NVIDIA announced that it is backing infrastructure capacity at the PORTS-Pike Technology Campus in Portsmouth, Ohio, to host NVIDIA compute for OpenAI. The arrangement involves SB Energy and shifts the AI infrastructure discussion from chips alone to a more physical layer: land, power, buildings and available electrical capacity.

What happened

In an official August 17 post titled “Securing the Infrastructure of Intelligence,” NVIDIA said it is partnering with SB Energy to secure “land, power and shell” capacity at PORTS-Pike. The company says the site will host NVIDIA compute and that OpenAI will be the tenant.

OpenAI’s official feed listed an August 17 entry titled “OpenAI joins PORTS-Pike project,” confirming the company’s connection to the project, although direct access to the OpenAI page was blocked during this verification. The Decoder, in independent coverage published August 17, reported that the arrangement is tied to a 20-year lease with SB Energy and that NVIDIA is backing part of the residual value of the project’s first phase.

The project’s scale

NVIDIA’s central figure is 4.25 gigawatts of initial AI factory capacity at PORTS-Pike. The company also said each generation of AI factory systems deployed there could represent roughly 1.5 million NVIDIA GPUs and about $150 billion to $200 billion in NVIDIA revenue. NVIDIA said it may also extend the arrangement to secure the remaining 3.75 gigawatts of the campus.

That language should be read carefully: NVIDIA did not say that all of that revenue has already been realized or that the full campus is already operational. The guarantee would become effective in phases as data centers are placed in service between 2028 and 2030. NVIDIA also said its support is limited to defined portions of lease and power payments, plus a specified residual-value commitment, not the full site cost or all tenant obligations.

Why it matters

The news shows a strategic shift in AI infrastructure. Much of the debate has focused on GPUs, models and software. NVIDIA is arguing that the bottleneck for frontier labs may also be the ability to secure long-term sites with enough power and buildings to sustain multiple generations of compute.

For OpenAI, the announcement points to a major expansion of training and inference capacity. For NVIDIA, it opens a model in which the company does not only sell chips, but also helps secure the physical environment where those chips will be deployed. That structure may strengthen future demand for its systems, while also drawing more attention to financial commitments, energy availability and infrastructure concentration.

What is confirmed and what remains open

NVIDIA confirms that PORTS-Pike will host NVIDIA compute for OpenAI, that SB Energy is involved, that the expected initial phase is 4.25 gigawatts and that the project uses LPS as a critical resource for AI factories. OpenAI also published an official entry about joining the project, visible in its RSS feed.

What still requires follow-up includes full contractual documents, the actual construction timeline, the capacity ultimately placed in service, energy costs, permits, local impact and how the associated financial commitments will be accounted for. For now, this is best understood as a large-scale AI infrastructure move, not capacity already available to end users.

Sources: NVIDIA Blog, OpenAI and The Decoder.

Written by Nova Rivera — Product and automation perspective.