In partnership with
⚡ THIS WEEK'S SIGNAL
Add up every hyperscaler nuclear deal on the board and you get somewhere between 9.8 and 13 gigawatts, depending on who's counting and how generously. It's a real number, and it's a large one. But strip out the plant restarts and the uprates at reactors that are already spinning, and what's left is a stack of reactors that don't commercially exist yet — arriving in the 2030s to answer a demand curve that's steepening right now. Nuclear is the announcement. Gas and the grid are still the power plant.
━━━━━━━━━━━━━━━━━━━━━━━
📰 MAIN STORY
Four hyperscalers have now committed to nuclear at a scale that would have sounded absurd three years ago. The question worth your time isn't whether the deals are real. They are. It's what fraction of that capacity is firm power you can plan around before 2030 — and what fraction is a promise about a reactor that hasn't been built, licensed, or run anywhere in the country yet.
Start with what each player actually wants. Microsoft, Amazon, Meta, and Google are not buying nuclear because it's the cheapest or fastest megawatt available — it isn't either. They're buying 24/7 carbon-free baseload that lets them keep a clean-energy story intact while their load doubles. The utilities and independent power producers on the other side — Constellation, Talen, Vistra — want two decades of contracted, investment-grade revenue to underwrite plants that spent years fighting for economic survival. The advanced-reactor developers — Kairos, Oklo, TerraPower, X-energy — want the one thing a startup reactor company has never had: an anchor customer with a balance sheet deep enough to make the capital markets take a first-of-a-kind build seriously.
Here's the contrast that the single portfolio number hides. Roughly 6.4 GW of the committed capacity is existing nuclear — restarts and uprates at plants like Crane (the former Three Mile Island Unit 1), Susquehanna, Clinton, Perry, and Davis-Besse. That's power you can put in a resource plan. The rest — Google's Kairos fleet, Meta's Oklo and TerraPower units, Amazon's X-energy modules — is small modular reactor capacity from a technology that has zero commercial units operating in the United States today. Same press releases. Two entirely different risk profiles.
The data, by counterparty, verified against primary sources:
Microsoft / Constellation (Crane, ex-TMI Unit 1): ~835 MW, 20-year PPA signed Sept. 2024. Existing-plant restart; ~$1.6B project, $1B DOE loan. Restart now targeted for 2027. (Note: the widely repeated "$16 billion" figure is the projected GDP impact to Pennsylvania, not the deal value.)
Amazon / Talen (Susquehanna): 1,920 MW PPA through 2042, signed June 2025, ramping to full delivery by ~2032. Existing plant, now front-of-the-meter.
Meta / Vistra + Oklo + TerraPower: up to 6.6 GW by 2035, signed Jan. 2026. Mix of existing Ohio/Pennsylvania plants and uprates (~2.6 GW, Vistra) plus new SMRs (Oklo, TerraPower) that start no earlier than 2030–2032.
Meta / Constellation (Clinton): ~1.1 GW, 20-year PPA signed 2025 — a separate deal, not part of the 6.6 GW.
Google / Kairos Power: up to 500 MW by 2035; first unit (Hermes 2, ~50 MW) targeted for 2030.
Amazon / X-energy: anchor investment toward up to ~5 GW of SMR capacity by 2039 (Energy Northwest, Dominion).
Two scenarios to watch from here:
Scenario A — the restarts and uprates land on schedule. If Crane connects in 2027 and the Vistra and Susquehanna volumes ramp as contracted, roughly 6 GW of firm, near-term nuclear reaches PJM this decade. That's a genuine, if modest, structural addition — and it validates the "restart economics" thesis without the SMR question ever being answered.
Scenario B — the SMR timelines slip, as first-of-a-kind builds tend to. If Hermes 2, Oklo's Aurora units, and the X-energy modules push right by even three years, the 2030s portion of the portfolio thins out, and the capacity gap it was meant to close gets filled by exactly what's filling it now: natural gas and solar-plus-storage. The nuclear number stays on the slide deck; the electrons come from a turbine.
