Advanced Thermal Management for Datacenters and AI Factories
8,760-hour cooling & power analytics
We transform cooling into compute power This simulator will show you exactly how
Your power envelope has one weak point: your mechanical cooling
(AKA chiller plant). It drives your peak PUE, and on the
hottest afternoon of the year it decides how much IT your site can carry. IceBrick changes how that cooling
operates — in a way no other system can. Give us a few details about your data center and we'll run all
8,760 hours of your year to show you exactly how much headroom you can reclaim: typically
5–15% more IT capacity behind the grid connection you already own.
Question 2 of 8
Where does your data center live?
Aside from compute, weather is the single biggest driver of cooling cost. Click the hub closest to your site —
we will synthesize 50 years of hourly dry-bulb and wet-bulb from that location's climate normals.
—21 US hubs + Shoham (IL) · Australia below
Australia6 hubs · click to open
Sydney · Melbourne · Brisbane · Perth · Canberra · Adelaide — BOM climate normals, synthetic hourly year
Question 3 of 8
Name Your Power Envelope
Every project starts from one of two numbers — the substation connection you were offered,
or the IT nameplate you want to install. Type yours into its box and continue from that box: the other
quantity is derived for you at the design day. The two sliders below the boxes apply to both paths.
Choose your Substation feed (MW)
Type a number above and this column reads it as
your substation connection.
Choose your IT-Nameplate (MW)
Type a number above and this column reads it as
your IT nameplate.
How much of the connection operations may actually use at the worst hour.
UPS losses, distribution, white-space fans, lighting, security, offices. A modern site
with efficient UPS lands near 8–10%; older estates run 12–16%.
Question 4 of 8
How do you reject that heat today?
Pick the cooling cascade that matches your plant. This decides how much of the year is free,
how much runs on water, and how much lands on a compressor.
True Waterless Cascade
What leaves the plant. Return is always 10 K warmer. The same water temperature as the hall sliders on the next question — move either.
Where are the adiabatic pads installed? The dispatcher only uses the
hardware you actually have:
Chiller condensersThe common retrofit — fog on the air-cooled chillers. Assists the compressors on hot
hours; cannot extend free cooling.
Dry coolersWet media on the heat-rejection coils — extends free cooling toward wet-bulb, so cool
nights run compressors-off and charge into the opened room.
BothThe full toolkit: night wet free-cooling digs charging room, condenser fog defends the
hot afternoons. Recommended for new installs.
Adiabatic systems can be dispatched for two different goals — choose
what IceBrick should optimise on this site:
Replace the water with iceToday’s peak is held exactly where the pads hold it — and then the pads are shut. The store and
the ice plant are sized so that every hour the trigger would spray runs dry on ice instead: the same
envelope, zero water. The results open with the three mechanical-cooling envelopes side by side —
no means, water, ice — and the storage the last one takes.
Power first, water secondThe ceiling, the store and the charging plant are sized for power. Then every night the
waterless charging circuit banks whatever else fits dry under the ceiling, and that ice is spent in the
hours the pads would otherwise spray, hottest first. Water is bought only to defend the peak, never for
kilowatt-hours — annual water lands well below what your plant’s own trigger spends today.
RecommendedNight-foggingThe pads run at night on the dry coolers to their full depth, opening compressors-off
charging room; where the wet-bulb is too close to the supply for the dry coolers to carry, the
condensers of the chillers still running (and the ice plant’s own) are fogged instead. By day the pads
work the condensers. The only water rule is a cap on the change in site
WUE: L/kWh added over the plant’s water today — set it to your
site’s own WUE and the night never spends more than the day does.
Question 5 of 8
Is your data hall air-cooled or liquid-cooled?
This decides what the next question even means. A liquid hall has a CDU between your facility
loop and the chips, and the temperature that matters is the one the cold plates are promised. An air hall has
chilled water going straight to coils, and the temperature that matters is what leaves the plant. Pick one and
we will draw your loop.
