Something shifted this week in the quiet corners of southern Ohio that could redefine how artificial intelligence actually gets built. OpenAI signed a 20-year lease for an initial 4.25 gigawatts of AI capacity at a new campus, with Nvidia stepping in as the heavyweight backstop. The numbers alone feel almost unreal. We are talking about infrastructure measured in gigawatts, guarantees capped at $105 billion, and a timeline that stretches well into the next decade. I’ve been watching these power-and-land deals for a while, and this one lands differently. It is not just another press release. It is a concrete bet that the hunger for compute will keep growing long after the current hype cycle cools.
Why This Ohio Deal Matters More Than Most AI Headlines
Most people hear “data center” and picture rows of blinking servers in some anonymous warehouse. This project sits on a different scale. The PORTS-Pike Technology Campus will rise on the site of the former Portsmouth Gaseous Diffusion Plant and surrounding land in Pike County. Part private, part federal. SB Energy will build it, own it, and run it. OpenAI gets the long-term rights to the capacity. Nvidia supplies the exclusive AI compute stack and provides residual-value guarantees so the whole thing can actually get financed.
The first phases are scheduled to come online starting in 2028. That is not tomorrow, but in infrastructure terms it is aggressive. Securing land, power, and the building shell for 4.25 gigawatts of IT load already required serious coordination with AEP Ohio, the Department of Energy, and the Department of Commerce. Nvidia also holds an option on another 3.75 gigawatts. If exercised, the campus could eventually reach 8 gigawatts of AI capacity. Eight. Let that number sit for a second.
In my experience following these projects, the hard part is never the servers. It is the electrons and the real estate. You cannot simply order a few hundred megawatts the way you order GPUs. Utilities need years to plan generation and transmission. Communities need to be brought along. Ratepayers need protection. This deal tries to address all three at once.
The Structure of the Lease and Nvidia’s Role
OpenAI remains the primary obligor on the lease payments. That point gets repeated for a reason. Nvidia’s residual-value guarantees only kick in under specific failure scenarios. If OpenAI becomes insolvent or stops paying, Nvidia steps in to cover the difference between a guaranteed minimum value and whatever SB Energy can recover from a new tenant or a property sale. Nvidia can also choose to assume the lease, push for a replacement tenant, start a sale process, or simply let the lease end. There is even a one-year pause option where Nvidia covers certain project costs while the situation is sorted.
The combined payment obligations under the guarantees cannot exceed $105 billion. Each guarantee generally becomes effective when the related lease section starts, provided the ready-for-service conditions are met. The guarantees can run until the 20th anniversary of the applicable lease, but they can also terminate earlier if OpenAI reaches a satisfactory credit rating or meets other exit conditions.
Nvidia CEO Jensen Huang addressed the circular-financing concerns head-on. OpenAI, not Nvidia, will make the lease payments. If OpenAI does not need the capacity, the site can be offered to other eligible customers in Nvidia’s network. Cloud providers, enterprises, AI labs, startups. The hardware is described as versatile and fungible. That flexibility is part of the pitch.
Nvidia compute is versatile, fungible and broadly adopted.
Alongside the guarantees, Nvidia is investing $1.5 billion directly in SB Energy. That puts the chipmaker alongside SoftBank Group and OpenAI as capital partners. The money supports data center projects and local commitments. It also signals that Nvidia is no longer content to simply sell chips. The company is locking down the physical footprint those chips will occupy.
Power Generation and Grid Reality
SB Energy and SoftBank plan to build at least 10 gigawatts of new electricity generation to support the 8 gigawatts of potential IT capacity. They have also committed to invest at least $4.2 billion in regional grid infrastructure through an agreement with AEP Ohio. The design goal is explicit: existing Ohio electricity customers should not carry the cost of this new load.
That protection language matters. Large data center projects have faced pushback in several states precisely because ratepayers feared subsidizing the power needs of hyperscalers. Here the developers are saying they will build the generation and the transmission upgrades themselves, or at least fund them in a way that isolates the cost. Whether that promise holds through the full construction cycle remains to be seen, but the public commitment is there.
Rich Hossfeld, co-CEO of SB Energy, framed the approach as building the site while protecting ratepayers and investing in southern Ohio. The project is projected to support about 35,000 construction jobs through 2032 and create roughly 2,500 permanent operating positions. An $80 million community benefits fund has been established, with OpenAI adding $40 million to SB Energy’s original commitment. The money is earmarked for affordable energy programs, workforce training, employment initiatives, and local economic development.
Sam Altman has said OpenAI wants Pike County residents to benefit through jobs, business opportunities, and community investment. That kind of language is standard in these announcements, yet the size of the fund and the dual commitment from OpenAI and SB Energy give it more weight than the usual vague promises.
