On August 17, Nvidia filed a current report with the SEC disclosing that it had agreed to guarantee up to roughly $105 billion in lease obligations tied to an 8 gigawatt data center campus in Pike County, Ohio, with OpenAI signed on as the tenant under a twenty year lease. Three days later, CNBC reported on an internal memo from the National Republican Senatorial Committee warning that data centers had become a “sleeper issue” for the entire midterm cycle, and it named the Ohio Senate race specifically. Same week, same state, two documents pointing in opposite directions. Only one of them shows up in anybody’s cash flow model.
That gap is the most interesting thing happening in the AI trade right now. For three years the bear case has been about demand and depreciation schedules. The risk that actually arrived this month is duller and harder to hedge. It is zoning boards, rate cases, executive orders, and a midterm electorate that has decided data centers are why the power bill went up.
A Guarantee Is Not an Investment
Nvidia did not write a $105 billion check. It entered into residual value guarantees with SB Energy, the SoftBank owned developer building the campus at the former Portsmouth Gaseous Diffusion Plant site, covering roughly 4.25 gigawatts of IT load, with discretion to extend credit support to about 3.8 gigawatts more. Payments come due only if OpenAI defaults or becomes insolvent. SB Energy and SoftBank have separately committed to build generation supporting 10 gigawatts and to put at least $4.2 billion into regional grid infrastructure. The first 800 megawatts is expected online in 2028.
We have written before about what these financing structures imply for the durability of AI chip demand, and Fortune noted that the disclosed commitment landed roughly $145 billion below what had been previously reported, which kept that argument going. Set the circularity debate aside for a moment, because a simpler point survives regardless of where you land on it. A meaningful slice of the buildout now rests on structures that assume a specific facility gets built roughly on schedule, in a specific county, under a specific regulatory regime, for two decades.
Twenty year assumptions and two year election cycles do not mix well.
The Politics Moved Faster Than the Concrete
Pike County sits in Ohio, where Democratic candidate and former senator Sherrod Brown has made opposition to data centers a centerpiece of his campaign against Republican incumbent Jon Husted. The campus in question could cost more than $500 billion at full 10 gigawatt scale, according to earlier CNBC reporting, and 10 gigawatts is roughly equivalent to the annual power consumption of 8 million American households based on a CNBC analysis of Energy Information Administration data. That is not an abstraction to an Ohio voter opening a utility bill.
The NRSC memo, first reported by Axios, was blunt about the stakes. If voter perceptions do not change quickly, the committee wrote, the campaign against data centers will expand well beyond Ohio, and if Husted loses over the issue, “the industry will be blamed.” For context on the backdrop, Democrats currently lead the generic ballot by about 6.4 percentage points in the RealClearPolitics polling average.
Look around and the pattern is not partisan, which is precisely what makes it a durable risk rather than a headline. In Pennsylvania, Democratic Governor Josh Shapiro signed an executive order this week imposing hard conditions on data center development, including a requirement that developers pay for all new electricity generation, transmission, distribution, and related infrastructure their projects require, plus community benefit agreements, local hiring commitments, and strict water and environmental standards. Shapiro had previously courted these projects. His Republican opponent, state treasurer Stacy Garrity, is running to his right on the issue and calling for a pause.
In Texas, Governor Greg Abbott issued a directive in June requiring companies to fund electrical infrastructure upgrades, reuse their own water, and reduce the cost of electricity, and announced this week that a number of firms had agreed to the standards. His Democratic challenger, Gina Hinojosa, wants a moratorium until the legislature can write guardrails. In New York, Governor Kathy Hochul signed an executive order in July establishing what amounts to the first statewide yearlong pause on large data centers while her administration builds a regulatory framework. And in Florida, Representative Byron Donalds won the Republican gubernatorial primary this week after introducing federal legislation last month aimed at ensuring private developers rather than taxpayers bear data center investment costs.
Republican governors, Democratic governors, House Republicans, Senate Democrats. When an issue polls that well across the aisle, it stops being a news cycle and starts being a cost input.
This Is Not a Messaging Problem
The industry’s instinct has been public relations. Microsoft published a blog post in January laying out steps to be a good neighbor in host communities. OpenAI followed days later with its own pledge to pay its own way on energy, minimize water use, and invest in local jobs. President Trump, weighing in this week, allowed that data centers “can use a little public relations help” while arguing communities should welcome them for the jobs and tax revenue.
The underlying data suggests the problem runs deeper than communication. Pew Research Center reported on August 18 that 52 percent of Americans now say they are more concerned than excited about the growing use of AI in daily life, up from 37 percent in 2021. Among adults under 30, that figure is 55 percent, and only about 11 percent of that group say they are more excited than concerned. Roughly 71 percent of respondents believe AI will lead to fewer job opportunities, up from 64 percent in 2024. The survey was fielded in late June.
Anthropic chief executive Dario Amodei offered his own read in a post on X last weekend, framing the hostility as a broader “crisis of trust” in institutions rather than something specific to AI. Whether or not you accept that diagnosis, it points at the same conclusion an investor should reach. Sentiment this negative, this widespread, and this stable is not going to be reversed by a press release before November.
Where This Actually Shows Up in the Numbers
Not as a ban. Almost nobody is proposing to outlaw compute. The transmission channel is cost and schedule, which is subtler and more consequential.
Consider what Shapiro’s order actually does. It shifts the cost of new generation, transmission, and distribution from the ratepayer base onto the developer. That is a straightforward reallocation of billions of dollars of infrastructure spending from a socialized pool to a private one, and it changes the economics of every project still in planning in that state. Abbott’s Texas directive points the same direction from the opposite party. Moratoria like New York’s do something different but equally real. They add time, and time is the entire thesis for an asset financed on a twenty year lease with a first delivery date in 2028. Every quarter of interconnection delay, every community benefit agreement negotiated line by line, every rate case contested by a consumer advocate compounds into a later revenue start and a lower return.
The second order effect is worth sitting with. If developers absorb grid costs that utilities previously spread across all customers, the announced capital numbers for these projects were understated, or the returns embedded in them were overstated, or both. Nobody has restated anything yet. But a promise made in a January blog post to “pay our own way” reads differently once a governor turns it into a permit condition with penalties attached.
There is a well funded counterweight, to be fair. AI aligned political action committees, including the Marc Andreessen backed Leading the Future, are spending real money, and Donalds received significant support from that direction. The fight has two financed sides. But an industry that has to buy its way to neutral on a bipartisan consumer issue is carrying a cost it did not carry a year ago.
For investors watching this, the read is less about picking a winner than about updating a variable. Anything in the AI infrastructure complex, the hyperscalers, the accelerator vendors, the independent power producers, the electrical equipment names, has been priced off a buildout schedule. That schedule now has a political input with a measurable calendar date attached. The Ohio Senate race in November is arguably the cleanest single test of how much that input is worth. Whether other governors copy Hochul or Shapiro in the months after, and whether the industry’s voluntary pledges to pay its own way become binding permit conditions, will tell you more about 2028 revenue starts than another quarter of accelerator bookings will.
The chips were never the hard part. Neither, it turns out, were the substations.
_________________________________________________________________________________________________________
This article is written for educational and informational purposes only and does not constitute financial or legal advice. The views and analytical frameworks presented draw on publicly available information and reported commentary from industry participants. Readers are encouraged to consult primary sources and form their own informed views on these complex topics.
