Changelog

Changes to calculations, parameters, data and starting assumptions. Wording and design changes are not listed.

• Oct 31, 2026. Packages 3.7.0. No water or power estimate changed. Utilities reporting no sales are no longer compared. Virginia. Lexington's rainfall is now Rockbridge County's, 44.79 inches, the county around it, since NOAA has no record for the city. It had the state's median.
• Oct 31, 2026. Data refresh, packages 3.6.2. No water or power estimate changed. Virginia. The 38 independent cities had no rainfall or design storms of their own. All 38 now have design storms and 37 have rainfall; Richmond's 100-year, 24-hour storm is 8.43 inches. Lexington still uses the state's median rainfall. Texas. Reported water use moved slightly in Hopkins, Rusk and San Patricio counties, from TWDB's current report. Rounding. Electricity in words and the ranges on the methods pages follow two significant figures, and evaporation is rounded to the same place as water withdrawn wherever it appears.
• Oct 31, 2026. Numbers the pages kept on their own now come from the open packages, and places where pages disagreed now agree. No model estimates change. Storms on the detailed calculator use curve numbers (NRCS TR-55), not the Simple Method scaled up: on 200 acres half under roofs and paving, a 1-inch storm sheds about 2.1 million gallons instead of 2.7, and a 2.57-inch storm about 7.2 instead of 7.0. Sewer plants. Both days are shown, and the level follows the hottest day on both pages: low under 10% of the plant's spare room, then a big share, then could fill it above 50%. It followed the average day before, and the detailed page had no top level. Wells. Both pages say a permit applies in Zone C from 1 to 2 million gallons a day. Homes. Michigan's 4,538,056 residential customers (EIA 2025) instead of 4,516,367 (2024). Ohio, Virginia and Texas now show the share of homes too. Proposal sizes. The size buttons are worked out from the contracts and current PUE: Saline Township about 1,090 MW instead of 1,130, Van Buren Township about 790 instead of 830. A place's own sewer plant. One rule in every state: a plant carrying the place's name goes first if it is within 10 km, otherwise the nearest plant. Michigan had put a name match first at any distance, which sent some places to a plant far away that only shared a name. Rain before a place is chosen is the state's median county, 33.86 inches in Michigan instead of 33.9.
• Oct 31, 2026. Data refresh. Rogers City's main watershed is now Lone Lake-Ocqueoc, where the city's center lies in the current boundaries, not Lake Huron. Water system shares moved by 0.01 for Bangor and Cedar Springs. No water or power estimate changed.
• Model 1.6, Nov 2026. For 300 MW in Michigan, regional medians. Middle estimate. Wastewater is now at the model's own median and water lost is what is left; wastewater was the remainder, and ran 13 to 23% above its median in the evaporating designs. Cooling towers: wastewater 0.38 to 0.34 million gallons a day, water lost 1.51 to 1.55, hottest-day wastewater 0.50 to 0.45. Hybrid: wastewater 0.024 to 0.020, hottest-day 0.34 to 0.29. Adiabatic pre-cooling: wastewater 0.019 to 0.018, hottest-day 0.24 to 0.22. Water taken moves under 1%. Electricity. A tower's fans slow to the air the hour's heat needs instead of running at design air whenever it evaporates: cooling towers 217 to 214 MW on average, and closed loop now uses about 1% more than towers on average, not about 1% less. Hottest-hour electricity moves under 1%. Power losses and lighting count as heat in the room's air. Septic. Condensate is assumed to go to a site drain or stormwater [A], not the drainfield: 7,100 to 7,900 gallons a day to a closed loop's septic system becomes 6,200 to 6,300 at 300 MW, and the state permit threshold moves back from about 400 MW to about 500 to 600. Lake-source. Homes on the peak day count what the lake loses, not the water passing through. Water system. A place is compared with a system covering at least half of it, otherwise the largest by population covering at least 5%, otherwise the largest by population [A]; it was the largest by population at any overlap. 27 of the 470 Michigan places with a mapped system change, such as Clawson (its own system, not Royal Oak's). Very small water systems. Where the main system has under 100 connections and covers under half of the place, no percent is shown; 13 Michigan places. Utilities. A reported peak below a utility's average load counts as missing. Assessed. The watershed table says whether the state assessed each watershed.
