OpenAI’s 8 GW Ohio Campus Begins With an 800 MW Reality Check

Home

Rows of server cabinets in an imgix data centre

In brief

OpenAI has joined an Ohio data-centre plan targeting about 8 GW of IT capacity. The first 800 MW is expected in 2028; permits, finance and power remain ahead.

Featured image: Server cabinets illustrate large computing facilities. Contextual photograph; not the PORTS-Pike site or completed OpenAI capacity. Photo: imgix / Unsplash. Unsplash licence.

An eight-gigawatt computing campus would be infrastructure on the scale of a power system, not a conventional property development. Yet its first useful test is much smaller: whether the initial phase connects on schedule.

OpenAI announced on 17 August 2026 that it had joined the PORTS-Pike project in Pike County, Ohio, with SB Energy, NVIDIA and the US Department of Energy. The parties describe an ambition for about 8 gigawatts of IT capacity through 2032. The first 800 MW is expected in 2028. These are forward-looking plans, not operating capacity. Read OpenAI’s announcement.

IT power is only one layer of the project

IT capacity is the electrical power available to servers and networking equipment. A campus also needs electricity for cooling, conversion losses, lighting and other systems. An 8 GW IT figure therefore does not represent the site’s complete grid demand, annual energy use or computing performance.

The first 800 MW is expected to rely largely on existing American Electric Power infrastructure. Later phases would require new power plants, including natural-gas generation, and new transmission. That sequence is the project’s physical reality check: the headline campus cannot arrive before its energy system. Power and phasing details.

Nor does a gigawatt describe intelligence produced. Hardware generation, utilisation, model architecture and operating software all affect the work obtained from a unit of power. Capacity is a constraint and a scale indicator, not a complete productivity measure.

Electricity substation infrastructure; representative photograph, not Calvert Cliffs.
An electricity substation illustrates the grid infrastructure required by data centres. Contextual photograph; not equipment at PORTS-Pike. Photo: American Public Power Association / Unsplash. Unsplash licence.

The commercial structure spreads ownership and risk

SB Energy is expected to own, build and operate the data-centre facilities under a 20-year lease. OpenAI says it will pay as completed capacity becomes available, and the compute infrastructure is intended to use NVIDIA systems exclusively. Commercial structure in the announcement.

NVIDIA plans a US$1.5 billion investment and credit support connected with an initial 4.25 GW of IT capacity. That commitment is subject to conditions including permits, environmental reviews, financing and construction milestones. A financing commitment should not be counted as delivered servers.

The structure separates the tenant, facility owner, chip supplier, land and energy relationships. It can allocate risk among specialised parties, but it also creates interfaces: a delay in power, finance or construction can hold up the entire computing schedule.

Water and public costs require measurable promises

The project says it plans closed-loop, air-cooled systems that recirculate water, with ongoing consumption expected to resemble office use. It also says SB Energy will fund required grid upgrades and transmission rather than shifting those costs to ratepayers. Both are commitments that can be tested as engineering and contracts become public. Community and resource commitments.

The development site is associated with the former Portsmouth Gaseous Diffusion Plant reservation, where federal environmental management work has addressed the legacy of uranium-enrichment operations. The data-centre project does not erase that separate remediation history. DOE background on the Portsmouth site.

The project website presents the campus as an economic redevelopment effort. Local job and investment forecasts are projections, so later reporting should distinguish temporary construction work, permanent roles and indirect estimates. PORTS-Pike project information.

The next milestone is not another round number

OpenAI projects 35,000 construction jobs during a six-year build-out and 2,500 operational jobs. Those estimates help show intended scale, but they do not yet show who is hired, for how long or at what stage. The same caution applies to the 8 GW target.

Useful evidence will include issued permits, financing close, generation and transmission approvals, construction progress, water data and energisation of the first phase. Annual public reporting promised by the project can make some of those claims auditable.

PORTS-Pike may become a vast AI campus. Today it is a conditional development programme whose credibility begins with 800 MW, not with treating an 8 GW ambition as an accomplished fact.

Join the discussion

Have a question or a different perspective? Share it below. Please keep comments respectful and relevant to the article.

Leave a Reply

Your email address will not be published. Required fields are marked *

FUTURETECHDOSE BRIEFING

Follow the technologies shaping what comes next.

Clear, source-led reporting across biotechnology, AI infrastructure, energy, robotics and emerging devices.

Latest reporting