At 3 AM, a solar farm delivers 0 MW. ☀️ That is physics, not a knock on solar, but it frames the whole argument about clean power. The real test of a grid is the dead hours, when demand is low but never zero and hospitals, subway pumps, and server halls all keep drawing.
In the United States, that hour runs on nuclear, gas, wind, hydro, and a fast-growing battery fleet. Small modular reactors are not on that list yet. 🔌 The first BWRX-300 in Ontario and the first Natrium plant in Wyoming are under construction, with first power targeted around 2030. This article scores solar, wind, and SMRs on the one metric that counts at 3 AM: megawatts delivered when nobody is watching.
Capacity factors expose the 3 AM problem
Capacity factor is the share of a plant’s maximum output that it actually delivers over a year. EIA’s 2025 figures, compiled in a SUN DAY Campaign review of EIA data, are blunt:
Nuclear: 91.0%
Natural gas: 58.4%
Wind: 34.2%
Utility-scale solar PV: 24.4%
Those are annual averages, so they smooth over the daily shape of output. Solar’s 24.4% works out to roughly 12 hours of usable power and 12 hours of nothing. 🌙 Nuclear’s 91.0% is close to flat around the clock, which is why engineers call it baseload. Developers design SMRs for the same steady output. No Western SMR has logged a full year of operating data yet, so for now that number is a design target and not a track record. ⚛️
Batteries narrow the gap, then run out of hours
Batteries are the strongest rebuttal to the 3 AM argument, and the numbers are not small. 🔋 Amperon’s mid-2026 comparison of ERCOT and CAISO shows how fast the fleets grew:
ERCOT (Texas): 20,948 MW of batteries, about 23% of the record 91,089 MW peak set in July 2026
CAISO (California): 17,023 MW, about 33% of system peak
CAISO average duration: roughly 3.8 hours at the end of 2025, per Modo Energy
ERCOT merchant default: two hours, according to the same Modo analysis
The catch is duration. A four-hour battery that starts discharging at 7 PM is empty by 11 PM, four hours before the 3 AM test begins. GridStatus data shows CAISO storage discharging from dusk to dawn on average in April, but the same analysis credits mild temperatures and muted demand. A cold, still January night is a different exam. 🥶
Wind is supposed to cover that night, and often does. Sometimes it does not. During a Dunkelflaute in early November 2024, Clean Energy Wire reports renewables supplied 30% of German public electricity from November 4 to 10, while fossil fuels covered the other 70%. 🌬️ Germany’s weather service counts about two cases a year when combined wind and solar output falls below 10% of nominal for 48 hours. That is 2024 data, but the pattern is seasonal and old, not a fluke. Grid operators reading this are welcome to say in the replies whether four hours of storage is enough where they work.
Where the SMRs actually stand
Only two SMR fleets run commercially today, in Russia and China. Every Western project is still a construction site or a permit file. 🏗️ The status of the four that matter, as of September 20, 2026:
OPG Darlington, Ontario: GE Vernova Hitachi’s 300 MWe BWRX-300, first of four planned units. The CNSC issued the construction licence in April 2025, lifted the first regulatory hold point on March 30, 2026, and received OPG’s operating licence application on March 25, 2026. The four-unit budget is C$20.9 billion, and the first unit is scheduled by the end of the decade.
TerraPower Kemmerer, Wyoming: A 345 MWe sodium-cooled fast reactor with molten salt storage that lifts output to 500 MW. The NRC voted the construction permit on March 4, and TerraPower started nuclear construction on April 23, 2026.
X-energy and Dow, Seadrift, Texas: Four 80 MWe Xe-100 modules, 320 MWe total. The NRC finished its environmental assessment in May, and X-energy’s 10-Q filing expects the construction permit in the first quarter of 2027 and first commercial delivery in the early 2030s.
CNNC Linglong One, Hainan: The 125 MWe ACP100. Construction started in July 2021, and World Nuclear News still lists it in pre-commissioning this month, so the first-half 2026 commercial target has passed.
Natrium is the most relevant design for a 3 AM debate. Its molten salt tank does inside the plant what batteries do for solar, letting output rise for the evening ramp. 📅 Every project, deal, company, and regulatory event we cover is logged, verified, and searchable in SMRbrief Pro. For readers new to why sodium, gas, salt, and water reactors behave so differently, SMRbrief’s breakdown of the five SMR reactor types covers the differences.
The price of certainty
Lazard’s 2026 report, summarized by Utility Dive, puts unsubsidized costs like this: 💰
Utility-scale solar: $40 to $98 per MWh
Onshore wind: $37 to $99 per MWh
Nuclear (new-build, not SMR-specific): $175 to $255 per MWh
That gap is real, and it is wide. Lazard also runs a firming analysis that adds storage equal to 50% of solar or wind capacity, four hours long. Four hours, again, does not reach 3 AM. 📉
The nuclear side has its own uncomfortable arithmetic. WNN puts Darlington’s four units at US$15 billion for 1,200 MWe, which divides out to about $12,500 per kilowatt, shared infrastructure included. Natrium’s reported cost of up to $4 billion for 345 MWe is roughly $11,600 per kilowatt. Neither number looks like the cheap, factory-built promise, and that deserves skepticism. Which figure should a city council weigh more, the $40 low end for solar or the 91.0% capacity factor for nuclear? Readers can make their case in the comments. ⚖️
The 3 AM verdict
Today, in North America, solar reaches 3 AM only through batteries, wind reaches it when the weather cooperates, and SMRs do not reach it at all. The steady overnight supply comes from existing large reactors, gas, hydro, and storage. ✅ Around 2030, if schedules hold, Darlington and Kemmerer add 645 MW of firm, round-the-clock capacity. That is meaningful for a region, but small next to the batteries already on the grid.
The honest reading is that the three are not rivals so much as different shifts. Solar and wind are the cheapest daytime energy, batteries cover the evening, and SMRs are the bet on the hours after that. 🌃 Four dates decide how good that bet looks:
The CNSC hearing on Darlington’s operating licence, date to be announced
The NRC’s final safety evaluation for Long Mott, targeted for November 2026
Linglong One’s first criticality and grid connection
Kemmerer construction progress toward its 2030 completion date
If the 3 AM test is the standard, which of these three would a city planner still be comfortable betting on in 2031? 🔭