• Model 1.5, Nov 2026. The cooling model now closes on energy and water in every hour. Water taken moves by up to about 4%, down in most places, and wastewater by up to about 16% (cooling towers about 7% down). Closed loop's small figures move most at 300 MW: peak-day water from about 7,000 to 8,700 gallons in Michigan, and peak-day wastewater from about 6,200 to 28,000, from condensate. Electricity moves by under 2%. Heat counts pump heat and outside air's heat and moisture, in place of a fixed sensible heat ratio. Outside air of 100 to 1,500 cfm per MW [A] adds condensate to the wastewater in humid air and humidifier water, at a share sampled from none to all [A], in very dry air. A hybrid's chillers reject heat by the wet share, and its drift follows the water through the wet part. Blowdown comes from the dissolved solids balance, so drift is no longer counted twice. A tower's air flow is capped by what air can carry. Lake-source cooling has its own electricity, with a heat factor of 1.05 to 1.17 (was 1.03 to 1.25), instead of cooling towers standing in. Water at power plants counts 5% line losses (EIA). Farmland sediment keeps the paved part of the land.
• Model 1.4, Oct 30, 2026. Electricity worked out hour by hour from local weather, in one heat balance with the cooling water, instead of one PUE per cooling design. For a 300 MW campus in West Michigan, average electricity: closed loop 216 MW instead of 222, dry coolers with adiabatic pre-cooling 214 instead of 222, hybrid 213 instead of 220, cooling towers 217 instead of 216. Peak: closed loop 387 MW instead of 372, dry coolers with pre-cooling 373 instead of 372, hybrid 361 instead of 351, cooling towers 360 instead of 348; the peak is now the hottest hour of a typical decade, not the top of a yearly range. The share of computers that are liquid-cooled is sampled from 0 to 100% [A]. Water. Evaporation now follows the heat rejected in that balance, replacing the heat factor. Water evaporated: cooling towers 0.4% less, hybrid 8.3% more, dry coolers with pre-cooling 0.7% less. Rounding. Water taken to two significant figures, water lost to the same place, wastewater the difference, so the numbers shown add up.
• Model 1.3, Oct 30, 2026. Parameters checked against primary sources for publication. For a 300 MW campus in West Michigan: closed loop now takes about 7,000 gallons a day instead of 2,700, hybrid about 74,000 instead of 97,000, and cooling towers about 1.9 million instead of 2.3 million; average electricity with closed loop is about 224 MW instead of 258 MW, or about 258,000 homes instead of 303,000. Utilization 50 to 80% instead of 60 to 90%, from measurements (EPRI 2026) and the IEA, which lowers electricity, cooling water and power-plant water about 13%. Water that is not for cooling. Staff water 10 to 30 gallons a person a day instead of 15 to 30. A new maintenance and cleaning term, up to 2,000 gallons a day plus 2 to 20 per MW, bracketed by applicants' projections, explains most of closed loop's rise. Loop volume 2,000 to 12,000 gallons per MW instead of 2,000 to 6,000. Cooling. Tower range 7 to 15 °F, drift 0.0005 to 0.02% of circulating water, the evaporation adjustment leaning low in cold weather (-0.10 to +0.05), economizer limits 5 to 12 and 14 to 22 °C wet-bulb, hybrid starting at 21 to 28 °C. Power plants. Michigan's mix 1.01 L/kWh evaporated (0.74 to 1.38) instead of 0.9, and 59 L/kWh withdrawn (38 to 90), from 2024 plant reports with missing and implausible ones filled and 2022 evidence on the same basis. Nuclear 396 gallons per MWh instead of 403; wind and solar about 0.2 instead of 1, by Michigan's mix of the two. Homes. 634.9 kWh a month (EIA 2025) instead of 618 (2024). Detailed calculator's assist design is now dry coolers with adiabatic pre-cooling, calculated from pad effectiveness, air conditions and fan slowing, instead of an assumed share of evaporation starting at Microsoft's 85 °F, which belongs to a different design; about 82,000 gallons a day at 300 MW instead of a few thousand.
