On Site
You cannot run a multi-gigawatt AI campus on the existing local grid alone. SB Energy plans to build approximately 10 GW of new generation capacity, of which 9.2 GW will come from natural gas. That new generation is designed to support up to 8 GW of IT load. In other words, the project is not just about servers; it is about creating a self-reinforcing energy island that can grow with demand.
SoftBank and SB Energy have also committed at least $4.2 billion toward regional grid upgrades in partnership with the local utility. The stated aim is to protect ordinary ratepayers from the cost of serving such a large new load. Whether that promise holds over twenty years remains an open question, but the public commitment is part of the political and regulatory package needed to move the project forward.
Why This Structure Matters For Investors
I have watched enough large infrastructure financings to recognize the pattern. When a technology company with strong cash flow and clear product demand stands behind a project, the cost of capital drops. Lenders sleep better. Equity partners gain confidence. The same logic that once applied to semiconductor fabs or undersea cables is now being applied to AI data centers.
There is also a strategic angle that goes beyond pure finance. Leading AI labs are growing faster than their own balance sheets and credit ratings can comfortably support. By locking in long-term land, power, and shell capacity, Nvidia is essentially reserving manufacturing capacity for its future products. It is the same discipline the company already applies to its supply chain: secure critical inputs when visibility into demand is high.
Critics have raised the circular-financing concern. Nvidia invests in or supports the very customers that buy its hardware. Jensen Huang has pushed back on that framing, arguing that the company is simply securing capacity the way any large manufacturer would. Whether markets fully accept that explanation will become clearer as more of these structures appear in 2026 and beyond.
The Path From 4.25 GW To 10 GW
The current commitment covers the first 4.25 GW. Nvidia has left open the possibility of supporting the remaining 3.75 GW later. That optionality is important. If OpenAI’s demand continues to scale and if power generation comes online on schedule, the campus could expand without requiring an entirely new site search or new regulatory battles.
Timing will be everything. Buildings are expected to reach service between 2028 and 2030. That window sits right in the middle of what many expect to be the next major wave of AI model training and inference demand. Missing the window would be costly for everyone involved.
What The Numbers Really Mean
Let me put a few figures side by side so the scale becomes clearer.
| Metric | Initial Phase | Full Vision |
| IT Capacity | 4.25 GW | Up to 10 GW |
| Nvidia Backstop | Up to $105 billion | Possible additional support |
| New Generation | Part of 10 GW total | 9.2 GW natural gas |
| Grid Investment | $4.2 billion minimum | Regional upgrades |
| Revenue Potential per Generation | $150–200 billion | Multiple cycles |
These are not small pilot projects. They are industrial-scale bets on the continued growth of generative AI. If the demand materializes, the returns for Nvidia, SoftBank, and the local economy could be substantial. If demand slows or power costs rise faster than expected, the residual-value commitments and phased guarantees will be tested in real time.
Local Impact And Political Reality
Pike County is not Silicon Valley. Bringing multi-gigawatt construction, permanent high-skill jobs, and new tax base into a rural Ohio county changes the economic landscape overnight. Local officials have generally welcomed the project, provided that ratepayers are shielded and environmental reviews are thorough.
The natural-gas generation plan will face scrutiny. Even with modern combined-cycle plants, the carbon footprint of 9.2 GW is significant. Proponents will argue that the alternative—importing power from distant coal or gas plants over congested transmission lines—would be worse. Opponents will push for more renewable capacity or stricter emissions controls. Those debates are only beginning.
Comparing To Other Mega-Projects
Large technology companies have already experimented with off-balance-sheet structures for data centers. The difference here is the combination of exclusive hardware commitment, on-site generation, and a single high-profile tenant. Most earlier projects mixed multiple tenants and relied more heavily on the existing grid. Ports-Pike is more vertically integrated from the start.
That integration carries both advantages and risks. Advantages include tighter control over power quality, latency, and security. Risks include concentration: if OpenAI’s needs change or if a future generation of Nvidia hardware requires different infrastructure, the long-term lease could become less flexible than a multi-tenant arrangement.
