Today on Concrete Compute: Vertiv just spent one point four five billion dollars in cash — with a ceiling near two point six billion — to buy its way around the utility. Is that the smart move, or a sign the grid itself is losing the race? Before that, in the headlines: a Louisiana datacenter campus lining up a two point eight eight gigawatt gas plant of its own, New York's stunning reversal on nuclear power, AWS laying new cable across the Pacific, and a Georgia column raising alarms about secrecy in datacenter tax deals. It's a power-and-politics kind of day, and honestly most days on this beat are lately — the throughline across all four is the same one you'll hear in our main story: everybody's racing to solve for power before the grid can. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It's Thursday, September 3rd, 2026. Let's get into it.

Let's start in Louisiana, where a company called ONE Nuclear Energy signed a binding letter of intent — that's a firm commitment to lock down the land, though not yet a construction contract — for site control on Project Cayman: a two point eight eight gigawatt gas-fired power plant plus a seven hundred megawatt battery system, sitting next to a data center campus near the RiverPlex MegaPark in Ascension Parish. Now, I want to flag the caveat myself: this is an LOI stage deal, not steel in the ground, and the company hasn't disclosed who the datacenter tenant actually is, or given a construction timeline, which matters because binding-on-paper and permitted-and-financed are two very different milestones. But the pattern is the story — ONE Nuclear's own strategy, per the release, is to build gas now and offer a path to small modular nuclear reactors later, serving the same industrial corridor that's also landing SpaceX's Starbase project and Hyundai's steel mill. This is the exact same bottleneck driving today's main story: developers building their own power rather than waiting on the queue, and betting that a public information campaign with local parishes can turn a paper commitment into an actual plant before some rival site beats them to the switch.

Now, sticking with power politics — New York's governor Kathy Hochul is calling for the biggest buildout of new nuclear reactors in modern U.S. history, five gigawatts, which would make New York the number three state in the country for atomic generating capacity. Here's what makes that stunning: this is the same state that shut down its largest carbon-free power source, the Indian Point nuclear station, back in 2021. Hochul told Canary Media that expanding nuclear capacity is part of her plan to keep the lights on and attract, in her words, twenty-first century economic development. Not everyone's on board — the group NY Renews says the state hasn't maximized wind and solar, and some state lawmakers have called for a nuclear moratorium, which tells you this isn't a partisan split so much as a fight over sequencing and priorities within the same broadly pro-clean-energy coalition. Worth remembering the host's standing nuclear rule here: a policy target is not a power plant, and five gigawatts is a decade-plus build if it happens at all. What's genuinely new is the reversal itself — a governor who inherited a state moving away from nuclear now staking her energy strategy on the opposite bet, with the datacenter load growth as the backdrop nobody's pretending isn't there.

Amazon Web Services is laying new cable across the bottom of the Pacific Ocean — a four hundred twenty terabit-per-second subsea line called Sta'O'Nuk, using twenty fiber pairs, running from Japan to a new landing station in Ocean Shores, Washington, described as the first new cable landing in that state in more than twenty-five years. AWS says it'll support things like distributed large language model training. It's not switching on until 2029, so file this as a construction commitment, not new bandwidth you can use tomorrow — three years is a long runway for a project this physically involved, laying armored cable at depths up to fifteen hundred meters in some stretches. Jimmy Yu of Dell'Oro Group made the point that hyperscalers were already the biggest buyers of subsea capacity before AI showed up — this is just AI adding fuel to a fire that was already burning. And it's a reminder that time-to-power isn't the only clock running; time-to-bandwidth is right behind it, because international cable capacity has limited value, as the reporting on this cable puts it, if operators can't move that traffic efficiently from the coast into the actual datacenters and inland network infrastructure that need it.