• Model 1.2, Oct 28, 2026. Water that is not for cooling, power plants, and proposal sizes. Water that is not for cooling is now two parts instead of one base use of 0.001 to 0.008 gallons per kWh of computing with 10 to 50% of it lost. Staff water: people on site (20 to 60, plus 0.1 to 0.5 per MW of computing) times 15 to 30 gallons a person a day, about 10% of it lost [A]; it is added at the campus's size, since it grows with the staff, not the megawatts. Loop top-up: 1 to 5% a year of a loop holding 2,000 to 6,000 gallons per MW, all of it lost [A]. Humidification is no longer counted, since these designs recirculate indoor air. A 300 MW closed-loop campus in West Michigan now takes about 2,700 gallons a day (likely 1,700 to 4,400) instead of about 26,000, and loses about 340 to evaporation instead of 6,200; with hybrid cooling it takes about 97,000 instead of 120,000, and with cooling towers about 2.3 million instead of 2.4 million. The old closed-loop figure was four to seven times what Michigan applicants have projected; the new one is a little below those projections and about a tenth of the metered figure at the Switch data center in Gaines Township (see the checks on the methods page). At the starting sizes, closed loop's wastewater is now under 6,000 gallons a day, so a septic system falls to the county health department rather than an EGLE permit, and about a dozen places start on a different wastewater destination with hybrid or cooling towers. Power plants. The coal and nuclear choices now use Michigan's own plants instead of plants with cooling towers: coal 250 gallons per MWh instead of 687, since Michigan's large coal plants are cooled with water passing through once, and nuclear 403 instead of 672, weighting Cook (once-through) and Fermi 2 (cooling towers). For a 300 MW closed-loop campus, coal is now about 1.5 million gallons a day at the power plants instead of 4.2 million, and nuclear about 2.5 million. The detailed calculator's power-plant paragraph used an older overall overhead of 1.2; it now uses the cooling design's PUE, as everywhere else. Proposal sizes. Saline Township's 1.4 GW and Van Buren Township's 1 GW are electricity contracted from DTE, not the computers' capacity. Their proposals now load the computers at the contract divided by the top of the design's PUE range [A]: Saline about 1,130 MW instead of 1,400, and Van Buren's open-loop third about 280 MW instead of 330. The size buttons read 830 MW (Van Buren) and 1,130 MW (Saline).
• Oct 27, 2026 (last, water source). Where water starts [A]. A place with no public water system mapped by EPA now looks for the nearest sizable system (more than 3,300 people) within about 10 km. If there is one, a closed-loop campus starts on it, by an extension, and the note and card name it; cooling towers and hybrid still start on new wells, with that system named as the alternative. With closed loop, 691 of the 1,541 places now start on public water and 850 on wells (before, 470 and 1,071); towers and hybrid are unchanged. With well water and the public sewer, the sewer card asks how the wastewater will be metered.
• Oct 27, 2026 (last). Where wastewater starts [A]. A place with no public water system mapped by EPA but a large sewer plant within about 10 km now starts on the sewer, reached by an extension or agreement, instead of the campus's own discharge (we have no sewer service map, and EPA's water system boundaries often stop at city limits). Where no sewer is within reach, closed loop's mostly sanitary flow now goes to a new choice, a septic system on the site, approved by the county health department or permitted by EGLE depending on flow (Michigan's Part 22 rules). Evaporative cooling's blowdown goes to the sewer only if the plant has spare design flow for it, otherwise to the campus's own permitted discharge. With closed loop, 838 of the 1,541 places now start on the sewer and 703 on a septic system; before, 470 started on the sewer and 1,071 on the campus's own discharge. Lowell Township: wastewater now follows its published plan, the City of Lowell sewer, whose plant averaged at or above its design flow last year.
• Oct 27, 2026 (late night). The split of water taken now always adds up: water sent back out is shown as water taken less water lost to evaporation (before, the middle values of the two came from separate model outputs and could miss the total by a few percent). "Used up" is now "lost to evaporation" for the campus's own water; nearly all the water staff use goes to the sewer.
• Oct 27, 2026 (night, later). Base use (the water a closed loop takes): the share lost is now sampled from 10 to 50% (middle 30%), not a fixed 50%, because most of it is sanitary use that returns to the sewer (about 10% lost) and Michigan needs little humidification. A 300 MW closed-loop campus uses up about 6,200 gallons a day instead of about 12,800 (middle of the runs); the water it takes is unchanged.
