NVIDIA CEO Jensen Huang has introduced a new acronym to the AI infrastructure conversation, and it’s one that investors and rival chipmakers will likely be repeating for years. In a lengthy post on X, Huang laid out why land, power and shell, or LPS, has become the resource that will determine how fast AI factories can actually get built, and announced that NVIDIA is now securing LPS capacity directly for its customers, starting with a massive site in Ohio built for OpenAI.

The post arrives at a moment when the entire AI industry is running into a problem that has nothing to do with chip design or model architecture. Companies can order all the GPUs they want, but if there’s no physical site with the land, the electricity, and the building shell ready to house them, that compute has nowhere to go. Huang’s framing positions LPS as the new bottleneck sitting between NVIDIA’s silicon and the AI factories the world’s biggest labs need to keep scaling.

What Is LPS?

LPS stands for land, power and shell, essentially the three physical ingredients required before a single GPU can be installed anywhere. Land refers to the site itself. Power means the electricity supply and grid connection needed to run a facility that can draw gigawatts of energy. Shell refers to the actual building structure, the data center shell that houses racks, cooling systems and networking equipment.

Huang describes LPS as sitting alongside chips, packaging, memory and networking as one of the core layers required for what NVIDIA calls an AI factory, a term the company uses for large scale data centers built specifically to produce AI compute at industrial scale. According to Huang, most of NVIDIA’s customers, particularly large cloud providers and enterprises with strong balance sheets, have historically secured their own LPS independently. What’s changed is that frontier AI labs, whose compute demand is growing faster than their financing capacity, increasingly cannot do the same.

The PORTS-Pike Deal With OpenAI

The centerpiece of Huang’s announcement is a new partnership with SB Energy to secure LPS capacity at the PORTS-Pike Technology Campus in Portsmouth, Ohio. OpenAI will be the tenant at the site, building and operating an AI factory there using NVIDIA’s full stack DSX platform, covering GPUs, CPUs, networking and infrastructure software.

The initial deployment is expected to deliver 4.25 gigawatts of AI factory capacity, with Huang noting that each generation of NVIDIA systems installed at the site could represent close to 1.5 million GPUs and somewhere between $150 billion and $200 billion in NVIDIA revenue. Huang also said NVIDIA may extend its arrangement at the site to cover the remaining 3.75 gigawatts of capacity, which would bring the total site potential to around 8 gigawatts.

Zooming out, Huang tied this back to OpenAI’s broader NVIDIA commitments, pegging OpenAI’s existing and planned NVIDIA deployments at roughly 12 gigawatts through 2030, with room to expand to 16 gigawatts if the PORTS-Pike arrangement grows. At that scale, Huang put the total opportunity at around $600 billion of NVIDIA compute by the end of the decade.

Addressing The Circular Financing Question

A large chunk of Huang’s post is dedicated to what he calls “the important questions,” and it reads like a direct response to the criticism that’s followed NVIDIA’s various OpenAI arrangements over the past year, including the earlier $100 billion investment plan that reportedly stalled before this new structure emerged. Huang was explicit that NVIDIA’s guarantee is not the full cost of the site, describing it instead as covering “defined portions of lease and power payments” along with a residual value commitment, phased in as data centers come online between 2028 and 2030.

On the circular financing question specifically, Huang’s answer was short: OpenAI will pay the lease. He framed NVIDIA’s role as applying the same supply chain discipline it already uses to lock in chip manufacturing capacity, just extended now to the physical real estate layer underneath it.

He also addressed what happens if OpenAI eventually doesn’t need the site, arguing that NVIDIA compute is fungible enough to be resold to any other customer in its ecosystem, whether that’s a cloud provider, an enterprise, or another AI lab, largely because CUDA gives every buyer a common software layer to work with regardless of who originally leased the building.

Why LPS Matters Now

Huang’s post lands against a backdrop where power availability, not chip supply, has increasingly become the thing slowing down AI buildouts globally. Data center operators across the US have spent much of the past year competing for grid connections, land near substations, and utility commitments, often facing multi-year waits just to get power delivered to a site. By stepping into LPS directly, NVIDIA is positioning itself not just as the company that sells the compute, but as a partner that can help remove the physical constraints standing between a customer and deploying that compute at scale.

Huang was careful to note that this won’t become NVIDIA’s default approach across its entire customer base. He said the company will remain “strategic and disciplined,” reserving direct LPS involvement for a small number of exceptional sites capable of hosting multiple generations of NVIDIA systems over decades, while the bulk of its customers continue securing land, power and shell arrangements on their own, the way they always have.