NVIDIA Guarantees OpenAI's Ohio AI Factory: The PORTS-Pike LPS Deal

NVIDIA is now guaranteeing land, power and shell payments so OpenAI can build a data center too large for OpenAI's own balance sheet to finance alone. On August 17, 2026 at 6:10 PM, NVIDIA CEO Jensen Huang posted "Securing the Infrastructure of Intelligence" on X — a long-form announcement that had crossed 1.6 million views by the time of writing. It names the site (PORTS-Pike, Ohio), the tenant (OpenAI), and, for the first time in NVIDIA's own words, answers the "is this circular financing" question that has hung over every NVIDIA-OpenAI headline since explainx.ai first covered the reported $250B Ohio backstop in July.

This is fundamentally an infrastructure-financing story. But the mechanism underneath it — NVIDIA backstopping the land, power and shell costs that have become the real bottleneck on frontier compute — is the kind of thing that changes what builders can access and pay for model capacity, not just what shows up on NVIDIA's balance sheet. That's the angle worth reading past the headline number.

TL;DR — What People Are Asking

What NVIDIA actually announced

NVIDIA's own framing, in Jensen Huang's words: "AI factories require a full stack of critical resources: advanced chips, packaging, memory, and networking – as well as land, power and shell." NVIDIA has spent years using its scale to lock up chip supply chain capacity. The new move is applying that same playbook to LPS — land, power, and shell — for frontier labs that can't secure it on their own credit.

The specific deal: NVIDIA is partnering with SB Energy to secure LPS capacity at the PORTS-Pike Technology Campus in Portsmouth, Ohio, with OpenAI as the tenant. OpenAI will build and operate the AI factory itself, using NVIDIA's full-stack DSX AI factory platform — GPUs, CPUs, networking, and infrastructure software.

This is very likely the same project explainx.ai wrote up in July as a reported ~$250B financing backstop for a southern Ohio ~10GW SB Energy campus — same developer, same tenant, same region (the Portsmouth Site referenced in that reporting is PORTS-Pike). What changed is that this is now an on-the-record NVIDIA statement with concrete gigawatt and dollar figures, rather than anonymously sourced reporting.

Why frontier labs specifically need this — and hyperscalers don't

NVIDIA draws a clear line between two kinds of customers in its own post:

"The world's largest cloud service providers and investment-grade enterprises have balance sheets, infrastructure expertise, and long-term contracts to secure LPS independently. […] This model will continue to represent most of NVIDIA's business."

"Frontier AI labs have extraordinary demand for training and inference compute, but many are growing faster than their balance sheets and long-term credit profiles can support. […] Their growth is increasingly constrained not by algorithms or customer demand, but by the availability of compute."

That's the structural claim worth sitting with: the bottleneck on frontier AI capacity is no longer model quality or GPU output — it's land, power, and the credit rating needed to finance decades-long infrastructure leases. OpenAI has fast-growing revenue but not the 20-year investment-grade paper that utilities and lenders want before they'll finance an 8-gigawatt campus. NVIDIA's guarantee substitutes its own balance sheet for that missing credit history.

The numbers: what's guaranteed, for how long

NVIDIA's Q&A section is unusually direct about scope, which is worth quoting rather than paraphrasing:

"NVIDIA is supporting the LPS infrastructure at PORTS-Pike for approximately 4 gigawatts over a 20-year term, securing a site on which NVIDIA compute will be exclusively deployed. Our support is limited to defined portions of lease and power payments, along with a specified residual-value commitment — not the full cost of the site or all of the tenant's obligations."

The most important structural detail is that the site outlives any single generation of chips. NVIDIA's own framing: "the LPS commitment secures a long-lived AI factory site, while the NVIDIA compute inside can be upgraded repeatedly." Twenty years of land and power backing multiple GPU refresh cycles is a very different bet than a 20-year commitment to today's silicon.

For a sense of how "gigawatt" numbers get inflated or clarified across different AI infrastructure deals, see explainx.ai's hyperscaler nuclear deals explainer, which applies the same stage-by-stage skepticism to power announcements generally — is a number an operating megawatt, a development target, or a partnership at zero purchased capacity? PORTS-Pike's 4.25GW is described as an "initial deployment," which is a real commitment, but the extension to 8GW and the 12-16GW OpenAI total are explicitly framed as options and opportunities, not signed contracts.

Is this circular financing? NVIDIA's answer, and the standing skepticism

This is the question every reader of this story is actually asking, and NVIDIA addressed it head-on rather than dodging it. Here's Jensen Huang's exact answer:

"No. OpenAI will pay the lease. NVIDIA uses its scale and long-term visibility to secure PORTS-Pike to host NVIDIA compute. This is the same discipline we apply to supply-chain management: we secure critical inputs when we have visibility into customer demand and when doing so enables long-term productive capacity."

