Powerstack: Google, Meta, Amazon all uprate nuclear

The Currence team has had a busy few weeks on the ground. During New York Climate Week, we hosted a load flexibility roundtable alongside our friends at Constructive, pulling together 50+ leaders to dig into what it actually takes to scale flexibility past pilot-sized programs. Then we hit the road and spent three days at Yotta in Vegas, where 7,000 data center operators traded new gold-rush energy with a lot of hard questions about power and social license. 

Clients can catch up on both: six takeaways from the flexibility roundtable and what we heard at Yotta. And ICYMI, our Geothermal Market Update - Q3 2026.

This is Currence’s weekly newsletter on the moves and motives shaping the load growth era. Not a subscriber yet?

What happened

The US nuclear industry showed even more signs of a renaissance last week, with three nuclear uprates signed in six days. 

Add that up, and the uprates squeeze about 1.5GW of new juice out of old nuclear. But the hyperscalers backing them get even more, nearly 6.9GW in total contracts.

We've said it before, we'll say it again: uprates are among the fastest clean firm megawatts around (see below). They skip siting fights and multi-year construction permits, and many grid operators’ interconnection tracks explicitly cover uprated capacity. Most uprates raise how much heat a reactor is licensed to produce, which still needs NRC sign-off. The agency targets six-month reviews for the small ones, which add under 2% using better instruments, and 12 months for the larger, which add up to 20% and need major equipment work. (Efficiency upgrades, like a new turbine that turns the same heat into more electricity, skip new licenses entirely.) Everyone wants the megawatts, sure, but in today's data center-fueled load growth environment, buyers are most excited by the speed.

They come at a reasonable price, too. Google's deal works out to under $5,000/kW of new capacity. That's above the $2,200-3,000/kW that new gas plants were running earlier this year, but large gas turbine slots are sold out into the 2030s. And it's a fraction of first-of-a-kind SMRs: Ontario Power Generation's four-unit BWRX-300 build at Darlington is budgeted at about $15bn for 1,200MW, or roughly $12,500/kW, and that average already assumes the later units come in cheaper than the first.

Uprates aren't new. Since 1977, the NRC has approved 172, adding about 8GW to the US fleet, starting with Calvert Cliffs, the same plant in Amazon's deal. Market pressures nearly killed the practice over the past decade, and the NRC hasn't approved a stretch or extended uprate since 2017. Now Washington wants them back. Executive Order 14302 directs the DOE to help deliver another 5GW of uprates by 2030, and the NRC expects applications for 30 units by then, with faster reviews to match. What's new is hyperscalers willing to fund them, and take a long contract on the rest of the plant in return.

You can't keep uprating forever, though. A report for Idaho National Laboratory from February puts what's left through conventional methods at roughly 4-5GW. Novel approaches could stretch that. Alva Energy says its second-turbine design could add 10GW across the fleet. That's nice to have, and still far short of the 300GW of new nuclear the administration wants by 2050. The new megawatts hit a limit, but the existing fleet they’re attached to is the bigger prize.

Mark’s take

Okay, so I’m going to be selfish for a second.

We called this. Or at least we’ve been suggesting this for a while now.

At Currence, we have this triple thesis, that in the AI power buildout:

At the top of this year, we started publishing a series called The Fastest MW (or Currence clients, here), where we explore pathways of how to get cleaner power faster – and we give real-live options in case you want to pursue one of them. (see the chart above) We’ll continue this series in the coming weeks.

We suggest things like buying Solar + BESS projects in ERCOT that already have an IA in place (and give 35 projects that could be no-brainer acquisition targets). Or we give geothermal sites where output has declined, there’s plenty of headroom at the substation, and the field geology looks like it could very much be uprated using EGS approaches. We were happy to see the announcement from Sage and Ormat, who will do this at Blue Mountain, which came in at no.16 on our list of candidate fields.

But nuclear uprating is a favorite, and in our analysis can provide 2GW+ over the next few years. It’s not as fast as some of the other options, but given the firmness, scale, and that it still fits in the pre-2030 window – ahead of many gas turbine wait times – it just makes so much sense, if you have the means. If there are ways to reduce or sidestep NRC approval, or fast track interconnection, it definitely needs to be on the table.

Excerpt from The Fastest MW section on Nuclear: Uprates

The limiter, as pointed out above is cost, as not everyone has $3-4bn lying around. But it makes sense for this hyperscaler / utility cohort that has those balance sheets (with help from Uncle Sam, in some cases) to go for it, even with the handsome speed premium attached.

Who this helps

Constellation and Vistra. Both get to turn aging merchant plants into long-term contracted assets. Twenty-year PPAs give them the revenue certainty to relicense, Constellation can fund more than $7bn of upgrades off the back of Amazon and Google’s commitments, and Vistra gets cheaper federal debt.

Hyperscalers with deep pockets. Google, Amazon, and Meta each lock in clean firm power in PJM for 15-20 years, much of it at plants that already run, and settle outside a capacity auction that keeps clearing at its cap. 

Who should be nervous

Data center developers without a $3bn checkbook. The cheapest clean firm power in PJM is getting locked up for decades by the three or four buyers who can fund the upgrades. Everyone else is left with the capacity auction, gas turbine queues stretching into the 2030s, or behind-the-meter workarounds.

SMR developers chasing the same customers. At under $5,000/kW, against roughly $12,500/kW for the first BWRX-300s at Darlington, uprates are the cheaper way for a hyperscaler to buy firm nuclear this decade.

