Today on Concrete Compute: SoftBank's data-center venture just told potential IPO investors it's handing OpenAI a multi-billion-dollar warrant package to keep it as a tenant — so who's really propping up whom here? And in Maryland, more than half the counties in the state have now hit pause or outright banned new data centers, with the industry warning that if it keeps going, AI hits a wall. Before that, in the headlines: a two-week power outage in Gary, Indiana is putting a very human face on what you lose when a federal solar resilience program gets canceled. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It's Monday, August 31, 2026. Let's get into it — because today's two lead stories are basically the same fight, told from opposite sides of the table.

Now, let's start in Gary, Indiana, where Canary Media reports residents went without power for as long as two weeks after violent storms and tornadoes tore through the region this month. NIPSCO, the utility serving northwest Indiana, called it the worst outage in company history, and Canary Media reports it knocked out power to somewhere around three hundred seventy to three hundred seventy-five thousand customers at its peak. Here's the part that stings: Gary was set to get about thirty-three million dollars from the federal Solar for All program — rooftop solar, batteries, and a resiliency hub where people could've charged phones and kept medicine cold during exactly this kind of outage. EPA Administrator Lee Zeldin canceled that program, part of the broader Greenhouse Gas Reduction Fund, following the federal reconciliation bill, and nearly two dozen states have since sued the EPA over it. Now, Canary Media is careful to frame this as a preparedness story, not a proven cause — those resilience hubs were still in planning, not built, when the storm hit. But you don't need an engineering study to feel what two weeks without power means for people who are underinsured, uninsured, or running medical equipment on batteries that just gave out.

Our main story today has a name, and it's this: the SoftBank subsidiary building giant AI data centers for OpenAI is telling its own IPO investors that its business is, quote, "substantially dependent" on the very company it's paying to stick around. The Wall Street Journal reports, citing draft IPO documents, that SB Energy — SoftBank's data-center venture — issued warrants to OpenAI now valued around five and a half billion dollars, up from an initial three point six billion dollars back in January, all to lock OpenAI in as an anchor tenant ahead of an IPO the Journal reports could raise somewhere between five and seven billion dollars as soon as next month. Now, warrants — for anyone who doesn't spend their days reading termsheets — are basically the right to buy stock later at a set price. Handing your biggest customer warrants worth five and a half billion dollars is, in plain English, paying them to keep renting from you, in a currency that only pays off if the IPO actually works. And this isn't the only money moving between these two companies. The Wall Street Journal's reporting lays out a web of financial entanglement: SoftBank Group and OpenAI each invested five hundred million dollars into SB Energy back in January, alongside a one point two gigawatt lease in Texas. Nvidia, separately, has put in one point five billion dollars and guaranteed up to one hundred five billion dollars to help OpenAI lease a data center in Ohio. And SB Energy has committed to buying at least fifty million dollars of OpenAI's own software by twenty twenty-eight. Every one of those numbers comes from the Journal's reporting on documents tied to this IPO process. So here's the question I keep coming back to: when the landlord invests in the tenant, and the tenant invests in the landlord, and the chipmaker guarantees the tenant's lease payments, and the landlord buys the tenant's software — who's actually taking the risk here? My read: SB Energy needs OpenAI's name on the lease to make its IPO story work, and OpenAI needs SB Energy's capacity to keep its compute promises. Vendor financing between a supplier and a real, paying customer is normal industrial practice on its own. But the prospectus reportedly saying SB Energy's business is "substantially dependent" on OpenAI is the tell — that's not a hedge, that's the whole thesis, printed for regulators to see. And the numbers behind that dependency aren't comforting. One report cited in today's coverage puts SB Energy's net loss at roughly three point two billion dollars for the first half of twenty twenty-six, versus about two hundred fifty million dollars a year earlier. That's a loss that grew sharply, and it's happening while the company tries to convince public market investors it's worth north of fifty billion dollars. That question is still open. But today's disclosure adds weight to it: the more financially tangled OpenAI gets with its own landlord, the more it matters who's actually steering site and power strategy on OpenAI's side. Now, a caveat that matters here: this is a single scoop from the Wall Street Journal, citing draft and unfiled documents — not a finished, audited public offering. The five-to-seven-billion IPO raise and the valuation could both shift once SB Energy actually files. And here's a question worth sitting with: what happens to a nine hundred megawatt site with no named tenant if this IPO wobbles? The Journal reports a fourth SB Energy facility, in Scurry County, Texas, still has no named customer — so not all the capacity behind this valuation is actually leased to anyone. Some of it is steel and permits, waiting on a tenant. Here's what would make this credible to me: an actual signed, audited filing with SB Energy's real customer concentration disclosed, not a leaked draft, and a tenant found for Scurry County before the roadshow. Until then, treat the five-point-five-billion number as what SoftBank wants investors to believe OpenAI is worth to them — not as a verified fact about OpenAI's staying power.

