ERCOT's board just approved Batch Zero — and the first thing it does is hand San Angelo a data center the city may not actually have legal authority to regulate. Welcome to The Data Center Daily. Today we're tracking a Texas city stuck between a grid-operator approval and a zoning void, PJM moving on two procedural tracks at once, a $16 billion Michigan groundbreaking with some very familiar community-benefit language, an AWS project tied to a named deep aquifer in Missouri, and Nvidia saying supply stays tight through 2027. So the week started with queue gigawatts as a math problem — now it has a zip code and a city council that can't really touch it. Grid approval and land-use authority are different tools. Today we're watching a city learn that the hard way. Here's Sarah Boates at San Angelo LIVE!:
Historically, ERCOT and the Public Utility Commission processed large power load requests one at a time. That serial approach has become inefficient amid the data center boom. ERCOT is currently reviewing requests to add 438 gigawatts of power in Texas — more than five times the amount of energy used during the state’s record peak demand.
ERCOT's Batch Zero framework cleared the board Monday and Tuesday — that's the feasibility fight we flagged last episode, now approved and heading to the PUC for final sign-off. But the San Angelo LIVE piece points to something the vote itself doesn't solve: three data center proposals in Tom Green County, and local government may have no regulatory authority over any of them. Skybox-Emergent on city-owned land, Beacon in Dove Creek, Cipher Digital's Colchis project — three sites, one municipality, and a grid-operator framework that got approved upstream while the city council apparently has no zoning lever on any of it. ERCOT approved a process. San Angelo didn't get a vote. That's the part worth naming: the venue shifted from county to grid operator to PUC, but the governance gap didn't go away. It just moved up a level and left San Angelo with the local consequences and no local tool to answer them. From Faizan Hassan Hajam at Zero-Emission Grid:
The PJM Planning Committee (PC) meeting held on June 02, 2026, primarily focused on PJM manual 14H revisions, including IEEE 2800-2022 adoption and expedited interconnection track (EIT). Key points from the meeting are summarized below.
PJM's Planning Committee met June 2 and moved two things at once — IEEE 2800 adoption for inverter-based resources, driven by FERC Order 901 and a NERC standard that goes live October 1, plus the Expedited Interconnection Track in the same meeting cycle where the board already sent twelve proposals to FERC. That's not routine housekeeping; that's the procedural machinery actually turning. The IEEE 2800 piece is the one I want to stress-test — voltage ride-through, frequency response, protection settings for inverter-based resources. That's the physical interface layer. Before anybody calls a PPA clean, I want to know whether the IBR on the other end of that wire can actually meet Attachment K once it lands in Manual 14H. And the Expedited Interconnection Track is the live question from that twelve-proposal package — is EIT actually a faster lane, or is it the same queue with a new label on the waiting room door? The Planning Committee is where that gets built, and we're watching it in real time. If EIT speeds up hyperscaler large-load requests while everyone else sits on the standard track, that's a policy choice dressed up as a procedural reform. I'd like to see that written plainly in Manual 14H before anybody calls it a win for grid reliability. Annemarie Mannion, writing in Engineering News-Record:
The 250-acre campus in Saline Township, about ten miles southwest of Ann Arbor, is being developed by Related Digital for tech firms Oracle and OpenAI, as a 1.4-GW-scale campus planned to have three 550,000-sq-ft, single-story buildings.
Saline Barn got its groundbreaking June 1 — $16 billion, 1.4 gigawatts planned, 250 acres in Saline Township, Walbridge as the GC, and ENR is calling it the largest single project in that firm's 110-year history. Governor Whitmer was on stage calling it Michigan's largest economic investment ever made. Altman said the data center will not increase energy prices and will use less water than a normal — and then the sentence got cut off, which is doing a lot of work here. 'Will not increase energy prices' is a claim, not a tariff rider. Is that in a utility agreement, or is it just what he said at the podium? That 'model for other communities' line is exactly what OpenAI used at the Louisiana groundbreaking two days ago. Same CEO, same framing, different state. Which is fine — until somebody asks what the model actually requires contractually and realizes it's a speech, not a document. This one's from ABC17NEWS:
Water will be sourced from deep within the Cambrian-Ordovician aquifer, with wells drilled to depths of 1,500 feet to avoid interfering with local private wells. According to a report by CDM Smith, a global engineering and construction firm, the Project Green campus will draw 2.9 million gallons of water annually.
AWS Project Green in New Florence, Missouri — 17-building campus, Cambrian-Ordovician aquifer, wells drilled to 1,500 feet, and a high-yield well permit in play. CDM Smith's engineering report puts the full build-out at 50 million gallons annually. That's the first time this week we've had a named aquifer and a named permitting application on the intake side of the water conversation — not discharge, not aggregate draw, actual well depth and volume on a specific project. 1,500 feet into the Cambrian-Ordovician aquifer — that's a confined aquifer, geologically isolated, slow recharge by definition. I want to know what the Missouri DNR's modeled recharge rate is, because 50 million gallons a year sounds manageable until you stack it against how long it takes to replenish. And did Montgomery County's water authority get a technical review, or did AWS just send them a site-plan PDF? The Missouri DNR Groundwater Section chief says high-yield wells at this volume are common across the state — that's the official framing. What the CDM Smith report still doesn't answer, at least in public, is cumulative draw if all 17 buildings come online at once versus the 2.9 million gallon figure, which appears to be single-phase. Here's Markus Kasanmascheff at Winbuzzer:
Huang said supply-chain bottlenecks may persist. Following that warning, customers are left with a practical question, not just an investor one: how much new AI capacity can the industry deliver on time across memory, optics, and finished racks? Nvidia’s AI infrastructure business is carrying much of the load.
Nvidia posted $81.6 billion in Q1 revenue and is guiding $91 billion for Q2 — that's the headline. But Jensen Huang's real signal is in the constraint list: memory, optics, networking, and finished systems. That's not one bottleneck, that's four, and Vera Rubin shipments in the second half are the first real test of whether any of them clear. Every hyperscaler capex number we've quoted this week — the $16 billion in Saline, AWS in New Florence — lands on a wait list if Huang's supply call holds through 2027. I'd like to know which of those announced buildouts have firm hardware allocations and which ones are just racing toward racks that haven't been manufactured yet. That's the distinction to keep in view: a signed PPA and a permitted site gets you a data hall. It still doesn't get you a Vera Rubin cluster if the supply chain is gated at optics and memory two years out. If you follow the infrastructure behind AI, you may also like Anthropic Pentagon Watch, a daily briefing on Anthropic's fight with the DoD over Claude, military AI use, autonomous weapons, and AI procurement blacklisting. Find it wherever you listen to podcasts.
You'll find links to every story we covered today in the show notes, so if something deserves a closer look, they're all there for you.
That's The Data Center Daily for this Wednesday, June 3rd. This is a Lantern Podcast.