Today on Concrete Compute: Nvidia was reportedly ready to guarantee a quarter of a trillion dollars to backstop OpenAI's giant Ohio data center — and in the last day or so, that number got cut by more than half. What happened, and who's actually left holding the risk? Before we get there, in the headlines: Chicago's mayor wants City Council to hit pause on new data centers, a former Missouri farm bureau chief is inviting developers onto his own land in open defiance of the moratorium wave sweeping the country, and the New York Times looks at communities everywhere demanding a real cut of the AI buildout's upside. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It's Sunday, August 16, 2026. Let's get into it.

Let's start in Chicago, where Mayor Brandon Johnson just signed an executive order kicking off a ninety-day rulemaking process on data centers, and he's asking the City Council to go further and pass a temporary moratorium on new projects altogether. The order creates an Interdepartmental Data Center Environment and Community Protection Task Force and immediately tightens permitting on diesel backup generators and noise rules, while the moratorium itself — which still needs Council approval and isn't law yet — sits pending. The mayor is citing water use, rising power costs, and air and noise complaints tied to the roughly thirty-nine data centers already running in the city. Not everyone in City Hall agrees. According to WBEZ Chicago's reporting, Alderman Gilbert Villegas pushed back publicly, arguing that data centers bring substantial property tax revenue and union jobs, and that the real fix isn't another executive order, it's renegotiating Chicago's franchise agreement with the utility, ComEd. Meanwhile Katia Balba, with the Alliance of the Southeast, reportedly called the order a good start — but said residents actually want the full moratorium, with teeth. This is the same jobs-versus-impact fight playing out from Denver to New York State, both of which took similar action earlier this year, and it comes back to a question I keep asking on this show: what does the host community actually keep, past the construction phase? Chicago hasn't answered that yet. Neither has anyone else.

In Atchison County, Missouri, a very different response to that same moratorium wave. Blake Hurst, who ran the Missouri Farm Bureau for a decade, wrote a Wall Street Journal op-ed inviting data center developers to come buy his land. His argument: the neighboring White Cloud Acres project — valued anywhere from roughly four billion to six point three billion dollars, depending on which outlet you read — hit a wall when Nodaway County commissioners were expected to pass a data center moratorium around July 21st, while they negotiate a development agreement with the project's developer, Scale Microgrids. But blocking it there, Hurst argues, just means it lands somewhere else, on a willing seller's land — like his. That's a real tension, because national polling has consistently found somewhere between sixty-three and seventy percent of people oppose a data center near them, and Data Center Watch counted at least seventy-five projects worth roughly one hundred thirty billion dollars blocked or delayed nationwide in the first quarter of this year alone, with active opposition groups more than doubling to eight hundred thirty-three across forty-nine states. Hurst's relocation theory rests on one example so far, not a broad pattern — so treat it as an argument, not a settled fact about how this wave actually plays out.

The New York Times ties a bow on both of those stories today, reporting that as data centers keep landing in more towns, the pressure is mounting on developers and utilities to actually share the financial upside — not just the construction-phase jobs and the standard tax abatement. The Times piece is light on a single dated deal or negotiation — this reads as a trend story, not a scoop about one town — but it's the same question Chicago and Atchison County are both wrestling with in their own ways: permanent jobs, tax base, and protection from higher electric bills, versus just being told to be grateful for the investment. And this is my standing line on this show: the companies spending the most on this buildout can afford to answer that question better than a press release. That's my bar, not a legal requirement, and it applies whether the company is popular this month or not.

Over on X, a thread worth putting in front of you. Daniel Roberts, co-CEO of the data center company IREN, posted a thread that picked up nearly two thousand likes arguing that every new AI data center creates more demand, not less — cheaper, better models unlocking use cases that didn't exist a year ago. In his words on X, quote, 'Demand grows at the speed of software, but supply cannot. Supply grows at the speed of the real world: power connections, transmission lines, concrete, steel,' end quote. Investors piled into the replies, several pushing back and sharpening the diagnosis: the constraint isn't chips anymore, it's power and transmission, which move on a timeline measured in years, not chip-fab quarters. Worth flagging — Roberts runs a company that profits directly from an insatiable-demand story, so weigh that alongside the argument. But it's the same power bottleneck we're about to run into in our main story, where the backstop for OpenAI's own gigawatts just got a lot more complicated.