Inside the data hall
What leaves the plant and enters the coils. The same slider as in the next question — move either.
Hot air that short-circuits the containment and lands on the server inlet. Raises the inlet and outlet shown here; the plant itself is priced on the chilled-water temperature, which this does not change.
Outdoors: the chillers’ own discharge recirculating over their intakes on the 2nd–5th day of a heat wave. Mirrors the slider in the next question and is applied by the engine above the trigger threshold.
What the CDU delivers to the cold plates. The same slider as in the next question.
The CDU’s heat-exchanger approach: facility water runs this much colder than the rack supply.
Question 6 of 8
What water do your racks actually get?
Your IT does not see facility water — it sees whatever comes out of the CDU, one heat
exchanger later. So we ask for the number your racks are promised and the step the CDU adds. Facility water is then whatever the plant has to make to deliver it, and that is the loop the whole analysis runs on.
Loop animation — the thermal spike, how you run now and with IceBrick tied in at the CDUcollapsed · click to open
—
Same loop, IceBrick tied in at the CDU: the DCIM sees the spike as it starts, the store is dispatched, and the cold wave reaches the CDU HX ahead of the heat.
On the 2nd–5th day of a heat wave the condensers stop seeing the weather station's
ambient — their own discharge recirculates over the intakes. Estimate the degrees that adds. It is
applied only in hours above 90 °F (32.2 °C), and only to the kW/ton of the chillers IceBrick
avoids: once the Storage cuts more than ~30% of the plant's load, the recirculation cell collapses and the
surviving machines get the real ambient back. Leave at zero to model no recirculation.
The ambient above which the recirculation cell establishes and the factor is applied.
Below it the condensers breathe the true weather-station air.
Question 1 of 8
What is this run for?
This simulator runs two distinct advanced thermal-management methodologies, for two separate
use cases. The IceBrick system can be deployed as a means of acquiring power flexibility — a firm
curtailment you commit to the grid for a demand-response event — or to optimize your power envelope
— more IT behind the connection you already own, every hour of the year. Pick the use case; the questions
that follow describe your plant, and the same weather, the same chiller and the same 8,760-hour baseline serve
either one.
Speed to Power — acquire power flexibilitySize IceBrick to the night before an event and commit a firm curtailment for a
demand-response window — the number you can put in a utility contract, on the day the grid
actually peaked.
Power Envelope OptimizationSize IceBrick to shave every design-week peak, unlock IT capacity under the same
connection, and report PUE, water and the freed megawatts hour by hour.
Where does the ice plant charge from?
The site's own connectioncharging must fit under the same ceiling the store protects
A separate non-firm feed — night onlya second substation or a non-firm agreement carries the charging draw; it never appears on the site meter, and it cannot be relied on by day
The ice circuit's own pumps while it discharges by day
On the site meterthe circulation pumps draw from the site while the store discharges — the non-firm feed stays untouched by day
From a Li-ion batterya small battery, charged at night from the non-firm feed, carries the pumps through every discharge and every event
Question 7 of 8
The demand-response event
Utility calls are usually 13:00–19:00 and run four hours. The store is sized to what the
night before can freeze under your own operational limit; the grid picks which ten hours of the year we
show you first.
hour the window opens · discharge ramps up over the 7 minutes before it
preset 4 h · adjustable 1–8 h
the site's default grid is preselected · official 2025 peak hours where published, the site's hottest days otherwise
Question 7 of 8
Go higher
The same loop with IceBrick tied in at the CDU. Watch the same spike again: the DCIM sees it
the moment it starts, IceBrick is dispatched, and the cold wave reaches the CDU HX ahead of the heat — the hall
never feels the hour. That interception, guaranteed under SLA, is what lets you commit to the warmest supply your
rack architecture allows.
