# Demand and 2035 supply checks

WattInfo, 18 September 2026. These are independent calculations using archived official data and explicitly stated assumptions. No probability of a shortage or final procurement margin is assigned.

## Weather coverage and supply

The base comparison uses all thirteen complete hydrological years available in the prepared weather/hydro input: October 2011–September 2012 through October 2023–September 2024. Each year has two fixed-charging runs:

- Routine operation: Gösgen 1.010 GW and Leibstadt 1.233 GW net; each has 30 assumed scheduled maintenance days and 98% availability outside those days.
- Prolonged outage: the same assumptions, but Leibstadt is absent for the first 90 calendar days of the ending year. This exceptional event has no assigned probability.

Both use central-model 2035 demand, 26 GWp PV, finite hydro and up to 2 GW imports. No future Beznau, new nuclear, added hydro projects, V2G, new stationary battery or fuel reserve is credited. The solar shape uses an eight-station radiation proxy scaled to 1,100 kWh/kWp in the 2024 reference year. Snow, new geographic siting, inverter choices and the distribution grid are unresolved. Thermal/wind supply retains the 2024 monthly totals with a 98% delivery assumption.

Natural reservoirs are pooled separately from four finite pumped stores. Monthly river generation and derived reservoir inflow correspond to each weather year. These are smoothed inputs, with no lake cascade, local waterway or internal-grid representation. Cyclic starting stocks are chosen with perfect foresight. The model first minimizes missing annual energy, then peak missing power, then imports and cycling. Different dispatches can share the same objectives. The monthly hydro and optimized stocks can understate real operational difficulty.

Annual and winter gaps, peaks, imports, exports, hydro flows and day/night partitions are retained for every case. Winter means October–March. Daylight uses the existing NOAA-based reference-location classification. The thirteen weather outcomes are neither actual historical shortages nor an independent random sample from which to calculate outage probability. Historical weather is a set of test conditions, not a climate projection for 2035 or 2050.

The base energy and power screen rounds upward to 0.1 TWh/0.1 GW the largest winter gap and annual hourly power gap across the thirteen prolonged-outage cases. Extra summer delivery is reported separately. A contract must specify the delivery profile, usable fuel, outages, legal operation and margin; buying the two headline numbers alone does not establish reliability.

The proposed initial supply-offer brief also includes the largest requirements from the three extra data-centre cases. It rounds upward to 0.1 TWh and 0.1 GW across the thirteen base outage cases and those three growth checks. This does not establish adequacy for all growth/weather combinations or supply procurement margins.

## Import and growth sensitivities

For 2011/12, 2016/17 and 2021/22 weather, compare the same prolonged-outage case with zero imports, or with an extra data-centre load. The selected years include the cold-peak example, the largest base winter-energy gap and the earlier dry-year comparison. They are sensitivity cases, not a complete joint uncertainty distribution.

The data-centre case uses the report's central 3.2 TWh total for 2030 and assumes it remains constant through 2035. The 2024 estimate of 2.08 TWh is assumed to be within retained service-sector electricity. Its embedded 2035 component grows by the same interpolated activity and intensity as that sector. Only the difference between 3.2 TWh and that embedded component is added. The added load is uniform within each hydrological year and network losses are applied once. Further post-2030 growth is unestimated.

Both all-services and all-industry mappings are provided for annual demand checks at 2030, 2035 and 2050. These are alternative accounting approximations, not established sector weights or a confidence interval; cooling and other end-use allocation remains unresolved. Small server rooms outside the study scope stay in the rest of the baseline. The high 3.5 TWh study case can arrive after 2030. Later constant-demand cases are analyst assumptions, not published forecasts.

## Population and recorded-demand comparisons

The population sensitivity changes only the original central model's 2050 population from 10.5 to 10.3065125 million, the BFE workbook's annual-average input. It leaves the rest of the model unchanged. This affects household demand through the existing population relationship, not industrial or data-centre activity.

The 2025 comparison evaluates the original central trajectory at 2025 and compares its final electricity with BFE workbook T25!H71. BFE's net generation already removes pumping; country consumption minus network losses gives final consumption. The model retains its 2024 weather basis and assumed equipment replacement. It was built after 2025, so the comparison is not a prospective forecast validation. Annual differences can reflect weather, activity, timing and compensating model errors.

## Heat-storage evidence

Høje Taastrup's measured 2024 thermal balance is 34,044 MWh charged = 29,917 delivered + 3,598 lost + 529 increase in inventory. The reported efficiency convention uses delivered heat divided by input less that inventory increase. Its 3,300 MWh capacity is heat capacity calculated from the storage temperature window. Repeated short-term district-heat dispatch does not establish a seasonal electricity service.

The Danish cost table records €12.04 million full-project investment, €1.34 million grants, and €10.70 million after grants, at implementation-year prices. Modelled annual income and estimated OPEX do not establish audited profitability. Drake Landing's municipal account illustrates the need for funded maintenance and adequate customer income. No Swiss national winter-electricity saving is credited from these cases.

## Reproduction

Use Python with the pinned packages in model_extensions/requirements.txt. Run `python -m model_extensions.research`, then `python -m model_extensions.validate`, from the repository or reproduction bundle root. The generator verifies evidence and produces the catalogue/public exports. The bundle contains code, prepared model inputs and all hourly outputs. Third-party PDFs/workbooks remain in the private archive; the demand source-cell audit needs the catalogued BFE 7519 workbook at the manifest path. Public publisher links identify original documents; current downloads may differ from the frozen edition.

Input hashes, software versions, all case assumptions and hourly conservation checks are recorded. Public output checks validate structure, arithmetic and archive integrity. They do not independently validate Swiss plant operating rights, owner economics or statistical reliability.