• Oct 27, 2026 (night). Electricity is compared with the local utility: the utilities serving each place from the MPSC service-area map, each one's average load and summer peak from EIA-861 (2025), and all of Michigan's average (about 11,500 MW). For closed loop, the extra peak against cooling towers for the same computers.
• Oct 27, 2026 (evening). Electricity peak: the hottest-day figure is now every computer running (full capacity) times the top of the published PUE range for the cooling design, the peak a utility has to supply. It was the average computing load at that PUE. A 100 MW closed-loop campus: about 86 MW on average, about 124 MW at peak (was about 93 MW).
• Oct 27, 2026 (afternoon). Closed loop: the base-use split (about half lost as humidifier vapor, half to the sewer) is marked as an estimate. Golf courses: a summer watering-season figure is added, the yearly median spread over May to September (153 days), an estimate. Data files are now rebuilt by one update process; a rebuild from the same downloads gives identical numbers.
• Oct 27, 2026 (late morning). Starting assumptions for every place: a township starts at 300 MW on 200 acres of farmland and a city at 100 MW on 50 acres already built on; closed loop is the starting cooling design where none is published (6 of the 9 tracked proposals have said closed loop); water comes from the public system if EPA maps one, otherwise new wells; wastewater goes to the public sewer if a system is mapped and a sewer plant is within 40 km, otherwise to the campus's own permitted discharge. A tracked proposal loads its published settings, or a stated assumption where it published none (hybrid where the design is undecided). Electricity: homes' worth on the hottest days uses the top of LBNL's PUE range for the design (an estimate); homes are rounded to the nearest thousand. Before and after: likely ranges (the middle 80% of runs) and numeric changes; "Same" only when the rounded numbers match. Groundwater: the local aquifer (glacial deposits or bedrock), well depth and water level from EGLE Wellogic household wells (1,333 places). Sewer plants now come from EPA's 2022 Clean Watersheds Needs Survey (394 Michigan plants, including lagoons and groundwater-discharge plants); ECHO's last-year flow is kept only where it is 0.1 to 1.3 times the design flow.
• Oct 27, 2026 (morning). Rain, runoff and water quality: runoff, sediment, phosphorus and nitrogen before and after (with Michigan's standard runoff controls and with infiltration), construction sediment, road salt, generator fuel and cooling-water chemistry; impaired waters and cleanup plans (TMDLs) downstream from EPA ATTAINS 2026; designated trout streams; nearby sewer plants with design and recent flows. A before-and-after summary of public water, water evaporated, the sewer plant, runoff, sediment, phosphorus, road salt, diesel and electricity, computed the same way in Northline's Water Review (a check keeps the two in agreement).
• Oct 26, 2026 (evening). Electricity by cooling design (LBNL PUE ranges); an "Electricity from" setting (Michigan's mix, gas, coal, nuclear, wind and solar, MISO) with water evaporated at power plants (Macknick et al. 2012).
• Oct 26, 2026 (afternoon). A "Wastewater goes to" setting (public sewer, river or lake, treated and reused). Lake-source (once-through) cooling for places near a Great Lake, with extra lake evaporation from the USGS heat-balance method and NOAA lake temperatures. County comparisons use EGLE facility data, with golf courses broken out.
• Oct 26, 2026. Three water sources (public water system, new wells, lake surface water within about 5 km of a Great Lake or Lake St. Clair). Comparisons with published thresholds (Michigan Part 327, the Great Lakes Compact) and the local system's connection count.
• Model 1.1, Oct 26, 2026. Eight climate regions instead of five (added Southwest Michigan, Saginaw Valley and the Thumb, and Northeast Michigan; the eastern Upper Peninsula now uses the nearer northern Lower Peninsula stations). Every Michigan city and township, with its county, climate region, public water systems (EPA), watersheds (USGS) and design storms (NOAA). Removed every assumed utility number (per-person use, plant capacity, current sewer flow); where no public dataset has a number, the page says so.
• Model 1.0, Oct 25, 2026. Weather-driven model aligned with published methods (USGS evaporation relation, seasonal heat factor, drift, four cooling designs, Monte Carlo ranges); taken / evaporated / returned split; power-plant water shown separately; county water-use context; validation table.
• Oct 20, 2026. Proposal tracker re-verified (10 entries); Lowell, Saline, Augusta and Ypsilanti entries corrected.

How each number is made: methods and sources. See something wrong? Suggest a correction.