What makes the old Ohio uranium site suitable for an 8 GW AI campus?
The former Cold War-era uranium enrichment plant in Ohio was selected as the location for a data center that OpenAI will lease for 20 years. The site is planned to start at 4.25 GW of capacity and scale up to 8 GW, which would make it the largest single AI data center in operation.
What stands out is how the project pairs this existing industrial footprint with fresh energy commitments. SoftBank and its subsidiary SB Energy will invest more than $4 billion in local energy infrastructure and construct at least 10 GW of new generation capacity. Nvidia has committed $105 billion in financing support for the overall effort and will supply the chips that run the facility exclusively.
This combination addresses the core constraint for large AI workloads: reliable access to massive amounts of power. An 8 GW campus requires dedicated generation and transmission upgrades that most sites cannot accommodate quickly. By anchoring the build at a former enrichment plant, the partners can layer new power assets onto an already permitted industrial location while meeting the timeline for a 20-year lease.
The result is a direct path to the power scale needed for frontier AI training and inference. OpenAI has also pledged $40 million toward local priorities as part of the arrangement. The Ohio site therefore functions as both a physical platform and a regulatory starting point for the energy buildout required to reach 8 GW.
How does Nvidia's $105 billion financing commitment actually flow in this deal?
Nvidia is supplying more than $105 billion to cover land, power infrastructure, and the basic building shell for an 8-gigawatt data center in Ohio. The money does not go straight to OpenAI. Instead, it supports SB Energy, which will own and construct the facility. Nvidia is also making a separate $1.5 billion equity investment directly into SB Energy.
OpenAI's role is limited to signing a long-term lease and making regular payments once the site is ready. This structure keeps the ownership and construction risk with SB Energy while Nvidia provides the large upfront capital commitment. The $105 billion figure covers the scale needed for an initial 4.25 gigawatts of capacity, with room to expand by another 3.75 gigawatts later.
The arrangement marks a change in how suppliers back their biggest customers. AI developers generate strong revenue but still require external capital to fund the power and real estate that support new clusters. By guaranteeing funds at this size, Nvidia reduces the barrier for OpenAI to secure the site while tying its own returns to lease payments that flow through SB Energy.
One clear lesson is that hardware leaders are now acting as project financiers, not just component sellers. This shifts capital planning away from pure equipment purchases toward long-term infrastructure deals that tie hardware supply directly to power and building costs.
Why must the entire facility run only on Nvidia chips?
Nvidia has committed up to $105 billion in credit support for the PORTS-Pike site in Ohio. The company is also investing $1.5 billion directly in SB Energy, the firm that will own and build the campus. In return, the facility is being secured specifically to host Nvidia compute hardware, with plans to house more than one million of its AI chips across an eventual 8 gigawatts of capacity.
This arrangement stems from Nvidia's supply chain approach. The company gains long-term visibility into customer demand from OpenAI and uses that insight to lock in physical infrastructure ahead of time. By tying financing to the site itself, Nvidia ensures the power, land, and shell are configured for its own systems rather than left open to other hardware vendors. The first 800 megawatts are scheduled to come online in 2028 under a 20-year lease where OpenAI is the main tenant.
The practical result is that design decisions, power delivery, and networking will align around Nvidia's architecture from the start. OpenAI gains access to massive scale without bearing the full financing burden, while Nvidia protects its position by controlling where its chips will operate at this size. The partial coverage of lease and power costs further shows how the deal focuses on securing dedicated capacity instead of subsidizing the entire project.
The core lesson is that hardware leaders now treat data center sites as critical inputs worth securing years in advance when demand forecasts are clear.
Who is building the 10 GW of new power generation and who pays?
SB Energy is handling construction and operations at the former uranium site in Pike County, Ohio. The company will develop the full campus under a 20-year lease that makes OpenAI the primary tenant. OpenAI holds a direct stake in SB Energy, and its CEO Sam Altman invested in the firm years earlier. This setup keeps day-to-day responsibility with SB Energy while tying the project closely to OpenAI's long-term needs.
