It took some 20,000 hours of modelling time on Alps, the CSCS supercomputer in Lugano, and 551 pages to reach one number per future: how many terawatt-hours Switzerland would have to import in winter. The study, Neue Nukleartechnologien – Eine Auslegeordnung für das Schweizer Energiesystem, was published by TA-SWISS on 27 October. It was led by Peter Burgherr at PSI, with Annalisa Manera (ETH Zürich) and Isabelle Stadelmann-Steffen (Universität Bern) among its authors. It is an instrument, not an oracle. The Kurzfassung puts it in five words: «Modellierungen sind keine Prognosen.» Three coupled models ask a least-cost what-if question out to 2070: one for the macro-economy, one for the Swiss energy system and one for life-cycle analysis. The team spent sixteen months, from April 2025 to July 2026, building that shared model, and they published it with a Zenodo DOI so anyone can check the arithmetic. That openness is the real contribution.

Why winter, physically

At 47.4° N, Zürich’s noon sun at the December solstice stands about 19° above the horizon. That is the Earth’s 23.4° tilt working against a city far from the equator, and it leaves roughly eight and a half hours of day. Solar fills the summer. Winter is when heat pumps draw hardest.

The study’s winter imports for 2050:

• «Offene Märkte»: just under 12 TWh without a new reactor, 8.5 TWh with early large reactors and 7.8 TWh with SMRs. At high construction costs, the import need falls «nur unwesentlich».
• «Autarkie», no reactor: 3.7 TWh, and 3.6 TWh in 2070. Both are under the 5 TWh winter-import ceiling in Art. 1 of the Stromversorgungsgesetz.
• «Weiter wie bisher»: solar grows by 17 TWh by 2050, and imports persist.

The reactor as an object

The modelled Gen III/III+ unit is 1,200 MWe, which the Kurzfassung describes as enough for three cities the size of Zürich. Its assumed cost is 9,500 CHF/kWe for a 2030 start, or 8,000 CHF/kWe for a 2040 start. The model assumes a ten-year build. It also notes that French, Finnish and UK builds since 1999 took two to three times their planned five to six years. Earliest operation would be «frühestens ab den späten 2040er-Jahren». Burgherr told SRF: «Wenn wir ein neues AKW erst 2050 oder später bauen, sind andere Technologien bereits weiterentwickelt worden.»

The cheapest move in the model is the oldest hardware. Running Gösgen and Leibstadt to 70 years cuts 2020–2070 system cost by 7–12 billion CHF, and PSI calls long-term operation currently the most cost-efficient option. The trade-off is physical. That cheap winter terawatt-hour depends on keeping pressure vessels commissioned in 1979 and 1984 qualified for decades more. The weight falls on inspectors and operating crews whose names never reach a headline.

←TODAY: Swiss demand of about 56 TWh/yr is modelled to reach 70–90 TWh by 2050, driven by heat pumps, e-mobility and data centres.
→3012: Every building in Zürich files its October–March balance the way it files its statics.
Fulcrum: The gap between 12 and 3.7 TWh is not only reactors. A large part of it is how buildings heat, store and charge in the dark months.

On the desk

That demand curve is a building curve. Each heat pump that replaces an oil boiler moves load into the months when the roof yields least. A Minergie certificate gives you the annual figure. Very few drawing sets give the winter one.

Atelier: A Zürich studio already runs SIA 380/1 for every Vorprojekt, yet the grid will care far more about the six cold months than about the annual total. The Monday move: add one line to the office’s energy-concept template, Oct–Mar demand in kWh against Oct–Mar on-site yield, and make it mandatory from the next project onward.

Hack: Turn meter readings into a winter ratio, the one number that says how hard your building leans on the grid when the sun is lowest. In an existing building, read the heat-pump meter month by month. In a project, feed the PV list from a Ladybug radiation study (Ladybug Tools, built by Mostapha Sadeghipour Roudsari and its open community) inside a GhPython component. The values below are placeholders. A ratio near 1 means the roof carries its own winter; a ratio of 3.75 means the grid carries most of it.

hp = [1450, 2300, 3100, 3300, 2700, 2000]  # kWh heat pump, Oct..Mar (placeholder)
pv = [ 900,  520,  310,  380,  700, 1150]  # kWh roof PV, Oct..Mar (placeholder)
ratio = sum(hp) / sum(pv)
print(f"winter draw {sum(hp)} kWh / yield {sum(pv)} kWh = {ratio:.2f}")

The ballot

On 18 June 2026, Parliament lifted the 2017 new-build ban as an indirect counter-proposal to the «Blackout stoppen» initiative. A referendum followed, and the vote is set for 28 February 2027. Under current law, the study counts three national Yes votes before any reactor stands. Burgherr summed it up: «Nur wenn alles zusammenkommt, wird ein neues Kernkraftwerk gebaut werden.» Stadelmann-Steffen described the electorate on SRF: «Die Hälfte ist dafür, die Hälfte ist dagegen.»

TA-SWISS states that the study speaks neither for new plants nor for keeping the ban. The Schweizerische Energiestiftung reads it differently. As reported by AWP via cash.ch, SES says the study assumes shorter builds and lower costs, omits a severe accident, and is unsuitable as a basis for political decisions. SES president Nadine Masshardt added: «Die Atomlobby will uns eine Technologie aus dem letzten Jahrhundert als Zukunft verkaufen.» Those are SES’s words. The numbers above are the study’s.

From the late 2070s, the lesson I would pass back is to carry the scenario name with every terawatt-hour, the way you carry the error bar with every measurement. Then open your current project, find its October–March kWh, and write the ratio on the drawing set.

Sources & Further Reading

• Primary: TA-SWISS — Novel nuclear technologies (study and Kurzfassung)
• Reinforcing: PSI — New nuclear technologies: analysing options for the Swiss energy system