Our second lead today has a name too: the Maryland test case, because right now Maryland is the clearest live experiment in the country for what a community actually gets, or loses, when it says no to a data center. Here's where things stand: the Baltimore Sun reports that as Maryland counties keep banning or pausing data-center development, AI and energy executives are warning the state risks losing jobs, tax revenue, and processing power. The moratorium wave isn't small — according to Governing, roughly thirteen counties, more than half the state, have now adopted some version of a ban or pause, including Baltimore City, Harford, Montgomery, Frederick, Calvert, and Prince George's County, which adopted the state's longest pause, at two years. The proof point everyone's pointing to is Amazon. In early August, Amazon formally withdrew its application for a roughly five hundred megawatt, two point four six million square foot data-center campus at Calvert Cliffs — the clearest sign yet that the backlash can actually kill a project, not just slow it down. Now, let's hear both sides in their own words, because this is a real fight with real stakes on both ends. The industry's argument, as the Baltimore Sun frames it, is straightforward: block enough projects and Maryland loses jobs, tax base, and the processing capacity increasingly underpinning the broader economy. The Maryland Tech Council goes further — Bethesda Magazine reports the group says local concerns are, quote, "misunderstood or vastly overstated," and that rushed policy risks, quote, "foreclosing on a generational economic opportunity." On the other side, residents and county officials aren't arguing against jobs for the sake of it — they're pointing to water use, pollution, and rising energy costs as reasons to slow down or stop specific projects. Those are real, local costs that show up on a water bill or a power bill before any tax revenue shows up in a county budget. So who's right? Honestly, nobody has the receipts yet to say for certain, and that's the most important sentence in this whole segment. Maryland's own state-commissioned study on the industry's impact on air, water, the Chesapeake Bay, and the grid still hasn't landed. Until it does, neither side's numbers are locked in by any actual regulation: Maryland's Public Service Commission is still drafting data-center rules, and the transmission and capacity reforms sitting at FERC remain pending. Everyone in this fight is arguing from projections, not from adopted policy. And there's a question hanging over all of this that nobody's answered yet: do counties that block projects actually recapture that tax base and those jobs somewhere else, or do they just lose them outright? The reporting today doesn't say, and until somebody tracks where these projects land instead, that's a real gap in the story, not a rhetorical one. Here's my read, and I'll own it as mine: the industry's warning about an AI wall is worth taking seriously — you don't want a state locking in a two-year pause and finding out the jobs and tax base just went to Virginia instead. But I also don't think residents pushing back on water and power costs are the villains here, and framing it that way isn't fair either. The companies spending the most on this buildout — and Amazon spends about as much as anyone — can afford to bring counties a real, binding deal before the fight starts: capped water use, rate protection for residential customers, permanent jobs written into the agreement instead of promised at a podium. That's my standard, not the state's, and it applies the same way whether the company is popular or not. Maryland didn't get that kind of deal at Calvert Cliffs, and now it doesn't have the project either.

Now, step back for a second, because these two stories are the same fight wearing different clothes. In Texas and Ohio, capital is racing to lock in gigawatt-scale AI tenants with warrants, guarantees, and reciprocal software deals — money chasing certainty. In Maryland, counties are racing the other direction, locking the door before the money even shows up. So what do these two stories actually have in common? Both sides are trying to control the same variable: who eats the risk if the promised value doesn't show up on schedule. SB Energy is betting its IPO on OpenAI's staying power; Calvert County bet its tax base on Amazon staying put, and Amazon left anyway. Time for the Hype Check. On this pairing, I'm putting it at a five. The SoftBank-OpenAI numbers are real disclosures in a real IPO process, and Maryland's moratorium wave is a real, documented policy shift — that's substance. But both stories are still running on projections nobody's tested: SB Energy's valuation depends on an IPO that hasn't priced, and Maryland's costs and benefits depend on a study that hasn't published yet. Ask me again once both of those actually land.

If you're the type who reads IPO filings for fun, or the type who just wants to know why the power bill might change, this show's built for both of you — follow Concrete Compute wherever you're listening, so today's episode finds you automatically tomorrow too. Maryland's state impact study still hasn't landed, and we'll bring you what it actually says about power, water, and the grid the moment it does. 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!