What OpenAI Will Actually Run There
The campus will use Nvidia’s DSX AI factory platform across the initial 4.25-gigawatt buildout, subject to limited exceptions. The platform packages facilities, hardware, networking, and software into a single system. OpenAI’s systems will run on Nvidia GPUs, CPUs, and networking gear. Huang has described land, power, and finished data center space as essential resources in an industry where companies must lock in large power commitments before they can even install the computing equipment.
In practical terms, this means OpenAI is securing long-lived infrastructure so it can deploy what it considers the most productive AI factories. The phrase “AI factories” is deliberate. It treats compute capacity less like a cloud service and more like industrial production capacity. You reserve the plant, you staff it, you keep the power flowing, and you run the models.
I’ve found that this framing changes how people think about the risk. A traditional cloud customer can scale up or down with relatively short notice. A 20-year lease with residual-value guarantees looks much closer to a manufacturing commitment. Once the power is contracted and the shells are built, walking away becomes expensive for everyone involved.
The Broader Power-and-Compute Squeeze
OpenAI’s lease is not happening in isolation. AI companies are racing to secure electricity, grid connections, and suitable land. Advanced computing projects often require years of preparation because utilities must add generation and transmission capacity before servers can begin operating. That lag creates scarcity. Scarcity creates premium pricing for any site that already has power or a clear path to it.
For crypto investors the same dynamic has sharpened interest in Bitcoin miners that control powered land and existing grid connections. Several miners have been converting existing sites or planned capacity into high-performance computing facilities. The infrastructure can serve AI customers as well as proof-of-work networks. The same substations, the same cooling, the same real-estate footprints can be redirected when the economics favor one use over another.
One notable fund managed by a former OpenAI researcher held positions in a cluster of mining companies as part of a deliberate power-and-compute strategy. The disclosed equity exposure at the end of the first quarter reached $13.67 billion. That is not a casual side bet. It is a recognition that the bottleneck has moved from chips to electrons and physical sites.
Perhaps the most interesting aspect is how quickly the market has adapted. Miners that once competed purely on hash rate are now marketing available power capacity and data-center readiness. The Ohio deal reinforces that narrative. If OpenAI needs 4.25 gigawatts just for its initial phase, other AI labs and cloud providers will need similar volumes. Anyone who already controls the power has leverage.
Financing Risks and Credit Considerations
The residual-value structure is designed to make the project bankable even if OpenAI’s credit profile is still evolving. Nvidia absorbs the tail risk of a default, but only up to the $105 billion cap and only after recovery efforts. OpenAI has agreed to reimburse and indemnify Nvidia for any amounts paid under the guarantees. That indemnity is important. It keeps the economic burden on OpenAI in most scenarios.
Still, the size of the guarantee is large enough that markets will watch the credit metrics closely. A 20-year lease on multi-gigawatt capacity is a long-duration obligation. If AI demand softens or if OpenAI’s revenue growth slows, the fixed nature of the payments could create pressure. Nvidia has given itself tools to manage that pressure, including the ability to step in and re-lease the capacity. The fungibility argument is the safety valve.
Huang’s comments make clear that Nvidia views the capacity as resellable. That assumption underpins the whole guarantee framework. If the hardware and the site can serve other customers without major reconfiguration, the residual value stays high. If the site becomes specialized in ways that limit alternative use, the residual value drops and the guarantee becomes more expensive for Nvidia to honor.
Community Impact and Local Economics
Pike County is not a traditional tech corridor. The decision to place a multi-gigawatt AI campus there is itself a statement. The former gaseous diffusion plant site carries history and existing infrastructure that can be repurposed. The surrounding rural economy stands to gain construction employment for years and permanent operating jobs afterward. The community benefits fund adds a direct financial channel for local priorities.
Thirty-five thousand construction jobs through 2032 is a meaningful number in a region that has seen industrial employment decline. Two thousand five hundred permanent roles is smaller but still significant for the local labor market. Training programs tied to the fund could help residents move into those operating positions rather than watching skilled workers arrive from elsewhere.
Whether the economic multiplier effects materialize as projected depends on execution. Large projects sometimes over-promise on local hiring and under-deliver on supply-chain participation. The dual funding of the benefits package and the explicit ratepayer-protection language are attempts to lock in better outcomes. Time will tell how well those intentions survive the construction schedule.
OpenAI’s Broader Capacity Strategy
This Ohio commitment sits alongside other moves OpenAI has made to diversify its compute sources. Earlier this year the company expanded access through additional cloud providers after adjusting its long-standing relationship with its primary partner. Models and coding agents became available through new platforms. The pattern is clear: OpenAI is reducing concentration risk by spreading capacity across multiple providers and now across owned long-term leases.
At the same time, the company has continued to prepare for a possible public listing. Reports earlier indicated executives were weighing a 2027 timeline rather than accepting a lower valuation in 2026. Prediction markets currently assign roughly a 20 percent probability to an IPO by the end of 2026. Those prices move with every new data point. A massive, multi-year infrastructure commitment of this size is one more data point that OpenAI is building for a long horizon rather than a quick exit.