Our main story today: Nvidia was ready to be the bank for OpenAI's biggest bet, and this week that guarantee just got a lot smaller. Here's the setup. OpenAI is trying to build a ten-gigawatt data center campus in Piketon, Ohio — a project on a scale that's genuinely hard to picture, with all of that power going to run AI models. The company actually building it is SB Energy, a unit of SoftBank. And here's the wrinkle that makes this whole story exist: OpenAI does not carry an investment-grade credit rating. Normally a project this size gets financed the boring, conventional way — banks, bonds, a return investors can underwrite against a rated borrower. OpenAI can't offer that, so instead, according to the Wall Street Journal, Nvidia has been negotiating to personally guarantee the financing — meaning if OpenAI ever defaulted on its lease or construction-debt payments, Nvidia would be the one covering the bill. The number the Journal first reported for that guarantee was staggering: roughly two hundred fifty billion dollars. That's not a chip order, that's not a data center budget — that's Nvidia agreeing to personally stand behind a quarter-trillion dollars of somebody else's construction debt. Now, here's today's delta: that guarantee has reportedly been cut to under one hundred twenty billion dollars, and narrowed so it only covers the project's first phase, not the whole ten gigawatts. The Journal's reporting says investors raised concerns about how much of that exposure was sitting on Nvidia's own balance sheet, and this renegotiation is apparently the response. There's a second piece here that makes the picture even stranger. According to The Information, Nvidia is discussing investing up to three billion dollars directly into SB Energy — the SoftBank unit actually building the Ohio campus — split between the deal-signing itself and SB Energy's planned initial public offering, which could come as soon as next month and might raise at least five billion dollars. Both of these pieces, the guarantee and the equity stake, are apparently part of the same roughly one hundred billion dollar credit-support negotiation. So Nvidia isn't just selling the chips that go inside this data center — it's discussing guaranteeing the debt that builds it, and putting equity into the company constructing it. Reuters and CNBC picked up the Journal's reporting today, and The Information's SB Energy numbers are being cited across outlets too — this is a story being pieced together from several newsrooms talking to people close to the negotiations, not from a single filing.

So here's my read on what all of this actually means. Start with the obvious tension: Nvidia's business depends on OpenAI buying enormous quantities of chips, and OpenAI's ability to keep buying enormous quantities of chips increasingly depends on Nvidia's willingness to guarantee OpenAI's own debt. That is a strange loop for a customer and a supplier to be standing inside together. If OpenAI's growth curve slows, or the model economics don't pencil out the way everyone in this deal is betting they will, Nvidia doesn't just lose a customer — it's on the hook for the construction debt behind ten gigawatts of infrastructure that might not have a tenant able to fill it. Now, before anyone treats these numbers as locked in, a serious caveat: none of this is signed. The Journal's own reporting says the guarantee deal could be signed as early as this weekend — could be, not is. Reuters and CNBC, covering The Information's three billion dollar SB Energy figure, both say explicitly they could not independently verify those numbers. There's no SEC filing, no FERC filing, no joint press release from Nvidia, OpenAI, or SoftBank confirming any of this. Every number in this story — the two-fifty, the one-twenty, the three billion, the hundred billion in credit support — is a negotiating position reported by people described as familiar with the matter, not a contracted, binding figure. A headline about a quarter-trillion-dollar guarantee and an actual signed instrument for under half that amount are two completely different stories, and this week we watched the first one start turning into something closer to the second. Here's what I'd want to see before calling any of this settled. First, does OpenAI ever actually sign the binding lease it's reportedly still negotiating on the full ten gigawatts, or does the project stay parked at phase one indefinitely? Second, does Nvidia's guarantee shrinking mean investors are getting nervous about AI demand specifically, or just about concentration risk sitting on one company's balance sheet — those are very different explanations for the same headline. And third, hold onto that IREN thread from earlier: if the real bottleneck in this industry has genuinely shifted from chip supply to power and transmission, as Daniel Roberts and the investors replying to him argued, then the risk in a ten-gigawatt Ohio campus was never really going to be whether OpenAI can pay its bills — it was always going to be whether this thing can actually get energized on schedule. A financing guarantee doesn't build a substation. Time for the Hype Check. I'm putting this one at a five. The signal here is real — Nvidia, OpenAI, and SoftBank are actually negotiating who eats the risk on an enormous project, and the guarantee getting renegotiated downward is a genuinely useful data point about where the market's appetite for this kind of exposure sits right now. But every headline number in this story traces back to anonymous people-familiar-with-the-matter reporting, nothing is signed, and the two outlets covering the biggest figure couldn't confirm it themselves. Call me cynical, but until somebody actually signs something, every number in this story — the two-fifty, the one-twenty, the three billion — is a negotiating position, not a fact.

The forcing event here: this financing guarantee is reportedly due to be signed as early as this weekend. If it lands with actual signatures instead of anonymous sourcing, that's the difference between a negotiating position and a real backstop for ten gigawatts in Ohio. If it slips again, that tells us something too. Follow Concrete Compute wherever you listen for whichever way that goes, and if today's episode changed how you think about who actually carries the risk in this buildout, pass it along to someone who'd want to know. This has been Concrete Compute, an AI-voiced podcast, created and built by a real human using today's cutting-edge technology. Nothing you heard on this show is financial advice. I'm Brian Lampert, and I'll catch you all tomorrow — take care!