OpenAI has signed a 20-year lease for the PORTS-Pike campus in Ohio, according to The Decoder, which cites Wall Street Journal reporting on the deal. The lease is with SB Energy, a SoftBank subsidiary, and covers roughly 8 gigawatts of information technology capacity for OpenAI. The Journal puts the gross project scale at 10 gigawatts when cooling and supporting infrastructure are included. The reported structure shows how large AI infrastructure commitments are moving beyond conventional cloud contracts. The Decoder says Nvidia is not guaranteeing OpenAI’s rent. Instead, Nvidia is backing the residual value of completed facilities in the first construction phase, which covers 4.25 gigawatts of IT capacity. Under the arrangement described by The Decoder, if OpenAI were to leave, SB Energy would first need to look for another tenant and then try to sell the facilities. Nvidia would cover any remaining shortfall in value, capped at $105 billion. In exchange, Nvidia becomes the exclusive chip supplier for the first half of the Ohio site and is investing $1.5 billion in SB Energy, according to the report. The physical site is also unusual. The Decoder reports that the campus sits partly on a former U.S. Department of Energy uranium enrichment facility. Its power is tied to a 9.2-gigawatt gas plant owned by the U.S. government and financed by Japan as part of a trade agreement, according to the same report. Nvidia CEO Jensen Huang framed the bottleneck as “LPS” — land, power and shell — rather than chips alone, The Decoder reports. In that framing, the limiting resources for AI buildout are the site, electricity supply and finished buildings that can host compute. Nvidia’s role in the Ohio lease suggests the chipmaker is using its balance-sheet strength to help secure the infrastructure that would ultimately consume its GPUs. The scale described in the report is large even by current AI data center standards. Huang expects about 1.5 million GPUs per system generation, with $150 billion to $200 billion in revenue, according to The Decoder. Across all sites, he puts OpenAI’s commitments through 2030 at roughly 12 gigawatts of Nvidia compute. If Nvidia exercises an option on the remaining 3.75 gigawatts in Ohio, the package would grow to about 16 gigawatts and be worth roughly $600 billion, the report says. OpenAI says it pays only for finished capacity, according to The Decoder. The first 800 megawatts are slated to come online in 2028. That timing matters because the economic obligation, physical power delivery and compute deployment do not all appear at the same time. The broader accounting issue is becoming material for investors. The Decoder cites a Wall Street Journal analysis finding that nine tech companies, including Alphabet, Meta, Microsoft and Nvidia, hold around $3 trillion in mostly AI-related obligations that are not yet reflected on their balance sheets. Leases are recorded once payments begin, and purchase commitments when goods are delivered; leases that have not started total $1.2 trillion, four times the level a year earlier, according to the report. The same analysis says Alphabet’s purchase commitments rose from $332 billion to $811 billion in three months. The Decoder also reports that Alphabet and Amazon recently posted negative free cash flow, and that Morgan Stanley analysts warned investors may have difficulty assessing actual debt levels as these contracts grow. Who benefits: OpenAI would secure a large block of future compute capacity without paying until finished capacity is delivered, according to the report. Nvidia would deepen its position as OpenAI’s chip supplier while helping unlock the facilities needed to absorb more GPUs. Who's exposed: Nvidia takes residual-value exposure capped at $105 billion for the first phase, as described by The Decoder. Investors in large AI infrastructure buyers are also exposed to obligations that may not yet be fully visible in conventional balance-sheet metrics.