Adiabatic regime at your site
Your climate's row of the supply-temperature map, computed live from this page's own physics
(design week · 30 MWth reference charging plant). The cursor follows the guaranteed loop your SLA sets —
.
Question 8 of 8
One last look before we run the year
This is the baseline operating matrix we will simulate. Change anything you like — you can also
come back and re-run with different answers at any time.
Fixed, and used for every compressor hour in this study — your plant is modelled
as N of these machines. We interpolate the unit's own published kW-versus-ambient curve and carry it from
the datasheet's 62 °F leaving water to your FWS through the lift. No rule-of-thumb kW/ton anywhere.
Run my 8,760-hour analysis →Builds the hourly weather, walks every hour through your cascade, and reports PUE,
peak PUE, water and your design day.
← Let me change an answerStep back through the questions.
Have a simulation ID?
Run history \u00b7 this browser
Hold tight — we are running your year.
8,760 hours, one at a time. Nothing averaged, nothing hand-waved.
Starting up…0%
8,760 analysis completed
This run
the operating matrix every number below is priced on
Your year, hour by hour
Change the water — the year re-runs in place
Every number below is priced on these two. Move one and all 8,760 hours are solved again.
re-running 8,760 h…
On top of what your own 90 °F trigger spends today. The dispatcher spends it on the peak-standing hours only, one hour at a time, until the peak hour has no pads left to give — never for kilowatt-hours. Move it and the IceBrick year re-runs.
The event — every number below re-solves in place
Window, grid and store size. The baseline year is already run; only the event and the night before are priced again.
—
The year, and the hours the grid actually peaked
Peak-hour reduction
—
— displaced by the ice at the peak kW/ton
— bonus on the remaining chillers
Peak kW/ton used
—
Charging plant to refill overnight
—
Storage per kW committed
—
Water in the event · adiabatic pads
—
Design day with the event
How much store is worth buying
Committed curtailment · firm
—
Energy shed per event
—
IceBrick storage
—
Li-ion equivalent
—
Baseline
Load profile
This is your whole year at hourly resolution — power stacked on the left
axis, water on the right, ambient behind it. Drag across the chart to zoom into any week, click a stream in the
legend to switch it off, and hit Reset zoom to come back out.
Facility power & water allocation · 8,760 h
Baseline
KPIs
Available capacity for IT under design-day conditions
—
Annual average PUE
—
Annual PEAK PUE · worst hour of the year
—
Peak facility demand
—
Annual cooling water · WUE
—
Design day. Chillers, switchgear and the grid connection are sized for the worst hour, not the
average one — the hottest day in the last twenty years of record (N20, the industry default).
———
Design day · hour by hour
Ready for the next level?
Same site. Same weather. Same cascade. One IceBrick thermal battery in the loop.
Results
Mechanical cooling, three ways: no means, water, ice
Results
Recommended design changes
IceBrick charges overnight — when the ambient is cold, the compressor
lift is small and power is cheap — and discharges through the day, taking the peaks off your chillers.
Nothing else about your plant changes. Here is what that does to the four numbers you just saw.
Annual average PUE
—→—
Annual PEAK PUE
—→—
Peak facility demand
—→—
Unlocked IT capacity · same grid connection
—
Design-day peak PUE
—
Daytime compressor avoided
—
Chiller plant you can retire
—
IceBrick storage size
—
Non-firm charging feed
—
Water use effectiveness · WUE
—
Side by side
Your year, before and after
Same axes, same scale. The red line is what the grid sees. Watch the
spikes disappear — and notice that the flat line sits barely above the old baseline, because the energy was
moved, not added. Blue is that move: load taken off the day and put back at night. Pink
is what the move buys on top of itself — the surviving chillers running at lower lift because the Storage is
post-cooling them, and the whole plant sliding down its part-load curve because it is carrying less. There is no pink
at night: the ice is frozen by a designated set of chillers on their own separate, colder circuit, so the machines
holding the facility never carry the charging duty and never feel it on their part-load curve. The night band is
pure blue — the ice circuit's own electric, and nothing else.
