NVIDIA to help secure power and land for OpenAI’s AI factory at Portsmouth, Ohio

NVIDIA said it is taking a more direct role in securing the physical “AI factory” infrastructure needed to run advanced computing at scale, announcing a long-term capacity arrangement tied to a major site in Portsmouth, Ohio. The company frames the effort as addressing a growing bottleneck in deploying artificial intelligence systems: not just chips or software, but the availability of large power and long-lived data-center sites where compute can be installed and upgraded over many years.

Under the plan described in an NVIDIA blog post, the company is partnering with SB Energy to secure “LPS capacity” at the PORTS-Pike Technology Campus. NVIDIA said it would do so exclusively for NVIDIA AI factories, with OpenAI named as the tenant that will build and operate the AI factory at the site.

NVIDIA said the initial deployment at PORTS-Pike is expected to provide 4.25 gigawatts of AI-factory capacity. It added that each generation of NVIDIA AI factory systems deployed at the campus could represent approximately 1.5 million NVIDIA GPUs, and that this scale could correspond to roughly $150 billion to $200 billion in NVIDIA revenue per generation, with multiple upgrade cycles over the life of the site.

The company also said it may choose to extend the arrangement beyond the initial 4.25 gigawatts to secure additional capacity of 3.75 gigawatts. Separately, NVIDIA stated that OpenAI has committed to substantial deployments of NVIDIA AI infrastructure through 2030, estimating those existing and planned commitments at roughly 12 gigawatts of NVIDIA compute, with an opportunity to expand to approximately 16 gigawatts if NVIDIA extends the PORTS-Pike arrangement.

NVIDIA’s blog describes the strategy as selective, focusing on “exceptional” sites where durable demand and long-term financing can support multiple generations of computing systems. It argued that, while many of its customers can secure power and land capacity independently, some fast-growing frontier AI labs may be constrained less by algorithms or customer demand and more by the ability to obtain decade-scale infrastructure commitments.

The company said OpenAI will build and operate a “world-class AI factory” at PORTS-Pike using NVIDIA’s DSX AI factory platform. NVIDIA described DSX as a full-stack offering that includes GPUs, CPUs, networking, and infrastructure software, positioning it as the technology foundation that can be upgraded as new generations of AI hardware become available.

In NVIDIA’s account, the economics hinge on separating the long-lived infrastructure commitment from the faster-evolving compute stack. It said the LPS commitment is meant to secure a long-duration AI-factory site, while the NVIDIA compute installed within it can be upgraded repeatedly over time. It characterized this as enabling improved production and better economics with each new generation of systems.

NVIDIA said its support is not the same as paying for the entire project. It described its role as supporting defined portions of lease and power payments and including a specified residual-value commitment, rather than covering the full cost of the site or all tenant obligations. It also said the guarantee would become effective in phases as data centers are placed in service between 2028 and 2030, and that NVIDIA’s remaining exposure would decline as OpenAI makes lease payments and capacity comes online.

What remains unclear from the announcement is the detailed structure and value of NVIDIA’s obligations, including the exact scope of the residual-value commitment and how the company calculates its maximum exposure under the arrangement. NVIDIA did not provide a full financial breakdown in the text, and it did not specify contractual triggers, termination terms, or whether additional capacity beyond the initial 4.25 gigawatts would depend on performance milestones or power availability.

Looking ahead, NVIDIA and OpenAI’s ability to move from commitment to deployed capacity will be the key developments to watch. The company said the site is designed for multiple generations and, it added, that NVIDIA compute is “versatile” and could be redeployed to another qualified tenant if needed. Still, the timeline for power delivery, data-center commissioning, and how quickly upgraded system generations are installed will likely determine whether the planned 2028 to 2030 rollout matches the scale NVIDIA described.