On one major count, solar out-generated nuclear worldwide in 2025. The quieter crossing matters more: batteries are turning a three-hour midday product into an all-day one, and the scarcest commodity on the grid is now the 8 p.m. hour.
| Feature: The Sun Learns to Stay Up Late | 3 |
| Breakthroughs | 4 |
| So What? + Field Notes | 5 |
| Against the Grain | 6 |
| Signals | 7 |
| By the Numbers | 8 |
| The Long View + Sources | 9 |
| Next Issue | 10 |
Every industry has a unit it was taught to count, and the unit quietly decides what it can see. Electricity was taught to count kilowatt-hours. A kilowatt-hour at noon and a kilowatt-hour at 8 p.m. are the same product on the scoreboard and wildly different products on the grid.
This week the scoreboard moved. By one authoritative tally, solar generated more electricity than nuclear worldwide in 2025; by another, it came within a whisker. Either way the headline is true and the lesson is elsewhere. Solar met 75% of the growth in global demand, and the next fight is over the hours after sunset.
That is this issue's lens: a commodity is whatever the market forgot to timestamp. Read the feature, then the contrarian column, and decide whether the battery is solar's accessory or its entire business model.
— The Editors
Each Friday: one deep dive, one contrarian read, three signals from other fields, and the numbers worth remembering. Every figure traces to a source on page 9. Where a result is interim, a forecast or a theoretical ceiling, we say so.
Solar has beaten nuclear on the world's scoreboard. The real contest is whether it can sell dinner as well as lunch.
Picture a restaurant with the best kitchen in the world that opens for exactly three hours a day, at lunch. It is cheap, it is excellent, and it is packed. Now picture the owner discovering a walk-in freezer that lets the kitchen cook at noon and serve until midnight. That, in one image, is the story of solar power in 2026: the cooking was solved years ago, and the freezer has just become affordable.
Ember's Global Electricity Review puts 2025 solar output at 2,778 TWh, a record increase of 636 TWh, or 30%. The World Nuclear Industry Status Report counts nuclear at 2,703 TWh, and finds its share of global electricity at 8.8%, the lowest in about 45 years and roughly half the 1996 peak of 17.5%.
Fossil generation did something it had not done for structural reasons before: it slipped, by 38 TWh (-0.2%), while demand rose 2.8%. Solar alone met 75% of the net increase in demand.
Solar is a time-of-day business. Ember's analysis of an average day in the first half of 2026 finds solar supplying about 25% of global demand between 11 a.m. and 2 p.m., and near zero between 8 p.m. and 5 a.m. In mature markets it covers about half of midday demand.
Grid operators do not buy terawatt-hours. They buy kilowatts at a particular minute. A midday glut is worth less every year it grows; an evening shortfall is worth more. The question for the industry is no longer how to make more sunshine-power, but how to move it.
Batteries are the mover, and this week's data shows how quickly the move is happening. Turn the page.
In the first half of 2026, Chile's solar supplied 71% of electricity at noon and 26% across the whole period. The Netherlands reached 58% at 1 p.m. and Germany 55% at noon. Midday is no longer a frontier; it is saturation.
In California, solar plus batteries now supplies more than 25% of demand between 7 and 9 p.m., up from 6.8% in the first half of 2023. Bulgaria reached 24% in the same window. The hour that used to belong to gas is now contested.
The global average installed cost of batteries fell 95%, from $2,634 per kWh in 2010 to $140 in 2025 (IRENA). BNEF expects 459 GWh of battery additions in 2026, up 50% from 307 GWh. In 2025, 74% of batteries were used for energy shifting, up from 47% in 2020.
Ember estimates batteries could theoretically shift 34% of new daily solar output in 2026, versus 18% in 2025 and 4% in 2021. Bulgaria (77%), Chile (76%) and Australia (60%) already have enough storage to move most of what they add.
Utilisation is the quiet frontier. China's standalone batteries ran 299 cycles a year in 2025, up from 146 in 2022, against a best-practice benchmark near 350. The hardware is arriving faster than the habits to work it.
This is not a laboratory story. The technology is shipping, which makes the strategic question a market-structure one: who captures the value of the hour?
Developers who sell solar alone sell into the cheapest hours and a falling price. Those who sell solar plus storage sell a shaped product, the evening included. Ember notes that co-located solar and batteries are already about a quarter of new utility-scale solar (IRENA).
Grid operators see the shape change first. India's fossil fleet fell to 125 GW at 1 p.m., down about 10 GW from 2023, but still ran 174 GW at the 7 p.m. peak. The EU's evening fossil peak is 101 GW, down only 5 GW. The evening is where the remaining fossil megawatts hide.