Drag to zoom either chart — the other follows, so you are always comparing the same hours.
Today's practice sprays the pads whenever the dry-bulb is above this. Move it and the whole study re-runs: the water bill, the peaks, the ice. Default 90 °F.
No means — pads never wetted, no IceBrick
Without IceBrick
Slide through the year
With IceBrick
Unlocked IT capacity
Slide through the year
Curious which hour sized this whole system?
Every size in this study traces back to a single
IceBrick-idle hour — the hardest of the year, and the day the ceiling is balanced against. Open the deep-dive
to see it.
The day that sized the system
The design day
Charge at night, coast through the peak
Your N20 design day, starting at 19:00 the
evening before so you can watch the charge begin. Switch between the cooling plant on its own and the whole
facility — and note the operational limit: the before-curve meets it on the worst hour, and the with-IceBrick curve never goes near it.
Design day with IceBrick
Month by month
Where the shift happens across the year
Compressor energy moved to night
daytime avoided vs night charging, MWh
What it adds up to
annual, at your operating matrix
Appendix
Heat-wave reserve
Appendix — heat-wave reserve: storage beyond one night
Every number in this report is for a storage sized to
one full night of charging — it starts and ends each day empty, and that is enough to hold the design day. A deeper
array banks ice on a mild night ahead of a heat wave and spends it on days whose own nights are too warm to
charge. Open the explorer to see how much extra peak each step of storage buys on this site — and where it stops paying.
Heat-wave reserve — how much extra storage is worth
Read the curve left to right: 1× is one night of charging, the size this report
recommends. Each step to the right adds storage that carries ice across days. The line is the annual peak that size can hold;
it flattens at the physical floor — the hour when IT, ancillary and pumps alone are the whole draw, which no amount of ice can
go beneath. Extra storage is heat-wave insurance down to that floor, and nothing past it.
automatic — chosen per hour
The dispatcher is free in every hour: it discharges wherever the meter would breach the
ceiling and charges wherever there is room beneath it, with perfect foresight of the year. No fixed
charge window exists any more.
———
Case study
No adiabatic option: trim with chillers or with IceBrick
Planning rate before the ITC (Investment Tax Credit). Includes the ice builders.
Design day · Option 1 · trim chillers
Design day · Option 2 · IceBrick, no trim chillers
The whole year · trim duty, ice builders and ice in store · 8,760 h
How this case study is computed
That is the next level.
Change a temperature, a location or a cascade and watch every number above move. The physics does not
care what you hoped for — which is exactly why it is worth looking at.
Take the whole thing with you
Every chart on this page at full resolution, every
number, and the assumptions behind them — one landscape PDF you can send to whoever signs for the
feeder. The charts are captured from this page exactly as you are looking at them.
The whole study — the baseline chapters and the
IceBrick chapters, one document. You will be asked to name the run first.
This run
Click to copy. Every input behind this page is in it — site, IT load, heat capture, cascade, hall type,
rack supply, SLA cap, CDU approach, facility supply before and after, ancillary, design return period and
Storage nights. Paste it alongside a screenshot and the exact run is recoverable, with nothing left to guess.
Name this simulation run
It goes on the cover of the report, along the foot of every page, and in the file name
— so whoever opens it knows which run they are holding.
Before we show you the numbers
The IceBrick comparison, the before/after charts and the downloadable report are gated —
tell us who's looking and they're yours for the rest of this session.
TRANSFORMING COOLING TO POWER
Screening-grade 8,760-hour model · synthetic weather from climate normals · swap in TMY3/AMY for bankable numbers
Chiller performance: YORK YVFA0309EVK46AAVBXO, Design Conditions Datasheet E.21.0.25809.0-D.89.0034 (YW21.00b) —
used for every compressor hour in this study