Riot has a massive AI lease. Now comes the awkward part: finding the power. On The Data Center Daily: Texas promises, Arizona guardrails, and FERC telling grid operators to explain themselves. First up: Riot Rockdale, and the delivery dates buried beneath the headline. Follow the show and the next briefing lands in your feed on its own. Here's Foley & Lardner LLP:
The lease, which runs through June 2048, is expected to generate approximately $9.1 billion in total initial contract revenue and includes two five-year extension options that could increase the total potential contract value to approximately $16.1 billion. The agreement provides for a phased deployment schedule, with the initial 96 IT MW expected in December 2027 and the full 191 IT MW deployment expected by June 2028.
There’s a real contract number here: 191 IT megawatts, 20 years through June 2048, and about $9.1 billion in initial revenue. The $16.1 billion figure only comes with the extensions, so keep that in the optional-upside column. The first 96 megawatts are due in December 2027, with all 191 by June 2028. Rockdale is in ERCOT, while Abbott has paused new data-center connections for that 474-gigawatt queue audit. Those dates still have to survive a very crowded power system. This is Riot’s second Rockdale tenant after AMD, putting the campus at 241 MW of contracted critical IT capacity and roughly $9.8 billion in long-term revenue. It’s bigger than Bitdeer’s Tydal arrangement, too—but watch the phased delivery dates, not the headline valuation. David Bell, writing in GilaValleyCentral:
PHOENIX — The electric co-ops serving Southeastern Arizona joined together to try to ensure that residential and retail customers aren’t paying more to subsidize large users, such as data centers. On Wednesday, Aug. 12, the Arizona Corporation Commission approved the application by Arizona Electric Power Cooperative (AEPCO) for a standardized process for approval of large-load customer agreements.
We just had a $9.1 billion Texas lease with delivery running into June 2028. In Arizona, co-ops are putting the customer agreement under a microscope before a big load leans on rural wires. Finally, paperwork with teeth. Under AEPCO’s August 12 process, the applicant has to disclose the business, location, interconnection, self-supply, interruptibility, and the actual agreement—before households get volunteered as the backstop. It’s a standardized approval process, not blanket approval. AEPCO explicitly says large-load inquiries include data-center developers, along with industrial, manufacturing, and greenhouse customers. One big gap: it covers electric supply, not water use. A campus can answer whether it brings onsite solar and still leave the local water fight for another room. Here's Sebastian Moss at DataCenter Dynamics:
The rapid buildout of the largest data center project in history is hard to keep up with. In the course of conversations over several months with OpenAI staffers about the trillion-dollar Stargate project, the scope and scale of the endeavor have morphed, growing ever larger and more expensive.
OpenAI’s Keith Heyde says the execution strategy is “constantly in flux.” On a trillion-dollar build, the land, power, shells, GPUs, debt, and partners are all moving targets at once. DCD’s reporting makes Stargate look less like one campus plan and more like separately negotiated infrastructure layers. We just covered Riot’s 191 IT MW lease with December 2027 and June 2028 delivery dates. Stargate has to coordinate that kind of timetable across a vastly larger portfolio. “Awe-inspiring compute demand” is a lovely phrase until somebody has to finance the power plant and transmission. DCD says NVIDIA’s Ohio residual-value guarantees already cap at $105 billion—and that’s one partner on one piece of this machine. The key disclosure is that the plan is still moving, not the trillion-dollar headline. If partner roles and execution paths keep shifting, an announced capacity figure needs a date, a signed power path, and an in-service commitment before it goes in the delivered column. Here's EnergyCentral:
This summer, federal regulators told the operators of America's largest power grids that their rules for connecting big new customers, what the industry calls large load interconnection, appear inadequate, and gave them a deadline to fix those rules or justify them. It's the most serious attempt in years to change how anything gets connected to the grid.
A children’s hospital and a gigawatt server farm sit in the same interconnection line, and we still call that a functioning system. FERC told all six organized markets their large-load rules look inadequate. BYOP is what happens when customers stop waiting for permission. And BYOP is bigger than a Texas data-center escape hatch. The piece argues hospitals, water authorities, and industrial parks may pursue self-supply too, meaning the grid’s biggest customers increasingly solve reliability outside the tariff. Right, but self-supply still needs equipment, land, permits, and often a wire somewhere. You don’t defeat a five-year interconnection queue by printing a brochure that says ‘bring your own power.’ Arizona’s co-op process offers another approach: connect the large load, but spell out who pays. FERC is pushing on the queue from above, while state regulators decide who carries the bill. Inevitable writes:
Yasir Arafat is the co-founder and CTO of Aalo Atomics, a company developing factory-manufactured modular nuclear power plants designed for AI data centers. Before founding Aalo, Yasir led the MARVEL microreactor program at Idaho National Laboratory. In this episode of Inevitable, Yasir explains how Aalo became one of four U.S. nuclear startups to achieve criticality by the July 4, 2026 deadline.
Aalo reached criticality by July 4—one of four U.S. nuclear startups to clear it. That’s a real reactor milestone, not another rendering with blue server racks behind it. And Yasir Arafat’s team says it sustained a fission chain reaction in less than a year, after building its reactor building in 36 days. For nuclear, those are unusually concrete schedule markers. But criticality doesn’t mean 24/7 power at an AI campus. Factory manufacturing may compress construction, but it doesn’t make the licensing, fuel, and operating case disappear. Physics remains stubbornly unpersuaded by hyperscaler capex. Put Aalo beside the BYOP pressure we just covered: with large-load grid rules under federal scrutiny, a co-located microreactor is a long-lead answer to a very immediate power problem. If you follow the infrastructure behind modern AI, check out 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.
What we’re watching next: Riot’s first 96 IT MW under the Rockdale AI lab lease is expected in December 2027, followed by the full 191 IT MW deployment in June 2028.
Links to every story are in the show notes, so take a look at the ones that caught your attention. That’s The Data Center Daily for today. This is a Lantern Podcast.