Chip & cold plate
Direct-to-chip cold plates take heat off GPU and CPU dies at up to 1,600 W per socket. Junction to coolant in a single conduction step.
≈ 45 °C outReady to deploy 5 Megawatt Hybrid Adapting Module.
Drag to orbit · pinch to zoom · scroll to read on
The whole chain
We engineer, supply and commission the complete heat-rejection path — not just the box on the roof. Every degree of approach between the chip and the ambient wet-bulb is ours to defend. At the ambient end the plant is a bank of identical 5 MW modules on common headers, so a 40 MW campus is eight of the same thing rather than one bespoke structure.
Direct-to-chip cold plates take heat off GPU and CPU dies at up to 1,600 W per socket. Junction to coolant in a single conduction step.
≈ 45 °C outTreated, filtered secondary coolant collects through vertical rack manifolds. Quick-disconnects, leak detection, N+1 flow.
45 / 37 °CA liquid-to-liquid plate exchanger with N+1 variable-speed pumps splits the clean technology loop from the facility loop and holds secondary setpoint independently of the outside world.
≈ 3 K approachA closed, clean primary circuit. Because the SEA-5 module coil is closed-circuit, no intermediate plate exchanger is needed here — and that saves 1.2 K of approach outright.
40 / 32 °CInside each 5 MW cell, finned V-bank coils take the first, sensible bite out of the return water. Whenever ambient allows they take all of it, and the wet section stays off.
zero waterBelow the dry coil, a closed-circuit coil sits in a modulating spray. Duty ramps 0–100 % against wet-bulb and load, so water is spent only where it buys temperature.
→ ambientThe unit of capacity
One cell, 5 MW of rejection, built and wet-tested in the factory and shipped complete. A 40 MW campus is eight of these on a common header rack — not one bespoke structure.
Size it · in 5 MW modules
Every figure below comes from an hourly simulation of the site you pick — 8,760 hours of synthesised dry-bulb, dew point and wet-bulb, run against a physical model of each machine. Capacity is built up in 5 MW modules; move the sliders and the whole table re-solves.
Basis. Figures are indicative and subject to site-specific engineering. Scope per system: tower or cooler, circulating pumps, in-plant pipework, water-treatment plant, controls and civil works — CDUs and in-hall distribution are common to all four and excluded. Setpoint 32 °C facility water supply (ASHRAE W32, chiller-less). Recirculating systems at 4.5 cycles of concentration; evaporation taken at 1.475 L per kWh rejected. The open tower carries an intermediate plate heat exchanger, because open-basin water cannot be sent to a CDU — that is worth 1.2 K of approach and a water-treatment plant. Adiabatic pads are once-through with no cycles of concentration, and the small tropical wet-bulb depression buys little cooling per litre passed over the media, which is why their annual water draw lands close to a modulating hybrid despite running dry more often.
Ambient conditions and the water temperature each machine can actually deliver, hour by hour.
Wet-section duty and fan demand over the same day, as a share of rated.
Why Southeast Asia is different
Temperate-climate playbooks do not survive the crossing. In Singapore the dry-bulb sits high, the dew point barely moves all year, and the gap between dry-bulb and wet-bulb — the entire budget an evaporative process has to work with — is often only 5 to 7 K. That single fact is what makes dry cooling expensive, adiabatic pre-cooling disappointing, and modulation valuable.
Annual hours by wet-bulb bin, split by what the SEA-5 module is doing in them.
Our edge
The wet section modulates continuously from full duty to fully off. Water is spent only in the hours and at the intensity that actually buy you temperature — typically cutting annual draw by around two thirds against an open tower on the same duty.
At the design wet-bulb the wet section goes to full duty and delivers the same approach an open tower would — without the intermediate plate exchanger, and therefore 1.2 K colder at the CDU. Dry and adiabatic machines simply run out of temperature and hand the problem to your chip.
For most of the year the wet section is running below full duty. That idle capacity is standing reserve: when a training job steps the hall from 40 % to 100 %, spray pumps ramp and the extra kilowatts land in under thirty seconds — no pad wetting lag, no basin transient, no thermal throttling.
Lose municipal supply and an open tower has only its basin — hours, at best. A SEA-5 bank falls back to dry operation and keeps running indefinitely at around 70 % of rated capacity. In a region where water permits and dry-season restrictions are real, that is a different risk class.
Process water runs inside a closed coil and never meets open air, so there is no aerosolised process water and a far smaller legionella regime to manage. Warm dry air off the top coil mixes with the saturated plume and suppresses it — which is often what makes an in-city or industrial-estate permit possible at all.
Every cell is the same SEA-5 module: factory-built, tested before it ships, set and connected to common headers in about two days. A 40 MW campus is eight identical things, so spares are one part list, the O&M procedure is one document, and the second hall does not need a second design.
It changes the money too. CAPEX arrives when the hall energises rather than on day one, redundancy costs 5 MW instead of a duplicated plant, and any single cell can be valved out and serviced while the rest of the bank carries the load.
We do not sell a nameplate. Every proposal is an 8,760-hour simulation against your site's dry-bulb and dew point, your load profile and your tariffs, with the dry/wet split chosen to put the crossover exactly where your economics want it. The model behind this page is the same one we hand you.
Talk to an engineer
We will come back with an hourly simulation for your actual site: water, power, delivered temperature and total cost across all four technologies — not just ours.