In partnership with
⚡ THIS WEEK'S SIGNAL
The Crane Clean Energy Center just cleared its last real transmission obstacle, and the NRC's own draft review found no reason to stop it — genuinely good news for the deal every nuclear developer cites. But look at what happened right alongside it: Meta split its latest nuclear buy into a safe tranche and a speculative one, and FERC still hasn't priced what a co-located data center owes the grid. The template works. Almost nobody else gets to copy it as written.
━━━━━━━━━━━━━━━━━━━━━━━
📰 MAIN STORY
Constellation just got easier. Everyone behind it just got harder. On June 1, the Federal Energy Regulatory Commission (FERC) approved a waiver letting Constellation transfer 760 megawatts (MW) of grid-interconnection rights from a retiring Eddystone gas plant to the Crane Clean Energy Center — the restarting Three Mile Island Unit 1 reactor that will sell its full 835 MW output to Microsoft under a 20-year power purchase agreement (PPA) signed in 2023. Without that waiver, full deliverability was stuck behind transmission upgrades not expected until December 2030. Days later, the Nuclear Regulatory Commission (NRC) released a draft environmental assessment finding no significant impact from the restart, with a final version expected in September. The template every subsequent hyperscaler nuclear deal gets measured against is, for the first time, genuinely close to working as advertised.
Here's what each side actually wants from it. Microsoft locked in two decades of firm, carbon-free power without taking construction risk — a restart of an already-licensed reactor, not a bet on an unbuilt one. Constellation wants proof the restart model scales, because it's the closest thing the industry has to a repeatable playbook. Meta, watching from behind, split its own January nuclear buy accordingly: a firm, 20-year, 2,609 MW PPA with Vistra for output from reactors already running, stacked next to earlier-stage development agreements with Oklo and TerraPower for reactors that don't exist yet. Amazon took a third path — co-locating load directly behind Talen's existing Susquehanna nuclear plant — and ran straight into the one lever regulators actually control.
The contrast is in how fast each path moves. Constellation asked FERC for its transmission waiver on March 31 and had an answer by June 1 — nine weeks for a bespoke, plant-specific fix. Amazon's co-location structure has been waiting almost two years for FERC to do something far simpler: set a number. The "Grid Reliance Charge" will determine what a co-located data center owes the grid it still leans on for backup and balancing. One incumbent got an individualized shortcut. The newer structure is stuck behind a generic rulemaking that still isn't finished.
The figures worth screenshotting: Crane's restart is estimated at roughly $1.6 billion — Constellation hasn't disclosed a final capital figure in an SEC filing — targeting power to the grid in the second half of 2027, with an NRC operating-license decision reported around May 2027. Meta's Vistra tranche, 2,176 MW of existing Perry and Davis-Besse capacity plus 433 MW of uprates, starts flowing this year; its Oklo (1.2 gigawatts (GW), Pike County, Ohio) and TerraPower (up to eight 345 MW Natrium units) commitments don't reach full scale until 2034. Meanwhile PJM's last two annual capacity auctions cleared at or above its $329-per-megawatt-day price cap — the scarcity price against which every one of these deals, however delayed, still looks cheap.
Scenario A: FERC sets the Grid Reliance Charge low enough this fall to keep co-location economically workable. Talen and Constellation's other co-location prospects move forward, and infrastructure investors get a second nuclear-adjacent structure, beyond restart PPAs, to underwrite. Scenario B: FERC sets it high, or keeps deferring. Co-location stalls as a template, new-build small modular reactor (SMR) risk becomes the only real option left, and ratepayer advocates get a cleaner argument that behind-the-meter nuclear was never going to shield the grid from data-center costs in the first place.
What to watch: any FERC order in Docket No. ER26-1088-001 that actually names a Grid Reliance Charge percentage — that's the moment the abstract policy fight becomes a number every co-location deal has to re-run.
━━━━━━━━━━━━━━━━━━━━━━━
ChatGPT gives you generic answers because you give it generic prompts.
You know the fix: longer prompts, more context, clearer constraints. But typing all that takes five minutes per prompt, so you shortcut it. Every time.
Wispr Flow lets you speak your prompts instead of typing them. Talk through your thinking naturally — include context, constraints, examples — and get clean text ready to paste. No filler words. No cleanup.
Works inside ChatGPT, Claude, Cursor, Windsurf, and every other AI tool. System-level, so there's nothing to install per app. Tap and talk.
Millions of users worldwide. Teams at OpenAI, Vercel, and Clay use Flow daily. Free on Mac, Windows, and iPhone.
━━━━━━━━━━━━━━━━━━━━━━━
🧾 DEAL DECODER
Meta's Triple Nuclear Agreement (Vistra, Oklo, TerraPower — announced January 9, 2026)
Who signs what: three instruments, three risk profiles, one press release. With Vistra, a 20-year PPA for 2,609 MW combining existing Ohio reactor output with plant uprates. With Oklo and TerraPower, earlier-stage development agreements to support reactor lines that haven't been built — not a PPA priced against completed generation.
Who fronts the capital: Vistra funds its uprates against revenue Meta has already contracted; Oklo and TerraPower fund reactor development themselves, layering Meta's offtake commitment on top of federal cost-share and private capital, the same basic design Constellation used with Microsoft.