The Broader AI Infrastructure Race
Every major cloud and AI company is racing to secure power and land. The winners will not be those who simply order the most GPUs. They will be those who can guarantee the electricity and cooling to run them at full utilization for years. In that sense, Nvidia’s move is less about charity and more about protecting its own future addressable market.
I have found that the most interesting stories in technology often sit at the intersection of silicon and steel. Chips are glamorous. Power plants and transmission lines are not. Yet without the latter, the former sits idle. The Ohio project forces everyone to confront that reality at industrial scale.
Potential Risks Worth Watching
- Construction delays that push first capacity beyond 2030
- Higher-than-expected natural gas prices over a twenty-year horizon
- Regulatory changes affecting residual-value guarantees
- Shifts in AI model architecture that reduce GPU intensity
- Local political pushback if ratepayer protections weaken
None of these risks is theoretical. Each has appeared in earlier large infrastructure projects. The difference this time is the size of the numbers and the speed at which AI demand is evolving.
What Success Would Look Like
If the project reaches full 10 GW capacity on schedule, if OpenAI continues to train and serve models at the expected intensity, and if Nvidia’s successive generations of hardware land cleanly inside the facility, then the structure will be viewed as a template. Other technology companies will copy elements of the land-power-shell approach. Lenders will grow more comfortable with similar residual-value commitments. Local communities that host these campuses will demand stronger ratepayer protections and clearer community-benefit agreements.
In my view, the most under-appreciated aspect is the twenty-year time horizon. Technology cycles usually run much shorter. By locking exclusivity and residual value over two decades, Nvidia is betting that its architectural advantage will persist across multiple product generations. That is a confident statement.
Looking Ahead To 2026 And Beyond
Expect more of these announcements. The combination of AI demand, constrained power availability, and willing technology balance sheets creates fertile ground for creative financing. Some will succeed. Some will be restructured. A few may quietly fail. The Ohio project is simply the largest and most visible example so far.
For ordinary investors the takeaway is straightforward. The companies that control the critical inputs—chips, power, and long-term sites—will capture a disproportionate share of the economic value created by the next wave of artificial intelligence. Nvidia has just reinforced its position in all three.
Whether you follow the story for the engineering challenge, the financing innovation, or the pure scale of the ambition, one thing is clear: the map of global computing capacity is being redrawn in places most people have never heard of. Pike County, Ohio, is now on that map.
Final Thoughts On Scale And Responsibility
There is something almost surreal about discussing 10 GW of new load in a single location. For context, that is roughly the electricity demand of a mid-sized American city. Concentrating that much power—and that much economic activity—in one rural county creates both opportunity and obligation. The companies involved have promised to protect ratepayers and invest in grid upgrades. Keeping those promises over twenty years will test more than engineering skill. It will test institutional memory and political will.
I keep returning to the residual-value commitment. It is the quiet mechanism that makes the whole structure work. Nvidia is saying, in effect, that if the worst happens and the campus is left with empty buildings, the company will still stand behind a defined portion of the value. That kind of long-term skin in the game is rare. It also means Nvidia has strong incentives to keep the site fully utilized with its own successive generations of hardware.
Perhaps the most interesting aspect is how ordinary this kind of deal may become. Five years ago a $105 billion infrastructure backstop would have dominated every business headline for weeks. Today it feels like another chapter in an accelerating story. That normalization itself tells us how quickly the AI infrastructure race has moved from experimental to industrial.
The Ohio campus will not be the last of its kind. It may not even remain the largest. But it has set a new benchmark for ambition, structure, and the willingness of a leading chipmaker to put its own balance sheet behind the physical foundations of the AI era. Watching how the project unfolds between now and 2030 will reveal as much about the future of computing as any product launch or model release.
For now the numbers are on the table, the land is secured, and the first phases of construction are planned. The rest is execution—on schedule, on budget, and with enough power to keep the lights on for the most demanding software the industry has ever written.
