Nvidia secures power sites for AI laboratories and dedicates Ohio park to OpenAI computing
Listen to this article
Read by Anchor
The scaling equation for foundation AI models is shifting from algorithmic design and chip procurement to securing land, power, and core structures, known as LPS resources. Against this backdrop, Nvidia announced a partnership with SB Energy to secure capacity for these resources at the Portside Tech Park in Portsmouth, Ohio, to host its computing infrastructure, with OpenAI serving as the operating tenant.
Details of the announcement indicate that leading AI laboratories face a structural gap compared to major cloud providers and investment-grade corporations, as these labs are expanding at a rate that outpaces their balance sheets' ability to underwrite multi-decade infrastructure contracts.The growth of AI laboratories has become constrained by compute availability and hosting sites rather than algorithms or user demand, prompting Nvidia to step in as an infrastructure backstop to sustain compute production and growing revenues.
The project involves OpenAI deploying an AI factory built on Nvidia's integrated DSX platform, including GPUs, CPUs, networking systems, and infrastructure software. The initial planned deployment capacity is 4.25 gigawatts, with each generation of deployed computing systems representing approximately 1.5 million GPUs, generating between 150 billion and 200 billion dollars in revenue for Nvidia per generation. The 20-year site commitment accommodates multiple technology refresh cycles to support output and economic efficiency, with an option to expand the arrangement by an additional 3.75 gigawatts to cover the full capacity of the campus.
More broadly, OpenAI has existing and planned commitments for Nvidia infrastructure deployments totaling nearly 12 gigawatts through 2030, which could reach 16 gigawatts if the Ohio expansion option is exercised, representing a commercial opportunity of roughly 600 billion dollars in Nvidia compute over that period. Regarding backstops, Nvidia is providing infrastructure support for approximately 4 gigawatts over 20 years, limited to specific portions of lease and power payments and residual value commitments, without assuming the full cost of the site. This backstop will phase in between 2028 and 2030 as the data centers come online and the tenant meets its obligations, gradually reducing Nvidia's remaining exposure.
Nvidia maintains that these arrangements do not constitute circular financing, as OpenAI pays the lease obligations as the actual tenant, while Nvidia leverages its scale and long-term visibility to secure critical capacity.Nvidia compute is fungible and reallocable thanks to the unified CUDA software ecosystem, allowing site capacity to be redirected to other tenants across its network of cloud providers and enterprises if the facility is underutilized in the future. While most customers continue to secure their own physical infrastructure directly, the move establishes a strategic foundation for supercomputing capacity across future hardware generations.