Big electricity users are the least-discussed beneficiaries. A load that can shift, whether cooling, charging, industrial heat or batch computing, can be scheduled toward the cheap sunny hours. Ember suggests smart EV charging alone could absorb about 8% of peak hourly wind and utility solar in the Netherlands by 2030.
The honest timeline: the midday is solved now; the evening is being bought in leading markets this decade; the long, dark hours of winter are a different problem and a different business.
1 · Harvest. Panels run flat out at midday, when power is abundant and cheap.
2 · Store. Batteries charge on the surplus that would otherwise be curtailed or sold at a low price.
3 · Release. At 7 to 9 p.m., when demand peaks and panels are dark, the batteries discharge into the grid.
Capacity factor. The share of a plant's maximum output it actually delivers over a year.
Duck curve. The dip in net demand at midday, then the steep evening ramp, once solar is large.
Battery cycle. One full charge and discharge; cycles per year measure how hard an asset works.
Shiftable solar. Ember's measure of how much new daily solar output batteries could move to later hours.
The consensus asks whether solar can beat nuclear. The better question is who will own the hours after sunset.
Treat the battery as the product and the panel as its feedstock. A panel sells a commodity that is cheapest exactly when everyone else's panels are also running. A battery sells the one thing the sun cannot, the evening, and it captures the spread between a glutted noon and a scarce 8 p.m.
On that reading, the terawatt-hour crossover is a vanity metric. Nuclear's real rival is not solar; it is solar plus a few hours of storage. And the leverage for everyone who runs flexible load, from data centres to cold stores, is in software: schedule work toward the sun and the price follows. That is our inference, not a finding of the sources.
The scoreboard is contested. Ember says solar was “nearly drawing level” with nuclear for all of 2025 while beating it from April to September; the nuclear status report, whose tallies are produced by a sceptic of the industry, says solar passed it for the year.
The ceiling is theoretical. The 34% of new solar that batteries could shift is a ceiling, not a delivery. Solar is still near zero from 8 p.m. to 5 a.m.
Growth is not guaranteed. EU solar installations fell 0.7% to 65.1 GW in 2025, the first decline in a decade.
Daily storage is not seasonal storage. Moving noon to evening is solved at scale; moving July to January is not. Firm power still has a job.
So the bet is real but bounded. If it pays, it will look less like a victory for solar and more like the birth of a time-of-day market.
CERN and Mistral AI signed a statement of intent to build “Flare,” billed as the first foundation model for collider physics, trained on billions of simulated particle collisions the way language models train on text. CERN holds about an exabyte of Large Hadron Collider data and expects collision volume to rise tenfold after its upgrade, around 2030. No timeline or budget was disclosed.
RIKEN researchers drove up to three silicon spin qubits at once through a single shared microwave line, holding single-qubit gate fidelity above 99.99%. With all five qubits of the array running in parallel, fidelity was about 99.9%. The catch: this covers single-qubit gates only, and higher drive power still causes decoherence that the new protocol does not fix.
In GIMEMA ALL2820, a phase III trial of 236 adults with newly diagnosed Philadelphia-positive leukaemia, a chemotherapy-free regimen (ponatinib then blinatumomab) reached 87% event-free survival against 71% for imatinib plus chemotherapy, at a median follow-up of 23 months. Complete remission was 94% versus 79%. These are interim results presented at ASH 2025, not a final peer-reviewed analysis.
Global solar generation in 2025, up from 2,143 TWh in 2024 (Ember)
Record one-year rise in solar output, 30% growth (Ember)
Nuclear's share of world electricity, lowest in about 45 years (WNISR)
Share of the rise in global electricity demand met by solar alone (Ember)
Installed battery cost in 2025, down 95% from $2,634 in 2010 (IRENA)
Share of California's 7 to 9 p.m. demand met by solar plus batteries, from 6.8% in 2023 (Ember)
Share of new daily solar that batteries could theoretically shift, from 4% in 2021 (Ember)
This week's three signals share a habit with the feature. Physics is handing its exabyte to a model because the bottleneck is no longer collecting collisions but reading them. Silicon qubits are climbing because the trick is not one perfect qubit but many, run at once. A leukaemia regimen drops chemotherapy by changing the order and the targeting of treatment. In each case the ingredients were not new; the sequencing was.
Solar is the clearest example. We spent twenty years making sunlight cheap and are now discovering that cheap sunlight at the wrong hour is worth less than expensive light at the right one. The next decade's winners will be those who sell time: the stored, the shifted, the scheduled.
The question to carry out of this issue: which of your own costs is a timing problem you have been treating as a quantity problem?
Which of your costs is a timing problem you have been treating as a quantity problem?
Researched, written & designed with Claude. Typeset in Poppins & Lora on the Anthropic palette.