Weekly Recap: Billions in Deals, Backlash in the Backyard

Today on Concrete Compute: this is the weekly recap, and the week's real tension isn't in Silicon Valley — it's in the fights breaking out in the towns hosting the machines. Before we get to the local pushback, the money side kept moving: a thirteen-billion-dollar cloud deal from a trading firm, a five-and-a-half-billion-dollar sweetener from SoftBank's data center arm, and a four-point-four-billion-dollar bet on power conversion hardware. And then there's the politics — California sending ratepayer protections to the governor's desk, Orange County floating a moratorium, Texas governor Greg Abbott touting the strictest rules in the country, and President Trump escalating his rhetorical war on anyone who says no to a data center. Both threads matter on their own, but together they tell you where this industry's pressure points actually sit right now — the capital keeps arriving faster than the permitting and power questions get answered. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It's Saturday, September 5, 2026 — this is the week in review. Let's get into it.

Let's start with the money, and the comparison before the raw figure: this week a quantitative trading firm signed a cloud contract sized like something you'd expect from a frontier AI lab, not a Wall Street shop. The number: Crusoe signed a five-year AI cloud deal with Jane Street Group reportedly worth around thirteen billion dollars — Jane Street now looks like Crusoe's most prominent customer, full stop. We flagged this earlier in the week, so the quick version: the figure comes from Bloomberg's anonymous sources, and neither company has confirmed it publicly, so keep that caveat attached — this isn't a filed contract number, it's a reported one. What it tells you is the theme holding all week — a non-AI-native company, a trading shop rather than a model lab, is now committing double-digit billions to compute. Why does a quant firm need that much silicon? Nobody's said on the record, and that's still an open question worth asking, especially since trading firms don't usually need training-scale clusters the way model labs do.

Sticking with the money question but flipping it to who pays: California lawmakers passed two bills this week, SB 886 and AB 2383, which would require the state's utility regulator to set up separate electricity tariffs and updated interconnection rules for large data centers — the goal being that their grid costs don't get quietly shifted onto everyone else's residential bill. We covered this as it moved through the legislature; the delta is that both bills have now cleared the legislature and sit on Governor Newsom's desk. He hasn't signaled whether he'll sign, veto, or let them become law without his signature, and that silence itself is worth noting given how closely this bill package has been watched. And even if he signs, the California Public Utilities Commission has until mid-2027 to actually finalize the tariff details, meaning the fight over how strict this gets is nowhere near over. This is exactly the ratepayer-protection story this show keeps tracking, and the real test won't come until those tariffs are actually written.

Here's a number that grew while nobody was watching: stock warrants worth three-point-six-billion dollars back in January were worth five-point-five-billion dollars by June — same warrants, same deal. The deal: SB Energy, SoftBank's data-center venture preparing to go public, gave OpenAI those warrants as an inducement to sign on as its anchor tenant. We touched on this one already this week; the update worth sitting with is just how far these tenant-anchoring structures go beyond the announced capex headlines — equity kickbacks, not just leases, are apparently how developers lock in a marquee customer now, and that's a structural detail the capex press releases never mention. One caveat that matters: these figures come from draft IPO documents reviewed by the Wall Street Journal, not final public filings, so the number could still move before any actual listing, and it's worth watching whether the eventual valuation confirms or deflates it.

Here's one that flew under the radar for us but not for the news cycle: President Trump escalated his rhetoric against data center opponents for the second time this week. Speaking Friday in the Oval Office, he said towns that don't approve data centers risk what he called poverty, crime and squalor, while towns that do approve them can, in his words, get rich and lower their taxes. The Hill posted the line straight on X: quote, Trump says data center opponents face poverty, crime and squalor, end quote. And @The_AI_Investor posted a longer verbatim version of the same remarks on X — Trump saying, quote, if you want to get rich as a state or as a town, and if you want to pay lower taxes, if you want to create tremendous wealth, if you want your house to be more valuable, and everything else, you're going to want a data center. If you want to go through poverty, crime, and squalor, I would say don't approve data centers, end quote. That's about as blunt as this argument gets. Now, it's worth stating the other side in its own terms, because opposition to data centers this year genuinely crosses party lines — Pennsylvania's Democratic governor Josh Shapiro has been tightening rules, and Michigan's Republican Senate nominee Mike Rogers has called for a moratorium. Experts cited alongside Trump's comments make a specific counterpoint: the long-term energy demands of these facilities could raise, not lower, home electricity bills, as utilities recoup the cost of new transmission and generation. So who's right? I'm not adjudicating that on air — but opposition showing up on both sides of the aisle heading into the midterms is itself the story, and it's the kind of split that makes this a genuinely hard one to predict district by district.

