The headline number is approximately 8 GW-IT — and the unit matters
OpenAI says it has entered an agreement to secure approximately 8 gigawatts of IT capacity at the PORTS-Pike Technology Campus in Pike County, Ohio, working with SB Energy, NVIDIA and the U.S. Department of Energy. NVIDIA describes an initial 4.25 GW-IT commitment with an option covering another 3.75 GW-IT.
That is planned information-technology capacity, not eight gigawatts of completed data-center load operating today. The distinction matters because the project is a multiyear buildout whose infrastructure has to arrive in phases.
The first 800 MW shows how the buildout becomes a utility problem
OpenAI says the first 800 megawatts are expected to become available in 2028 largely through existing AEP infrastructure. It also says later development will require new power plants connected to the grid, including natural-gas generation, plus new transmission lines and associated infrastructure.
That sequence moves AI capacity planning upstream. The delivery path includes generation, transmission, substations, switchgear, transformers, interconnection studies and construction schedules before the compute equipment can become productive capacity.
Land, power and shell are becoming strategic AI inputs
NVIDIA explicitly frames the partnership around securing land, power and shell capacity. The company is providing credit support for the buildout and says OpenAI will be the customer while SB Energy builds, owns and operates the data center under a 20-year lease.
The commercial architecture is notable because it treats the physical site almost like another critical supply-chain layer. The AI factory is no longer only a semiconductor procurement problem; it requires long-duration commitments across real estate, energy and industrial construction.
The project says its grid costs will not be shifted to ratepayers
OpenAI says the project will pay its own energy and infrastructure costs and that SB Energy will pay the full cost of grid upgrades and new transmission lines required to serve the data center. That is a project commitment and should be evaluated against implementation as the buildout proceeds.
The claim also highlights the policy question surrounding very large data-center loads: who finances incremental generation and network reinforcement, and how utilities allocate those costs while maintaining reliability for existing customers.
AI electricity demand is no longer a side issue
The International Energy Agency projects global electricity generation serving data centers to rise from about 460 terawatt-hours in 2024 to more than 1,000 terawatt-hours in 2030 in its base case. The exact trajectory remains uncertain, but the direction reinforces the infrastructure problem exposed by individual gigawatt-scale campuses.
As compute demand grows, bottlenecks can migrate outward from chips into grid connection queues, generation capacity, transmission equipment, cooling, construction labor and permitting. Different regions will encounter different constraints first.
The RFDELTA takeaway
PORTS-Pike is a useful systems signal because it makes the physical dependency chain explicit. Frontier AI is converting model demand into commitments measured in gigawatts, years of construction and major grid works.
The next competitive advantage in AI may therefore be partly electrical: the ability to assemble chips, land, generation, transmission and site infrastructure quickly enough to turn planned compute into energized compute.
Watch the original Signal
The concise video version is designed for discovery; this page preserves the sourcing, caveats and deeper context.
Memorable path: https://rfdelta.com/053
Video transcript
OpenAI has reserved approximately eight gigawatts of IT capacity at an Ohio AI campus. That is not eight gigawatts of finished compute today. The buildout is planned over years. But the infrastructure scale is the signal. NVIDIA says it is securing land, power and shell capacity, while OpenAI says the first eight hundred megawatts are expected in 2028. Later phases require new generation, transmission lines and grid upgrades. OpenAI says the project will pay its own energy and infrastructure costs. AI demand is now forcing compute planning upstream into utilities, substations, generation, construction and permitting. The next AI bottleneck may not sit inside the server rack at all. It may be the ability to energize the site. Follow RFDELTA for what comes next.
Frequently asked questions
Is PORTS-Pike already consuming 8 gigawatts of electricity?
No. OpenAI describes an agreement for approximately 8 GW of IT capacity that is planned to be delivered over time. It says the first 800 MW are expected in 2028; later phases require additional generation and transmission infrastructure.
Who is building and operating the data center?
SB Energy is expected to build, own and operate the data center under a 20-year lease to OpenAI, while the campus is intended to exclusively host NVIDIA AI compute infrastructure.
Why is this a power-grid story rather than only an AI-chip story?
Because multi-gigawatt compute deployment requires generation, transmission, substations, interconnection, electrical equipment, site construction and permitting before servers can operate. Those physical systems can become binding constraints.
Primary sources
Continue exploring RFDELTA
RFDELTA Signals map the hidden systems, technology transitions and operational dependencies underneath fast-moving headlines.