Meta's $3 Billion Utah Bet Amid a Wall Street AI Slowdown Scare

Today on Concrete Compute: an AI lab CEO spent the weekend saying maybe the industry should slow down — and Wall Street spent Monday deciding what that actually means for the buildout. Before that, in the headlines: Meta is putting another three billion dollars into its Utah data center campus, and New Mexico's Department of Justice says it still isn't satisfied that ratepayers are protected from Blackstone's utility deal. Welcome back to Concrete Compute, your daily brief on the AI infrastructure buildout. It's Tuesday, September 15th, 2026. Let's get into it.

Let's start in Utah, where Meta just announced its fourth expansion at the Eagle Mountain data center campus outside Salt Lake City. This is not a new site — Meta broke ground here back in 2018, the campus went live in 2021, and it's added phase after phase since: a five-hundred-thousand-square-foot expansion in 2019, a nine-hundred-thousand-square-foot expansion in 2021, and two more buildings totaling two million square feet in 2022. Today the campus spans four and a half million square feet across seven buildings. With this latest phase, Meta says it's pushing its total Utah investment past three billion dollars, and the new buildings will use closed-loop liquid cooling paired with dry cooling — basically a cooling system that recirculates the same water instead of constantly drawing fresh supply, which matters a lot in a state that worries about drought. Now here's the jobs number, and I want to give you both halves of it like I always do: the company says this expansion supports more than twelve hundred construction workers at peak, and more than three hundred jobs at the data center once it's completed. That's a real, tangible commitment to a town that's watched this campus grow for eight years straight. Just remember these are company projections, not audited outcomes, and Meta hasn't said yet whether this newest phase runs on the regular grid or a dedicated power contract.

Now, sticking with money and who's actually on the hook for it — New Mexico's Department of Justice just threw a wrench into the Blackstone-TXNM saga. Quick refresher: TXNM Energy owns PNM, the state's largest electric utility, serving nearly five hundred fifty thousand customers, and Blackstone wants to buy the company. Back in July, the New Mexico Public Regulation Commission declared that stock sale — four hundred million dollars — void and of no effect, finding that TXNM never got the prior commission approval the transaction required. The commission's order also required TXNM to certify that costs from the transaction wouldn't be charged to New Mexico ratepayers. TXNM has now finished reversing that sale — it borrowed the four hundred million through a Wells Fargo term loan to pay back Blackstone's affiliate and canceled the shares. But the DOJ's new filing says TXNM's compliance report still doesn't answer the real question: could the replacement financing quietly raise costs for PNM ratepayers down the line? As @UtilityDive put it on X, regulators are seeking more information related to potential indirect costs to PNM ratepayers. The DOJ wants PNM to be ready to book a specific liability in its next rate case just in case. TXNM's side, through spokesperson Eric Chavez, maintains the rollback doesn't create costs for New Mexico customers. One commissioner, Greg Nibert, dissented from the original order entirely, arguing that unwinding the sale could hurt TXNM's finances — and by extension, ratepayers too. This one's still very much unresolved, and it's the same beneficiary-pays question that keeps surfacing anywhere private equity meets a regulated utility — hold that thought, because our main story today is about exactly who ends up paying when the AI buildout wobbles.

Our main story: what a one-day stock selloff actually tells us about who's holding the risk in the AI buildout. Here's what happened. Anthropic CEO Dario Amodei wrote over the weekend that AI labs should slow the pace of capability improvements — basically, that the race to build smarter models is moving faster than the safety work can keep up with. Sam Altman and Elon Musk both voiced agreement. And on Monday, Wall Street reacted like it meant something — AI-linked stocks sold off. Now, today's reporting doesn't hand me a clean breakdown of exactly which stocks moved and by how much, so I'm not going to pretend to a precision the numbers don't give me. What it does hand me is the real question sitting underneath all of it: do investors actually believe the physical buildout — the chips, the power contracts, the concrete and steel going into these campuses — is about to slow down? Or is this a safety-optics moment, a weekend of public soul-searching from three of the most powerful people in the industry that doesn't actually change anyone's spending plans? That's the tension the whole trading day hinges on, and it's worth sitting with, because those are two very different stories with two very different consequences. One version says the AI capex supercycle just hit its first real speed bump, and every gigawatt of planned data center capacity needs to be re-priced against a slower model-improvement curve. Both of those stories are plausible. Neither one is proven yet. And remember the story we just left — New Mexico regulators fighting over who eats the cost of a private equity deal at a regulated utility. That's the same question one level up: when AI's biggest backers place their bets, who's actually exposed if the bet goes bad? Is it the labs themselves, the chipmakers selling the picks and shovels, or the utilities and communities providing the power? Today's stock reaction doesn't answer that. But it's the right question to keep asking every time one of these headlines lands.

Now let's stress-test this, because a lot of ink got spilled on this over one weekend memo. And the same reporting flagged what the real signal to watch would be — not a CEO's weekend essay, but an actual canceled data center order or a scrapped construction deal. As of today, we don't have one of those. So where does that leave me? My read is that a stock move driven by a CEO's weekend memo is exactly the kind of thing that looks dramatic on a Monday and means nothing by Friday, unless it's followed by something concrete — an actual chip order pulled, an actual site paused, an actual interconnection agreement torn up. Sentiment isn't nothing — it moves capital, it moves headlines, and it can genuinely spook a market for a session or two. But sentiment isn't steel. It doesn't cancel a transformer order. It doesn't un-sign a power purchase agreement. Those things move on their own timeline, driven by permitting queues and equipment backlogs that don't care what any CEO wrote on a Saturday. Now, is this an actual inflection point for the buildout? I'm not there yet. One weekend memo and one day of selling is not the same evidence as an actual canceled interconnection agreement or a paused groundbreaking. Until I see that, I'm treating this as a sentiment story about safety optics, not a capex story about slowing steel. Time for the Hype Check. I'm giving this one a four out of ten on substance. The fact that AI-linked stocks sold off Monday is real — that's a genuine market event, and CEOs saying out loud that the industry should slow down is a genuine story. But the underlying claim that gets attached to it, that this signals an actual slowdown in the physical buildout, has zero confirmed evidence behind it yet. No developer has canceled a groundbreaking. This is a mood, not a moratorium — for now.

So here's what I'd watch for next: not another CEO essay about AI safety, but the boring paperwork — an actual canceled chip order, a construction contract that quietly disappears, the kind of filing nobody tweets about. That's the number that would turn this from a Monday mood into a real story, and until it shows up, read every 'AI slowdown' headline with that filter. If you want the buildout explained this way every weekday, follow Concrete Compute wherever you're listening, and if today's episode helped you make sense of a confusing news day, send it to the one person in your life who keeps asking you what's actually going on with AI stocks. 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!

I also host Space Stakes: the business of the new space race, every day. What actually flew, what the contract is really worth, and who has customers. Find it wherever you get your podcasts.