Nvidia building GPUs is old news. Nvidia securing 20-year land leases and guaranteeing power capacity for AI factories is something else entirely. Jensen Huang announced a partnership with SB Energy to lock in land, power, and shell capacity at the PORTS-Pike Technology Campus in Portsmouth, Ohio, with OpenAI as the anchor tenant. The initial deployment targets 4.25 gigawatts of AI factory capacity. At roughly 1.5 million GPUs per generation cycle, that translates to an estimated $150 to $200 billion in Nvidia revenue per deployment cycle, across a site designed for multiple upgrade cycles over two decades.
Why Nvidia is moving into infrastructure
The logic here is worth unpacking. Nvidia has long relied on cloud providers like AWS, Azure, and Google Cloud to handle the physical infrastructure layer. Those companies have balance sheets, construction teams, and decades of experience signing long-term power contracts. But frontier AI labs are different. Companies like OpenAI are growing faster than their credit profiles can support. They need massive compute capacity now, but may not qualify for the investment-grade financing required to secure large-scale data center infrastructure independently.
So Nvidia is stepping in. Not to build or operate data centers directly, but to backstop the infrastructure commitments that make sites like PORTS-Pike viable. Nvidia’s guarantee covers defined portions of lease and power payments, plus a residual-value commitment. It does not cover the full cost of the site. And as OpenAI makes payments and capacity comes online between 2028 and 2030, Nvidia’s exposure shrinks.
The OpenAI commitment and what it means at scale
This is not a one-off deal. OpenAI has committed to substantial Nvidia compute deployments through 2030, covering approximately 12 gigawatts of capacity, with potential expansion to 16 gigawatts if Nvidia extends the PORTS-Pike arrangement to include the remaining 3.75 gigawatts. At those levels, the total opportunity represents roughly $600 billion in Nvidia compute revenue through the end of the decade. That number is large enough to matter even for a company with Nvidia’s current scale.
The fungibility argument and competitive context
Nvidia’s defense against the obvious risk question, what if OpenAI stops using the site, rests on CUDA. The argument is that Nvidia compute is fungible because CUDA gives developers a common platform across hardware generations. If one tenant exits, another can step in. Cloud providers, enterprises, sovereign AI programs, and startups all run on the same stack. This is a real competitive moat that AMD and Intel have not yet matched at ecosystem depth.
Still, this is a meaningful strategic shift. Nvidia is now tying capital and balance sheet exposure to specific physical sites. That is a different risk profile than selling chips. It puts Nvidia in closer competition with the infrastructure strategies of hyperscalers, while also deepening dependency with a single customer at a massive scale.
- 4.25 GW initial capacity at PORTS-Pike, expandable to 8 GW
- ~1.5 million Nvidia GPUs per generation deployment
- 20-year site commitment supporting multiple upgrade cycles
- OpenAI as tenant, with up to $600B in compute commitments through 2030
For developers and founders evaluating the AI infrastructure market, the takeaway is straightforward. Nvidia is not just a chip company anymore. It is positioning itself as the full-stack platform for AI production, from silicon to the physical ground the servers sit on. That is either a very strong moat or a very concentrated bet, probably both.
