Two point seven gigawatts of power lined up for one Google campus in Wyoming. The entire Wyoming Electric system peaked at four hundred thirty-nine megawatts this summer. So the data center gets its own grid, and the utility collects a management fee. Neat trick. Welcome to The Data Center Daily for Thursday, October 8. Coming up: Google's power stack, Amazon at Homer City, Oracle's Wisconsin line back at square one, an Australian alliance that fell apart, and SpaceX shopping for chip debt.
Black Hills Corp., in its own announcement, as carried by Finviz:
The project, exclusive of additional transmission system expansion investments, is anticipated to be served by a total resource mix of 2.7 GW, including reserve margins, and is planned to begin taking energy service in late 2027 and ramp to the project’s peak load in 2030. Black Hills will provide up to 590 MW of grid-connected energy service through company-owned generation and market energy.
It's a press release, so read the parts. Announced Tuesday, effective September 30, running through 2048. Black Hills puts one point eight billion dollars into five hundred sixty-four megawatts nameplate of new gas at its Cheyenne Prairie Generating Station, plus twenty-six megawatts from the market and retail service. That's the five-ninety. And the other two point one gigawatts? Third-party Wyoming resources, run through a private microgrid under Black Hills' Large Power Contract Service tariff. AI Industry Today notes Black Hills hasn't disclosed the technologies or the counterparties behind them. So close to eighty percent of the power stack is a blank. Wind, gas, or a promise? I'd like to know. AI Industry Today's other point matters: two point seven is the resource portfolio, not the load. Black Hills previously described the campus as one point eight gigawatts, which leaves roughly nine hundred megawatts of headroom. Even the load alone is more than four times Wyoming Electric's all-time peak, set on July 20. CEO Linn Evans says Google bears all costs of serving the campus for the life of the project. Google already advanced three hundred ninety-nine million, refundable, for long-lead equipment. That's how you get turbines in this market. Same day, per The Elec, Google signed with Constellation: a twenty-year PPA for eight hundred ninety megawatts of new capacity from uprates at eleven existing nuclear units, first one in 2028, plus a fifteen-year supply agreement covering two point seven gigawatts of existing PJM generation. Thirty-five ninety megawatts in all. Only the eight-ninety is new supply.
Mike Bavto, reporting on the Indiana County filing:
Amazon Data Services, Inc., the company's development arm, filed a Preliminary Land Development Plan with the Indiana County planning commission on August 28 under the project name PIT100 Masterplan, laying out 36 data center buildings at the Homer City Energy Campus 50 miles east of Pittsburgh.
The filing dates to August. What's new is the county's posted package naming AWS as the anticipated end user, first reported Tuesday by DCD. Until now Amazon had only acknowledged talks. About twenty-two hundred acres, construction starting in December, final phase done in June 2036. And the power is the whole pitch. This is the old Homer City coal plant, two gigawatts, shut in mid-2023. The replacement is a four point four to four point five gigawatt combined-cycle plant on seven GE Vernova 7HA.02 turbines, with EQT agreeing in principle to supply Marcellus gas and first power targeted for 2027. Largest gas plant in the country, if it gets built. Scale check from Baxtel: the entire Pittsburgh market tracks two hundred eighty-four megawatts. Homer City shows thirty-seven hundred reserved for data centers. More than a hundred residents packed the September 9 planning meeting over noise, water and property lines, and commissioners tabled it. AWS says it'll take power through FirstEnergy, backed by the on-site plant. If the halls ramp slower than the turbines, who's holding that utility hookup? So the long-doubted project now has a tenant and a calendar. Still a preliminary land plan, though. Not a lease, and nothing energized.
Wallstreetcn, via Longbridge, summarizing a new report from research firm Aterio:
An administrative law judge subsequently determined that ATC failed to comply with orders regarding listing changes, and the PSC closed the case on September 10 without ruling on the substantive content. ATC resubmitted its application on September 18, and the statutory review clock restarted from zero.