Who eats the loss if the load never shows up: with Vistra, almost no one — the plants exist and run regardless. With Oklo and TerraPower, the developers carry construction and fuel-supply risk a restart never faced; Meta hasn't disclosed equity exposure, so its downside looks more reputational than financial.
Precedent it sets: bundle one cash-flowing tranche with two speculative ones in a single announcement, so the easy part buys headlines for the hard part. Think of it as a startup wrapping seed-stage bets inside a press release about its most reliable, already-profitable product line — every hyperscaler comparing notes on its next reactor is now asking for the same split.
━━━━━━━━━━━━━━━━━━━━━━━
⚡ QUICK HITS
SpaceX Agrees To Phase Out xAI's Unpermitted Turbines: Under an agreement with Mississippi regulators, SpaceX will begin removing the unpermitted gas turbines powering xAI's Colossus data centers as early as this month, with full removal by July 2027 as the site transitions to a permitted 1.2 GW gas plant. The turbines have drawn NAACP and Southern Environmental Law Center litigation over air permits in a Memphis-area airshed already ranked among the nation's worst for asthma. [Quartz / TechCrunch]
Entergy's $1.8 Billion Gas Plant Buy Gets Shakier: An equity analyst called Entergy's proposed purchase of the Cottonwood gas plant "increasingly in doubt" after Louisiana regulatory staff found it could add roughly $7 a month to a typical residential bill — a figure that surfaced alongside public questions about whether Meta, which is building data centers in Entergy's territory, should be covering more of the cost directly. [Utility Dive]
TerraPower Broke Ground On The First New Reactor Type In Decades: TerraPower's Natrium demonstration plant in Kemmerer, Wyoming received the first NRC construction permit ever issued for a commercial non-light-water reactor in March, and broke ground the following month. It's a genuine industry milestone — though it's a separate project from whichever reactors TerraPower ultimately builds under its Meta agreement, a distinction worth holding onto as the two get conflated in coverage. [U.S. Department of Energy]
FERC Still Hasn't Priced Co-Location: PJM's further compliance filing on how much co-located data centers must pay to use the grid (Docket ER26-1088-001) has been sitting with FERC since May 18, and the Commission has not yet set the Grid Reliance Charge percentage that will decide whether behind-the-meter nuclear arrangements still pencil out. Every unresolved month is a month co-location deals get modeled on a placeholder number. [FERC docket filings]
NuScale's Design Got Approved; Its Customer Problem Didn't: The NRC approved an uprated version of NuScale's small modular reactor design, the second SMR design cleared for U.S. use — a regulatory win that arrives without a single hyperscaler-contracted NuScale project on the books, three years after its one flagship customer contract collapsed. [U.S. Department of Energy]
━━━━━━━━━━━━━━━━━━━━━━━
🧭 THE EXPOSURE MAP
If you hold equities: Nuclear-adjacent names are increasingly being priced on deal-closing probability, not headline gigawatts — the gap between Meta's firm Vistra tranche and its speculative Oklo/TerraPower tranche is the same distinction credit analysts are starting to draw. Watch Constellation's and Vistra's Q3 earnings calls for any language on how co-located load gets treated in PJM's capacity auction, since the unset Grid Reliance Charge is a direct input to both. Nothing here is a security recommendation.
If you run projects at a utility: If your organization has floated a co-location or behind-the-meter arrangement, ask whether it's modeling the Grid Reliance Charge as zero, as PJM's proposed rate, or as unknown — FERC's own paper-hearing record suggests "unknown" is still the honest answer, and a workplan built on any specific number right now is a guess dressed as a schedule.
If you set or influence policy: FERC granting Constellation a bespoke, fast-tracked waiver in about nine weeks, while a two-year-old co-location rate question sits unresolved, previews a two-speed system — incumbent restarts get individualized relief, new entrants wait on generic rulemaking. Watch whether that pattern draws a fairness challenge inside the ER26-1088-001 docket itself.
If you pay a utility bill or buy AI compute: None of this week's nuclear news changes a bill in the near term — Crane's power doesn't reach a meter until at least the second half of 2027, and Meta's Vistra tranche uses plants already running. The mechanism to actually watch is capacity-auction pricing: PJM's last two annual auctions cleared at or above its price cap, and that scarcity value is what makes even a delayed nuclear PPA look cheap by comparison a few years out.
━━━━━━━━━━━━━━━━━━━━━━━
🔧 TOOL / RESOURCE OF THE WEEK
NRC's Crane Clean Energy Center Info Finder: The Nuclear Regulatory Commission's public tracker for the restart, listing every inspection report, license amendment filing, and comment period in one place. Utility program managers and policy staff following the restart timeline can join the CCEC correspondence listserv directly from the page instead of waiting on secondhand reporting. → NRC Info Finder: Crane Clean Energy Center
━━━━━━━━━━━━━━━━━━━━━━━
💬 CLOSING THOUGHT
Crane clearing its last real hurdle is a genuinely good sign — for Crane. What it tells us about the next reactor is close to nothing: a fast, bespoke waiver for an existing licensed unit doesn't predict how FERC treats a brand-new one, and a firm tranche riding alongside two speculative ones isn't the same deal wearing different letterhead. For the optimism to travel past this one plant, the Grid Reliance Charge needs an actual number, and one new-build reactor needs to hit its date on schedule.
So here's my open question: once a single restart clears every hurdle, does that de-risk the next ten new-build reactors — or just the next restart?