In my view, the combination of long-term power leases and residual-value guarantees from a major chip supplier is a signal that OpenAI expects sustained high utilization of its models. You do not lock in 4.25 gigawatts for twenty years if you think demand might plateau. You do it when you believe the next generation of models will require even more compute than the current generation.
Implications for the Wider AI Supply Chain
Nvidia’s dual role as both exclusive compute provider and residual-value guarantor tightens the relationship between chip supplier and model developer. The $1.5 billion equity investment in SB Energy further embeds Nvidia in the physical layer of the stack. SoftBank’s presence as both investor and power-generation partner adds another layer of capital alignment.
Other AI companies will watch the structure closely. Some may seek similar residual-value arrangements with hardware suppliers. Others may prefer pure cloud contracts that keep the balance-sheet exposure lighter. The Ohio model is capital intensive and long duration. It suits a company that already has strong commercial traction and expects that traction to continue.
The power side of the equation is equally consequential. Building 10 gigawatts of new generation is a multi-year undertaking. Grid upgrades of the scale contemplated here require regulatory approvals, equipment lead times, and skilled labor. Any delay on the power side cascades into delayed AI capacity. That interdependence is why the project partners are emphasizing simultaneous progress on generation, transmission, and data-center construction.
What Could Still Go Wrong
Long-duration infrastructure projects carry execution risk. Construction costs can rise. Equipment deliveries can slip. Environmental reviews can extend timelines. Community opposition can surface even when local benefits are promised. The 2028 start date for initial capacity already builds in several years of work. Further slippage is possible.
Credit risk is another variable. OpenAI’s ability to service the lease depends on continued revenue growth and access to capital markets. Nvidia’s guarantees provide a backstop, but they are not unlimited. A severe downturn in AI spending or a major technical setback for OpenAI’s models could test the residual-value assumptions.
Regulatory risk sits in the background as well. Large power loads and data-center clusters attract scrutiny on energy use, water consumption for cooling, and local environmental impact. The project’s framing as ratepayer-protected and community-benefit-oriented is an attempt to stay ahead of that scrutiny. Success is not guaranteed.
Finally there is technology risk. If a future generation of AI hardware becomes dramatically more efficient, the economic value of a multi-gigawatt campus built around today’s architecture could decline. Nvidia’s argument that the capacity is fungible and broadly adoptable is the counter to that risk. The hardware can serve other workloads. The site can host other tenants. The residual value holds if those claims prove durable.
How Crypto Markets Are Reading the Same Signal
The intersection with crypto mining is more than a footnote. Miners that already control power and land have spent the past two years exploring high-performance computing conversions. The same electrical infrastructure that supports proof-of-work can support training and inference clusters. Cooling systems, substations, and fiber connectivity transfer reasonably well.
When a company of OpenAI’s profile commits to 4.25 gigawatts of dedicated capacity, it validates the scarcity thesis that many mining investors have been advancing. Powered sites are not abundant. Grid interconnection queues are long. Any operator that can offer near-term available megawatts has a product that AI companies need. The Ohio deal does not create that scarcity, but it makes the scarcity more visible.
Some mining companies have already signed hosting or joint-venture agreements with AI and cloud customers. Others are still evaluating the capital required to convert. The residual-value and long-lease structure used in Ohio may eventually appear in modified form in mining-to-AI deals. The details will differ, yet the core idea—locking down power and space for multi-year AI workloads—is the same.
Looking Ahead to 2028 and Beyond
The first capacity is scheduled to come online in phases beginning in 2028. That leaves roughly two years of heavy construction, equipment installation, and grid work. Progress updates will become important markers. Any public disclosure of power-purchase agreements, generation project milestones, or early tenant commitments will move the narrative.
Nvidia has indicated that the full guarantee agreements will be filed as an exhibit to its Form 10-Q for the fiscal quarter that ended July 26, 2026. Those documents will give investors a clearer view of the precise triggers, caps, and termination conditions. Until then, the high-level description already paints a picture of carefully structured risk sharing.
I’ve found that the most durable infrastructure projects are the ones that align the interests of the technology user, the hardware supplier, the developer, and the local community. This deal attempts that alignment. OpenAI gets long-term capacity. Nvidia locks in a major customer and residual-value exposure it believes it can manage. SB Energy and SoftBank get a large development opportunity with equity partners. Pike County gets jobs, investment, and a community fund.
Whether the alignment holds through the full twenty years depends on execution and on the continued growth of AI demand. For now the commitment itself is one of the largest single AI infrastructure leases announced to date. It sets a new benchmark for scale and for the financial structures that make such scale possible.
The real test will arrive when the first servers light up and the power meters start spinning. Until then, the Ohio campus stands as a concrete expression of how far the industry is willing to go to secure the physical foundations of artificial intelligence. Land, power, and time. Those three constraints now sit at the center of the race. OpenAI and Nvidia just placed a very large, very long bet that they can manage all three.