Now, one more story worth your attention before we get to the main event. A Georgia Recorder opinion column — and I want to be clear that's what this is, one columnist's take, not a straight news report with the state or industry weighing in — argues that non-disclosure agreements and generous tax breaks are letting a wave of unpopular datacenter proposals move forward before residents can fully evaluate them. That's exactly the question this show keeps coming back to: what does the host community actually get? Jobs that outlast construction, a real tax base, protected power bills? The column doesn't name the specific NDAs or incentive deals, so I can't tell you which projects it means, and that vagueness is itself a limitation worth naming rather than papering over — but the underlying tension, secrecy paired with public subsidy, is one we'll keep watching as more of these deals surface. And to be fair to the other side of that argument, which this particular column doesn't quote directly: state and industry officials generally frame those same NDAs as standard practice for protecting competitive site-selection information, not as concealment from the public — the column simply doesn't put that rebuttal on the record here, which is itself worth noting.

Our main story today: Vertiv just bet one point four five billion dollars that the fastest way to get AI datacenters power isn't through the utility at all — it's around it. Vertiv, the company that's long sold datacenters their power and cooling gear, announced it's acquiring UtilityInnovation Group, a firm that designs microgrids, onsite generation, and what's called behind-the-meter power architecture — meaning electricity generated and managed right at the site, before it ever touches the public grid. The deal is one point four five billion in cash at closing, with up to another one point one five billion contingent on UIG hitting earnings targets over the next one and two years. Add those together and you get the headline number everyone's quoting, two point six billion — but I want to be precise here, because that's a ceiling, not a signed check, and Vertiv's own release says the multiple should look significantly lower if the full earnout actually pays out, which is its own quiet admission about how uncertain that upper figure really is. Vertiv also hasn't closed the deal yet; they're targeting the fourth quarter of 2026, pending regulatory approval. And notice what UIG actually sells: designs that engage a customer before any equipment is even chosen, meaning Vertiv wants influence over the power blueprint at the earliest possible moment, when those decisions shape everything downstream — grid-connected sites, what the company calls bridge-to-grid setups, and fully islanded sites running entirely on their own generation.

So why does Vertiv want this? Vertiv's CEO, Gio Albertazzi, put it plainly: competitive advantage for AI datacenter operators increasingly depends on how fast they can move from site selection to first token — the moment a chip actually starts producing usable output. UIG's founder, Sidney Hinton, described his company's mission as solving power challenges through flexible, technology-agnostic architectures — meaning designs that don't lock a customer into one generation source. Put those two statements together and the strategic logic is obvious: Vertiv wants to own the customer relationship from the utility interconnect all the way down to the rack, whether that site ends up grid-connected, bridge-to-grid, or fully islanded and running on its own generation. My read: this is Vertiv formally admitting that chip supply stopped being the bottleneck a while ago, and grid interconnection queues are the new critical path — so badly that even the equipment vendor wants a piece of the workaround, rather than just selling the gear that sits downstream of whoever solves it. The multiple tells you something too — thirteen times UIG's expected 2027 earnings for the upfront piece alone, a price a power-equipment giant doesn't pay for a nice-to-have, and Vertiv itself says the deal should add to earnings in the very first year after it closes, which is a fast payback clock for an acquisition this size. And this fits a pattern we just heard in Louisiana: developers and now vendors building their own generation because they don't trust the queue to move fast enough. The open question nobody's answered yet is how much of UIG's actual project pipeline is grid-connected work versus fully islanded microgrids — because those are very different bets on how broken the interconnection system really is, and Vertiv hasn't broken that split out for anyone yet. Time for the Hype Check. I'm giving this one a 6. The deal itself is real, it's signed, and the strategic logic is sound — but the two point six billion figure is doing a lot of unearned work in every headline I've seen, when one point one five billion of it depends on earnings targets nobody's hit yet, and the whole thing hasn't even closed.

So the next time you see a datacenter operator announce dedicated gas plants or microgrids of their own, don't read it as a backup plan — read it as an admission about how slow the front door has gotten, whether that operator is a developer, a nuclear startup, or now an equipment vendor like Vertiv trying to get ahead of the same queue. We'll be watching whether UIG's pipeline skews toward the grid or away from it entirely. If you've been enjoying how this show connects the dots between deals like today's, go ahead and follow Concrete Compute wherever you're listening, so today's episode finds you automatically next time out. 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!