OpenAI’s PORTS-Pike announcement reads like a local jobs story at first. Pike County, Ohio, gets a new technology campus on and around the former Portsmouth Gaseous Diffusion Plant. OpenAI says it will secure about 8 gigawatts of IT capacity from SB Energy, with Nvidia as the exclusive compute supplier. The first 800 megawatts are expected in 2028, and the full buildout runs toward 2032. The public numbers are large enough to make the press release feel abstract. Thirty-five thousand construction jobs. Two thousand five hundred operating jobs. At least 10 gigawatts of new generation. At least $4.2 billion of regional grid infrastructure. An $80 million community benefits fund once OpenAI’s $40 million is added to SB Energy’s earlier pledge.

Strip away the civic language and the deal is still more interesting than a normal data-center lease. SB Energy will build, own, and operate the campus under a 20-year lease to OpenAI. Nvidia is putting $1.5 billion into SB Energy and providing credit support for land, power, and shell capacity tied to the initial 4.25 IT-gigawatts, with an option on the remaining 3.75. OpenAI says it will begin paying only as completed capacity becomes available, funded through revenue, business growth, and investor capital. That sentence does a lot of work. It says the AI lab wants the option on future compute without carrying the whole construction project directly today.

That is the right steelman. Frontier AI demand is uncertain in level, but not in direction. A lab that waits for ordinary utility planning cycles will lose the ability to train and serve models when demand arrives. A chip supplier that only ships accelerators can still lose sales if customers do not have land, transmission, substations, cooling, and permits ready. A developer with power access can turn a former industrial site into a toll road for compute. Each participant is buying protection against the same failure mode: demand shows up, and the physical system cannot deliver.

The economics are messier than the industrial-policy frame admits. An 8 IT-gigawatt campus is not merely a big building. It is a claim on generation, grid capacity, gas plants, transmission lines, transformers, local permitting, water management, labor, and twenty years of utilization. The useful accounting question is not whether the project creates jobs. It probably does. The question is where the downside sits if AI revenue arrives slower than the infrastructure schedule.

The public structure pushes part of that risk away from OpenAI’s face balance sheet. OpenAI is the customer, not the owner. SB Energy owns the asset and carries the development role. Nvidia supports the land, power, and shell buildout and locks in the compute stack. SoftBank sits behind SB Energy. AEP Ohio and public agencies shape the grid path. Local residents get the promised fund, jobs, and tax base, but also the land, water, and grid consequences of a huge new load.

This is how the AI capex cycle is starting to look across the industry. The Wall Street Journal reported this week that nine major technology companies have about $3 trillion of AI-related commitments that do not yet appear as ordinary balance-sheet debt, including future leases and purchase commitments for data centers, chips, energy, and equipment. Exact classifications matter, and many obligations are disclosed in filings rather than hidden. Still, the direction is clear. Investors who look only at quarterly capex miss a growing book of future claims on cash flow.

That does not make the buildout fake. A lease is not fraud because it is a lease. A purchase commitment is not debt just because it binds future spending. Project finance exists because infrastructure has different cash-flow timing from software. The problem is simpler. AI companies are selling a software-growth story while buying an infrastructure-duration problem. The revenue curve is supposed to compound quickly. The cost curve is being locked in through contracts measured in decades.

Nvidia’s role is the cleanest signal. Earlier in the cycle, Nvidia had the scarce asset: accelerators. Customers lined up, margins expanded, and the bottleneck looked like chip supply. PORTS-Pike shows the next constraint. The valuable bundle is land, power, shell, and a guaranteed path to deploy Nvidia systems at gigawatt scale. Nvidia is no longer only selling into the capex cycle. It is helping finance and de-risk the capacity that will buy its own systems.

That circularity can be rational. If Nvidia’s support gets a campus built sooner, OpenAI gets capacity, SB Energy gets financeable demand, and Nvidia gets a larger future equipment market. It can also make demand harder to read from the outside. Some orders reflect end-user AI revenue today. Some reflect expected demand tomorrow. Some reflect a vendor-backed ecosystem trying to make tomorrow’s demand investable today. These are different risks, even if they all show up as capacity plans.

The community-benefit language deserves the same treatment. OpenAI says it will pay project-specific energy and infrastructure costs, and Nvidia’s release says the AEP Ohio partnership is designed to protect ratepayers. Good. That is the minimum structure a project of this size needs. But ratepayer protection is not a slogan; it is a tariff design, a cost-allocation fight, and a long series of regulatory decisions. If 10 gigawatts of new generation and $4.2 billion of grid infrastructure are built for one class of load, someone must stand behind the fixed costs. Contracts can assign the risk. They cannot make it disappear.

The local labor math is similar. Thirty-five thousand construction jobs through 2032 is a real headline. Two thousand five hundred permanent operating roles is a smaller, more durable number. The political exchange is obvious. A region that once hosted nuclear-industrial infrastructure gets a new industrial role; in return, it accepts the footprint of the AI economy. That may be a good trade for Pike County. It should still be priced as a trade, not wrapped in the language of inevitability.

My prior is that OpenAI is right to secure capacity early. Frontier demand is easier to lose from undersupply than from holding too many options. The cost of being short compute during a product takeoff is severe. But the distribution is wide. Better model efficiency, slower enterprise adoption, regulation, local opposition, higher power costs, or a cheaper rival architecture could all reduce the value of a 20-year capacity claim. The bullish case does not remove that tail. It funds it.

PORTS-Pike is useful because it makes the AI race less metaphorical. Frontier competition now includes model quality, GPU allocation, credit support, lease duration, grid tariffs, gas generation, substations, and the right to convert Ohio electricity into tokens from 2028 onward. Balance-sheet exposure will not answer every question, but it will show which parts of the story have become fixed obligations.