AI load is hitting grid rules, gas queues, and gigawatt campuses all at once. And the calendar is not exactly helping. It’s The Data Center Daily, Friday, May 8th. Today: FERC delays, Stargate expands again, a ten-billion-dollar Texas lease, and gas just jumped wind in the ERCOT interconnection queue for the first time in a decade. That’s the one I’m on, because it tells you who’s actually getting stuck with the bill for this AI buildout. And it’s not the hyperscalers. All right, plus Kevin O’Leary gets a permit. Let’s get into it. From HyperSinc:
The Federal Energy Regulatory Commission just announced it will make a final decision on the most consequential grid policy in decades, and it is going to miss the deadline by two months. On the RM26-4 docket, which will determine how data centers, AI manufacturing facilities, and industrial loads larger than 20 MW connect to the U.S. power grid, FERC said it will act by June 2026, not April 30.
FERC missed its April 30 deadline on Docket RM26-4, the rulemaking for how data centers and other large industrial loads actually hook into the U.S. grid. New target is June. In the meantime, 86 gigawatts of planned generation is sitting in queue waiting for rules that do not exist yet. Eighty-six gigawatts. That’s not a pipeline, that’s a parking lot. And all those gigawatt-campus announcements keep coming while the interconnection math stays unresolved, because until FERC decides on participant funding, nobody actually knows who pays what. The participant-funding piece is the one to watch. FERC is floating a 100 percent cost shift onto large loads for grid upgrades, and if that lands, it rewrites the economics of every co-location deal signed on some promise about “available capacity.” And then the RTOs still have to implement whatever FERC actually publishes. So a June decision does not mean June interconnection. It means June is when PJM or MISO starts turning a final rule into real queue positions, which is a whole different thing. Here's Brand Vision:
OpenAI and Oracle on Tuesday unveiled the next phase of their sprawling Stargate initiative: a 4.5‑gigawatt expansion that will deploy hyperscale data‑center campuses across at least four states and house two million AI accelerators. The announcement lifts Stargate’s total committed capacity to just over 5 GW—already the world’s largest single AI infrastructure program—and strengthens Oracle’s bid to catch Amazon and Microsoft in the cloud‑services race.
OpenAI and Oracle are adding 4.5 gigawatts to Stargate, which takes the program to just over 5 GW of committed capacity. Two 800-megawatt campuses are set to break ground in Texas and Michigan early next year, and Wisconsin and Wyoming are still under negotiation, so they are not in the column yet. Four states, 4.5 gigs, and “tens of billions over the decade” — that’s a press-release budget, not a construction budget. Tell me which site has a signed interconnection agreement, and then I’ll start paying attention. The clean-energy pitch is hydropower and small-modular nuclear against a 2030 carbon-free target. And SMRs at commercial scale by 2030 is a very optimistic read of where that technology is right now. And “long-term clean-energy contracts” is doing a lot of work when the article does not say a single one is signed. That’s the oldest trick in the hyperscaler announcement playbook. Akash Sriram, writing in The Star:
The agreement covers 352 megawatts (MW) capacity in the first phase of the project with an undisclosed tenant. Hut 8 said the client would install computing equipment at the site to support large-scale AI training and operations.
Hut 8 just signed a 15-year, $9.8 billion lease — take-or-pay, triple-net, no termination for convenience — for 352 megawatts at its Beacon Point campus in Nueces County, Texas. That’s phase one of a planned one-gigawatt build. The tenant is undisclosed, but Genoot called them high-investment-grade. Undisclosed tenant, $9.8 billion commitment — that’s the number everybody’s going to print. Almost nobody will mention that 352 megawatts in Nueces County still has to get powered. What’s on the grid out there, and who’s paying for the interconnection? Take-or-pay with no termination for convenience is genuinely load-bearing language. That is not a letter of intent in a hard hat. The structure is real. Whether the gigawatt campus ever gets past phase one is the next question. Stock up 25 percent on a lease signed to somebody we can’t name, for a campus that’s mostly still dirt. I’m not saying it’s not real — the contract language is tight — I’m saying the market is pricing in the full gigawatt before a single rack gets racked. Here's Chain Tech Daily:
Box Elder County commissioners in Utah voted unanimously on May 4 to approve the Stratos AI campus backed by Kevin O’Leary Digital, the infrastructure arm of O’Leary Ventures. The approval came over the objections of hundreds of residents who chanted “Shame!” as the vote was announced and who said they had been given too little time to raise concerns before the decision.
Following up on the Box Elder scrutiny we flagged last edition: commissioners voted unanimously on May 4 to approve the Stratos campus — 40,000 acres, 9 gigawatts at full buildout, phase one at 3 gigawatts, and a natural gas pipeline on-site. Hundreds of residents were in the room chanting “Shame” as the gavel came down. Nine gigawatts. Utah’s entire grid is about 4. So O’Leary is proposing to build more than twice the state’s current generation capacity, privately, on-site, burning gas — and the county said yes over loud public objection and, apparently, a very short comment window. O’Leary framed it on Fox Business as national security — China built 400 gigawatts of AI-capable power, and we need to respond. That’s a real argument. It just still does not tell you where the water comes from in a high-desert county. And “approved by Box Elder County” is not the same as permitted, financed, interconnected, or built. That 9-gigawatt number is a full-buildout ceiling on a county land-use vote. Somebody still has to site a gas pipeline, pull water rights, and get FERC and PacifiCorp comfortable with whatever this thing does to the regional grid. AOL, with Brandon Mulder:
For the last six months, the volume of gas generation in the Texas grid’s interconnection queue — the yearslong waiting list for electric generators wanting to connect to the grid — has surpassed wind. It’s the first time since January 2016 that gas has overtaken wind in the queue, a shift that reflects the policy and economic headwinds facing the wind industry and data centers favoring gas power as they seek to cash in on the artificial intelligence boom.
First time since January 2016: gas has overtaken wind in ERCOT’s interconnection queue. Six straight months, and the driver is exactly what you’d expect — data centers chasing 24/7 baseload, not capacity factors. Queue, though. Queue. Lawrence Berkeley says only 22 percent of ERCOT queue projects ever get built. So we’re really talking about what developers are fantasizing about, not what’s going to be running wire by 2028. Fair — announced is not energized. But the directional shift still matters. Wind had held that queue lead for a decade, and that’s a real signal about where the capital thesis is pointing, even if most of these projects never close. And who’s eating the transmission costs when half these gas plants queue up next to an already congested load pocket? The hyperscalers sign a PPA and call it clean energy, the gas developer collects the interconnection slot, and the ratepayer gets stuck with the grid upgrades. Got a tip, a story idea, or a correction for The Data Center Daily? Send it to datacenterdaily at lantern podcasts dot com. We read every note, and your feedback helps shape the show.
You’ll find links to every story we covered today in the show notes, so if something caught your ear, they’re all there for a deeper read.
That’s The Data Center Daily for this Friday, May 8th. Thanks for listening. This is a Lantern Podcast.