The core of that defense: NVIDIA isn't handing OpenAI money to buy NVIDIA chips (the classic "vendor financing" circularity concern raised about Microsoft's OpenAI revenue and NVIDIA's own $500 billion compute-as-asset-class push). NVIDIA is instead backstopping a lease and power contract that OpenAI is still contractually obligated to pay, framed as the same supply-chain discipline NVIDIA already applies to chip inputs like wafers, packaging, and HBM.

NVIDIA also addressed what happens if OpenAI walks away: "NVIDIA compute is versatile, fungible and broadly adopted. The capacity can be resold to another qualified tenant across NVIDIA's global ecosystem of cloud service providers, enterprises, AI labs and startups." CUDA lock-in, in this framing, is what makes the guaranteed capacity redeployable rather than a stranded asset if OpenAI's demand doesn't materialize.

That's a real and coherent argument. It is also not the whole picture, and the standing skepticism voiced widely across AI-infrastructure commentary is worth stating plainly rather than dismissing: the guarantor and the chip vendor are the same company. NVIDIA's revenue is directly tied to how much compute OpenAI buys, and NVIDIA is now also underwriting the infrastructure OpenAI needs in order to keep buying it. Ordinary supplier financing (a chipmaker offering payment terms) doesn't usually extend into backstopping a customer's 20-year real-estate and power lease. Whether that distinction matters more than NVIDIA's "OpenAI still pays the lease" defense is a judgment call this piece isn't going to make for you — the mechanism and the skepticism both deserve to sit on the page together, and the same tension underlies the AI-companies-off-balance-sheet-debt discussion explainx.ai covered in July.

What this means for what you build or pay

This is where a financing story earns a spot on a builder-focused blog: does the mechanism actually change what you can access or afford?

Short version: this is a bet on loosening compute scarcity, not a guarantee of it, and not a guarantee it happens soon. A few things follow from NVIDIA's own numbers:

• Nothing changes for pricing or rate limits before 2028. NVIDIA is explicit that "the guarantee will become effective in phases as data centers are placed in service between 2028 and 2030." If you're budgeting for GPT-5.x or Claude API costs this year, this deal has zero effect on your invoice today.
• The scarcity story is the real target. NVIDIA's own diagnosis — labs are "increasingly constrained not by algorithms or customer demand, but by the availability of compute" — is the same constraint that shows up downstream as tight rate limits, waitlists for new model tiers, and token prices that don't fall as fast as model efficiency improves. If NVIDIA's LPS strategy actually gets more gigawatts online faster than frontier labs could manage alone, the multi-year effect should be more available inference capacity and, eventually, price competition among labs that finally have enough compute to compete on price rather than ration it.
• It's still a bet on OpenAI's growth, not a floor under the market. NVIDIA's guarantee shrinks as "OpenAI makes lease payments and capacity comes online" — meaning the whole structure depends on OpenAI's revenue and usage actually growing into the commitment. If demand disappoints, NVIDIA's fallback is reselling the capacity to another tenant, not eating the loss forever — which is a reasonable hedge for NVIDIA, but it means builders shouldn't read this as OpenAI-specific capacity locked in regardless of what happens to OpenAI's business.
• Watch 2028, not 2026. The realistic signal to track is whether the first PORTS-Pike phase actually reaches commissioning on schedule. Power and permitting, not chip supply, are historically the slower leg — see explainx.ai's data center backlash map for how often "announced" gigawatts slip against permitting and interconnection realities.

If you're choosing which frontier API to build on today based on price or availability, explainx.ai's token pricing explainer and model-selection energy math are more useful right now than this announcement — PORTS-Pike is a 2028+ supply story, not a today's-invoice story.

Honest limitations

• This is Jensen Huang's own X post, not an SEC filing or a signed, publicly disclosed contract — the numbers are NVIDIA's characterization of the deal, not independently audited figures.
• The 8GW ceiling at PORTS-Pike and the 16GW OpenAI ceiling are both described as options NVIDIA "may" exercise or extensions OpenAI has "an opportunity" to reach — not committed capacity.
• "Approximately $600 billion of NVIDIA compute through 2030" is a headline opportunity figure spanning multiple deals and years; it is not a single signed purchase order.
• NVIDIA's residual-value commitment and the specific lease/power payment portions it's guaranteeing aren't quantified in the public post — "defined portions," not a disclosed percentage.

Related on explainx.ai

• NVIDIA-OpenAI $250B Ohio backstop — the reported version of this deal, from July
• Microsoft's AI revenue is mostly OpenAI paying its own bill — the other circular-financing debate
• NVIDIA's $500B plan to make compute an asset class
• AI companies' off-balance-sheet debt, and the Enron comparison
• Every hyperscaler nuclear deal — decoded
• The AI data center backlash, mapped
• AI token pricing, explained
• Model-selection energy math

Sources and further reading

• Jensen Huang on X — "Securing the Infrastructure of Intelligence" (August 17, 2026)
• NVIDIA newsroom
• OpenAI

This analysis reflects NVIDIA's public statement as of August 17-18, 2026. Contract terms, guarantee percentages, and phased capacity figures may be clarified or revised as the deal is formalized in filings.