A quick read on the numbers shaping the market. The capex, the contracts, the regs, all anchored in the so-what.

80% // MI regulators approved 80% min. billing for DTE’s Google data center contracts, above the 50-65% range in the utility’s standard large-load rate. The 20-year term requires 15 years of minimum charges if Google exits early, posted collateral against the risk, and fully paying for up to 1,600MW of new renewables and 480MW of storage. Currence data shows more than 50 Michigan townships and cities have passed data center moratoriums this year, even though state law bars them from banning the projects outright.

5.8GWh // Antora and Pratt Energy are building a multi-day thermal battery project at Pratt’s Kansas biorefinery, backed by Copenhagen Infrastructure Partners, Grok Ventures, and University Pension Plan Ontario. It’s Antora’s second major deployment this year after its 5GWh Big Stone, South Dakota project, and infrastructure fund capital showing up twice in a year continues the momentum.

$4.5bn // Investor-owned utilities filed rate increase requests in Q3, a quarterly record that brings the nine-month total to $23.1bn. The split between data centers and wildfire or storm costs isn’t out yet, but it lands as ratepayer protection bills hit Congress and a wave of state large-load rules moved forward, so expect commissions to cite these totals when they decide how much data center cost to push onto other customers.

$225m // Chesapeake Utilities lined up a six-bank syndicate for a new ATM equity program. Barclays, Citizens JMP, Ladenburg Thalmann, Morgan Stanley, RBC, and TD all came in with proceeds earmarked for capex, paying down revolver debt, and acquisitions. It’s small change next to this year’s equity raises (NextEra’s $2bn equity units, AEP’s $2.6bn follow-on plus forward, Constellation’s $3.1bn secondary, RWE’s roughly €4bn raise with QIA and Norges Bank cornerstoning it) but it’s the same utility-equity playbook running.

Explore more Signals on Currence here.

The policies, rulings, and company moves worth watching.

A new Texas coalition lines up the supply chain against data center opposition. The Lone Star Infrastructure Coalition counts Blackstone’s QTS, NextEra, Siemens Energy, Quanta, the Nuclear Energy Institute, and Lancium as members, and plans to take the economic case to residents, mayors, and county officials as ERCOT works toward a December deadline to finish Gov. Abbott’s audit of self-powered projects.

Portland General becomes the second utility live in CAISO’s Extended Day-Ahead Market. EDAM lets neighboring utilities tap into California’s day-ahead dispatch instead of running their own, and PGE’s interties with CAISO and PacifiCorp convert into shared transfer points under the market’s transition rules. We already covered how Bonneville, which sells nearly a third of the Northwest’s power, is deciding between EDAM and the rival Markets+ platform by year-end, and how PGE’s first settlements go will help shape that call.

Pennsylvania’s PUC advances a rule barring “critical load” status for data centers. The commission’s tentative order would bar utilities from shielding new large loads from being shed first in a grid emergency unless they’ve secured their own capacity.

Virginia’s Lt. Gov. breaks with the NextEra-Dominion merger, saying the $67bn takeover isn’t in Virginians’ best interest, days after AG Jay Jones asked the SCC to reset the review clock over the companies’ 14 Sept. amended filing, doubled bill credits, and a new NextEra office tower, which already pushed the evidentiary hearing back roughly two months.

The latest research, features, and data drops on the Currence platform.

Overheard at Yotta 2026. Our team spent three days at the industry’s biggest trade show, now pulling 7,000 attendees, and came away with one throughline: demand for AI infrastructure is real, but a lot of what’s moving through the market is optionality dressed up as commitment on both sides of the meter. Developers told us about deals that looked signed until the moment they asked for a signature, and social license, not power, is emerging as the bottleneck capital can’t buy its way past. Read the full recap here.

Geothermal Market Update, Q3 2026. This report scores next-gen geothermal against the seven attributes we use to define bankability, whether a project can raise capital on terms that let it scale. EGS is advancing fast and its contracted prices are closing in on the sub-$100/MWh mark, but it’s still a one-company market. Read the full outlook here.

Where the market is meeting, and where to find us

📅 TechCrunch Disrupt // San Francisco, Oct 13-15 // The startup world's biggest fall gathering, with tracks on AI, climate tech, and infrastructure. We’ll be speaking, so give us a shout if you’ll be there!

📅 Norrsken Impact/Week // Barcelona, Oct 14-15 // Founders, funders, and policymakers gathering on climate, energy, and impact investing. 

📅 RE+ // Las Vegas, Nov 16-19 // North America's largest clean energy event, covering solar, storage, hydrogen, and grid edge tech.

This week is the niche-iest of niche-y. No real reason or deep meaning this week, but this is for those of us who used to hang out at the DC rock clubs back in the 2010-ish era. Maybe I’m just feeling nostalgic, but think the Black Cat, DC9, the Rock & Roll Hotel, and even 9:30. So, one of my favorite shows from that era was Daddy Lion, a local from GWU. So, here they are with Appearance.

But, tbh, the absolute best show I ever saw in DC during those years, no question, hands down – The Dears. Went in never having heard of them, left a lifelong fan. Here’s one from them, too.

Full Powerstack B-Sides on Spotify.

Interested in diving deeper? Talk to our team and learn how the teams that finance and build the energy system use Currence to stay ahead in the energy and AI buildout, including Southern Company, Tokyo Gas, Jefferies, Galvanize, B Capital, and others.