The one thing to watch: Kairos's Hermes 2 in Oak Ridge — the first commercial-scale unit under any of these deals. Its construction pace is the closest thing we have to a real clock on whether the SMR half of this portfolio converts.
━━━━━━━━━━━━━━━━━━━━━━━
Everything GTM. One platform.
Small teams don't have time to stitch together five tools and hope it works.
Apollo gives you everything you need to find leads, reach them, and close deals — all in one place:
230M+ verified contacts
AI-powered outreach
Data enrichment
Inbound lead capture
Meeting scheduler
And more
Stop juggling tools and start building pipeline that scales.
With Apollo, the AI revenue engine powering 4M+ users.
━━━━━━━━━━━━━━━━━━━━━━━
⚡ QUICK HITS
The Crane restart has a regulatory catch nobody's talking about: To deliver its full ~835 MW at restart, Constellation asked FERC to transfer capacity interconnection rights from a fossil plant it planned to retire — and PJM's independent market monitor filed in opposition in April 2026. Whether all 835 MW can actually reach the market on day one is now a live FERC question, not a settled one. Utility Dive
Meta's 6.6 GW leans on old plants first, not new reactors: The near-term megawatts come from Vistra's existing Perry and Davis-Besse plants (~2.6 GW ramping from late 2026), while the Oklo and TerraPower SMR units don't start until 2030 and 2032 at the earliest. The "advanced nuclear" headline is front-loaded with conventional nuclear. Utility Dive
Google put a shovel in the ground: Kairos broke ground on the Hermes 2 demonstration plant in Oak Ridge, Tennessee, in April 2026 — the first commercial-scale step under the 2024 Google agreement, with ~50 MW targeted to reach the TVA grid around 2030. It's the portfolio's first physical pace-setter, and worth tracking as the real test of SMR timelines. World Nuclear News
Carnegie sized the whole thing — and it's smaller than it sounds: A June 2026 Carnegie Endowment report concluded that if every announced hyperscaler nuclear deal is fully realized, it adds roughly 13 GW — about 100 TWh a year — by the mid-2030s, against data center demand that could top 550 TWh annually by 2035. Nuclear, at its most optimistic, covers a fraction of the load, and arrives late. Carnegie Endowment
The biggest existing-plant deal got restructured by regulators: Amazon's Susquehanna arrangement moved to a front-of-the-meter PPA after FERC rejected the original behind-the-meter interconnection amendment in late 2024 — the transition completes after the plant's spring 2026 refueling outage. It's a reminder that even a signed nuclear deal can be reshaped by how the power is allowed to flow. Utility Dive
━━━━━━━━━━━━━━━━━━━━━━━
🔧 TOOL / RESOURCE OF THE WEEK
Carnegie Endowment — "Beyond the Hype: Assessing Hyperscaler Nuclear Commitments Against U.S. Energy Realities": A June 2026 report (Pendleton & Schuessler) that maps the announced deals against actual U.S. energy realities — fuel supply, spent-fuel storage, and timelines. It's the cleanest independent sizing of the portfolio available, and it's the antidote to counting nameplate ceilings as delivered power. → carnegieendowment.org
━━━━━━━━━━━━━━━━━━━━━━━
💬 CLOSING THOUGHT
The honest read is that these deals are doing two jobs at once. They're adding a few real gigawatts of firm nuclear this decade through restarts and uprates — that part is solid. And they're trying to finance an entire generation of reactors that don't exist yet, on the bet that a deep-pocketed anchor customer is what the SMR industry has always been missing. That second bet might work. It's also the one that fills a 2020s power gap with a 2030s solution.
So, here's what I'd actually want to know from anyone underwriting a nuclear-adjacent site right now: are you pricing the SMR timelines as your base case, or as optionality you'd be glad to have but wouldn't stake the campus economics on?