Here's one flying under the radar this week: down in Orange County, Florida, Vice Mayor Kelly Martinez Semrad is asking commissioners to consider a moratorium of at least one year on new data center applications. Her reasoning is straightforward — the county's land development code has gaps when it comes to large-load facilities, and there's a projected regional groundwater shortfall she wants addressed before more of these get approved. Now, under Florida law, a large-scale data center means one anticipating peak loads of at least fifty megawatts, so this isn't aimed at some small server closet, it's aimed at the gigawatt-class projects that actually strain a local grid and water system. What's important to say clearly: this is one commissioner's proposal, not yet a commission vote, so nothing is actually paused yet. But it's a good example of the question this show keeps asking about host communities — what does the community actually get, and does it have the water and power to support what's being asked of it? Orange County apparently doesn't know the answer yet, which might be the most honest thing a fast-growing county can say before it approves anything else. Whether the full commission actually adopts this when it comes up is still genuinely unresolved.

Now to Texas, where Governor Greg Abbott brought his own version of this fight to Lubbock. At a roundtable there, he said Texas now has the strictest AI data center requirements in the country, framing it as protecting rural communities — his words were that these facilities, quote, cannot build them in neighborhoods or rural communities that do not want them, end quote. The substance behind that claim: directives to the state's utility commission and grid operator requiring data centers to pay their own grid costs, report their water and power use, and reduce their impact on nearby communities. That's a Republican-led version of the same ratepayer-protection instinct driving California's bills this week — different party, same underlying question of who pays for the grid upgrade, which tells you this argument has stopped being a partisan one at the state level. But here's the caveat that matters: Abbott's strictest-in-the-country claim is his own characterization, and a chunk of what he's describing rests on voluntary commitments from companies like Skybox, Digital Realty, and Mara, not passed statute. Full codification has to wait for the twenty twenty-seven legislative session. So is this binding law or a set of promises with a governor's press conference attached? Right now it's genuinely both, and that gap between directive and statute is the real question heading into next year's session.

Now a supply-chain story that matters as much as any capex headline, and the comparison before the number: Flex just spent as much on a single power-conversion company as some hyperscalers spend announcing an entire new campus. The number: four-point-four-billion dollars, for a company called EPC Power, adding eight-hundred-volt DC and grid-forming technology — hardware that helps convert power between the grid and the chip — to Flex's data center infrastructure business. Here's why this one's not just a footnote: as rack densities keep climbing, the equipment sitting between the grid and the GPU is turning into a strategic bottleneck, and real money is now flowing into it, in a part of the supply chain that rarely makes headlines on its own. Flex plans to spin that whole infrastructure unit off as an independent public company in the first quarter of twenty twenty-seven. The deal itself isn't finalized — it's still subject to regulatory approval and expected to close in the fourth quarter of this year. Given Flex's recent string of power-infrastructure acquisitions, the open question is whether regulators raise any antitrust concerns before that close. If they don't, this quietly becomes one of the more consequential hardware bets of the year, because you can build all the gigawatts you want, but somebody still has to convert the electrons.

Last story of the week, and it's the one that ties the whole reliability question together: utilities are eyeing small modular nuclear reactors as an answer to grid reliability, but here's the twist — tech companies and defense customers are the ones actually first in line, not the utilities themselves. The deep-pocketed private buyers are the early adopters testing this technology, while utilities are watching from the sidelines to see whether the siting and cost-recovery obstacles — basically, who's allowed to build one where, and who pays for it if it goes over budget — can actually get solved before this scales to the broader grid. This adds texture to a steady drumbeat of nuclear headlines all week, and the honest framing here is exploratory: this is utilities watching closely, not a wave of orders, which is a meaningfully different story than the one the headlines sometimes imply. So the real question, and nobody's answered it yet in this reporting: if hyperscalers and defense customers prove the technology out first, does that de-risk it for ratepayer-funded utilities to follow, or does it mean regular electricity customers end up bearing the risk of a technology built and tested on someone else's dime? That's the fork in the road for small modular reactors right now, and it's genuinely unresolved.

That's the week — thirteen billion here, five-and-a-half billion there, and governors in three different states all arguing about who gets to say no to a data center. If this episode helped you make sense of the week, follow Concrete Compute wherever you listen, so today's episode isn't the last one you catch. 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!