Nvidia supplies the financing and the hardware. A securities filing shows the chip maker extending up to $105 billion in credit to support an initial 4.25 gigawatts of capacity, with an option to add another 3.75 gigawatts later. The first phase of 800 megawatts is scheduled to reach operation in 2028, and the rest will follow in stages. Nvidia also delivers the Nvidia AI chips that will fill the buildings.
The financial split is straightforward. OpenAI pays through the lease for power, space, and compute access. SB Energy covers building and running the site. Nvidia covers the upfront capital and equipment costs in exchange for steady demand from its largest customer. The arrangement spreads risk across three parties instead of loading it onto any single balance sheet.
The practical result is that OpenAI gains access to up to 8 gigawatts of dedicated capacity without owning the land or the power plants outright.
What does the 20-year lease term reveal about OpenAI's long-term capacity plans?
OpenAI has agreed to a 20-year lease on the former uranium plant site in Ohio. The company will take space only as new capacity comes online, so payments start when the facility actually delivers usable power and chips rather than on a fixed schedule.
That length stands out because most technology leases run far shorter. It signals that OpenAI expects demand for high-end compute to remain steady or grow for decades, not just through the next product cycle. The deal also ties into plans for 10 gigawatts of supporting power and at least $4.2 billion in regional grid upgrades, numbers that require years to build and amortize.
The long commitment changes how the company thinks about infrastructure. Instead of renting capacity in short bursts and moving when prices shift, OpenAI is locking in a large, steady supply that can scale in phases. This reduces the risk of sudden shortages but also creates fixed costs that must be covered even if training runs slow down.
Nvidia’s pledge of up to $105 billion in credit and chips further anchors the arrangement, showing suppliers are also betting on sustained need. OpenAI’s own statements frame compute as the new scarce resource that every major model depends on.
The clearest takeaway is that OpenAI now treats large-scale compute the same way utilities treat power plants: as a multi-decade asset that must be secured well in advance.
How will SoftBank's ownership change day-to-day operations for OpenAI?
SoftBank owns the Ohio site through a subsidiary that is building the data center, while OpenAI holds a 20-year lease on the completed capacity. The lease structure means OpenAI pays nothing until sections come online starting in 2028, and it takes capacity only as it becomes available. This setup keeps the physical construction and ownership with SoftBank, but OpenAI controls which workloads run on the eight gigawatts once they are ready.
The arrangement limits immediate shifts in daily work. OpenAI continues to manage its own models and training jobs without needing to handle site construction, power contracts, or maintenance of the buildings. Nvidia's financing covers 4.25 gigawatts of the total, which reduces some of the capital burden but does not alter who decides how the chips are used once they are installed.
Over time the main operational effect appears in scale rather than process. Teams at OpenAI gain access to far more computing power than before, yet they still schedule jobs, monitor performance, and allocate resources through the same internal systems. The gradual rollout means engineers will add capacity in stages instead of all at once, which spreads out integration work but does not require new ownership rules or staffing changes on OpenAI's side.
The clear lesson is that ownership stays with the builder while the tenant focuses on usage.
What local investments is OpenAI making and what do they actually cover?
OpenAI has committed up to 105 billion dollars toward a data center project in Pike County, Ohio. The company signed a 20-year lease for the site, which forms part of a larger effort to secure computing capacity for its models. This sum represents one of the biggest direct financial pledges from an AI lab toward a single location.
The money does not go toward building the facility itself. Instead, it supports lease payments and power costs at the site. Nvidia is supplying the financing to back those obligations, using the funds to cover the ongoing expenses that OpenAI has agreed to meet. The total project cost could reach 500 billion dollars, so the OpenAI portion focuses narrowly on sustaining operations once the center is running rather than funding construction.