On Monday we had Oracle offering to take a slice of Point Beach nuclear for Project Lighthouse, its one point three gigawatt Port Washington campus with Vantage. Today it's the wire. American Transmission Co. needs Wisconsin PSC approval for a new high-voltage line, and on August 7 the commission unanimously rescinded its completeness finding after ATC filed or refiled five hundred sixty-four documents. So the build-by-end-of-2027 plan is gone. Aterio's base case is full power in October 2028. The delay case is April 2029, and Aterio says that's the one history backs: comparable three forty-five kV cases took three hundred fifty-five to three hundred fifty-eight days from completeness to decision, not one-eighty. The line got pricier, too. ATC's refiled interconnection project is estimated at two point four eight to two point seven two billion dollars, up about seventy-five percent, and roughly one point one billion of that is a grid stability system Oracle has committed to cover. Two buildings have already topped out. Next marker: the completeness ruling, due by October 19. Tom Content at the Citizens Utility Board notes Wisconsin's biggest customer today draws under a hundred megawatts. This campus could peak at thirty-five hundred. Oracle hasn't responded publicly. One sourcing caution. This is a translated aggregator write-up of Aterio's analysis, and the deferred-revenue figure, eight-fifty to nine hundred million a month, is the write-up's own math off a Goldman Sachs per-gigawatt estimate. It's not an Oracle disclosure.
Robert Clark, writing in Light Reading:
Nvidia-backed Australian data center alliance Project Southgate has broken up after just 12 months with a fraction of its planned 1.6GW buildout completed. CDC Data Centers has ended its partnership with Firmus – claimed to be worth as much as 73 billion Australian dollars (US$50.7 billion) – due to fears over Firmus' aggressive Asian expansion plans.
Southgate was unveiled in October 2025: Firmus AI factories on CDC data center infrastructure, scaling to one point six gigawatts by 2028. What actually got built is forty-two megawatts, all in a single CDC facility in Melbourne. Forty-two out of sixteen hundred. Under three percent. CDC's chief executive Greg Boorer told Rampart they're not doing that rollout with Firmus because Firmus made other choices about its own developments. Firmus co-CEO Oliver Curtis tells it differently: the two mutually agreed earlier this year to end it, and it doesn't touch Firmus' plans. Those other choices, per Light Reading: a three hundred sixty megawatt Nvidia-backed site on Batam Island in Indonesia, two Malaysian sites for OpenAI, and two more in South Australia. None involve CDC. And Firmus has an IPO pending, with the price reportedly dropping from eleven Australian dollars a share to nine on weak demand. Mostly unbuilt capacity, and the buyers noticed. That fits the pattern we laid out Tuesday with SB Energy and Holtec: investors marking down developers whose backlog is mostly paper.
Amit Chowdhry, summarizing a Reuters report:
SpaceX is seeking approximately $40 billion in financing to purchase Nvidia artificial intelligence chips, with the Elon Musk-led company discussing the potential financing with banks and asset managers, according to a Reuters report. The proposed financing would consist of approximately $10 billion in bank loans and $30 billion of investment-grade debt.
Apollo is expected to lead the deal and distribute the debt, and PIMCO is among the parties in talks. The money buys Nvidia hardware. Musk recently said xAI's Colossus 2 could more than double its Nvidia chip count by December, and that the company plans to build exclusively on Nvidia. Thirty billion of investment-grade debt to buy chips. Chips age a lot faster than turbines or transmission lines, and this is getting structured like infrastructure anyway. On Saturday, October 3, we covered xAI renting out its surplus GPUs. Now Musk's side wants forty billion more of them. Context from the same piece: Morgan Stanley estimates AI infrastructure will need about one and a half trillion dollars of external financing by 2028, even as lenders get choosier. Line it up with Firmus and Oracle today: money for the chips shows up fast. The wires and the permits are what's slow. Keep in mind it's a reported negotiation, not a closed deal, and none of these terms come from SpaceX.
Black Hills and Wisconsin both come down to who pays when a giant load lands on a small utility. If that's the thread you follow, our sister show Power Bill covers residential electric rates and utility cost shifts every weekday, so you know what hits your bill next month before it does. Search for Power Bill in your podcast app.
On the watch list: Taylor, Texas, where the council takes up its six hundred sixty-four acre annexation vote today, and the Wisconsin commission's completeness ruling on ATC's line, due by October 19. And whether anyone names the two gigawatts behind Google's Wyoming microgrid. Sources for every story are linked in the show notes. The Data Center Daily is a Lantern Podcast. Talk to you tomorrow.