This arrangement shows how AI companies are relying on chipmakers to bridge gaps in their own ability to carry long-term debt. Labs like OpenAI grow their computing needs faster than traditional balance sheets allow, which forces new financing structures involving hardware suppliers. For local infrastructure in Ohio, the result is a steady revenue stream tied to power and lease agreements that lasts two decades.
The key point is that OpenAI’s investment secures access to capacity without owning the physical plant or covering every capital cost upfront.
Why are AI labs now relying on chip makers for infrastructure financing?
OpenAI's new data center lease in Ohio includes a $105 billion guarantee from Nvidia to cover lease and power payments to SB Energy. Nvidia will also supply every chip at the site under a 20-year agreement. This arrangement shows how AI companies now need chip makers to help close the financing for massive builds.
The structure works because Nvidia brings both the hardware and the long-term commitment that lenders require. The "land, power and shell" model lets the site accept new Nvidia systems over time without new financing rounds. Jensen Huang has stated that OpenAI remains responsible for the payments, while Nvidia's scale and visibility make the project bankable for the owner.
This shift matters because building at former nuclear sites, like the Ohio location or the planned Kentucky project at Paducah, demands enormous upfront capital for power and construction. Traditional financing often falls short when demand forecasts stretch decades ahead. Chip makers can step in because their revenue depends on these facilities running at full capacity with their latest hardware.
The result is faster project timelines and more upgrade flexibility for operators. AI labs gain access to sites they could not fund alone, while chip suppliers secure guaranteed demand for multiple product cycles. The pattern is likely to repeat as power-hungry campuses move to other legacy industrial locations.
What happens to capacity if the project expands from 4.25 GW to full scale?
The campus under discussion is a planned 10 gigawatt facility. Moving from an intermediate level of 4.25 gigawatts to the full target would more than double the available power for AI workloads. The first phase alone is set at roughly 800 megawatts and is scheduled to come online in 2028, which means any larger build out would stretch across multiple additional stages after that date.
Power capacity at that scale changes the economics in a direct way. The completed campus carries an estimated price tag of at least 500 billion dollars based on current chip, labor, and electricity costs. OpenAI would take on payment obligations through a 20 year lease once the site is running, committing to at least tens of billions of dollars across the full term. That arrangement shifts the risk profile from construction to long term operational spend.
The practical result is a single site that could support far larger model training runs and inference fleets than current clusters allow. Because the land is federal and the project already carries Nvidia financial interest, further expansion would likely depend on continued coordination among those parties and steady access to transmission upgrades. The core constraint remains the same: each additional gigawatt still requires matching generation, cooling, and networking infrastructure before it can be used.
In short, reaching the full 10 gigawatt level turns the site into a multi hundred billion dollar, multi decade commitment that sets the upper bound on how much AI compute OpenAI can run from one location.
How might similar supplier-backed deals reshape future data-center projects?
OpenAI's push for a 10-gigawatt campus in Ohio, with costs reaching at least $500 billion, shows how one company is trying to secure its own power and space instead of renting from big cloud providers. The talks include the possibility that Nvidia could guarantee the lease payments, which would give lenders more confidence to fund such a large build.
This approach changes the usual order of events. Normally a data center operator lines up power and land first, then finds tenants. When a chip supplier steps in to back the lease, the project can move faster because the financial risk shifts. OpenAI would still owe tens of billions over 20 years for the site alone, plus another $350 billion for the chips needed to fill it. A guarantee from Nvidia could make banks more willing to lend against those future payments.
The same pattern could appear in other large projects. Suppliers of power equipment, transformers, or cooling systems might offer similar credit support to lock in long-term demand. That would let developers start construction on sites like the former uranium plant without waiting for every piece of financing to close. Over time the result would be fewer delays and more direct control by AI companies over their infrastructure, rather than depending on cloud contracts that can change terms or capacity limits.
The main lesson is that hardware makers may become active partners in funding the buildings that run their own products.