Skip to content
Photograph from orbit at night: the thin blue and orange line of sunrise along the curve of the Earth, with city lights below

CS1 · 150 kW thermal module · In development

The thermal layer for orbital compute

A plug and play heat rejection module for high power spacecraft. Cold plate, pumped loop and deployable radiator under one part number, built on a flight proven class of architecture.

Orbital sunrise seen from the International Space Station. Image: NASA

01 / The problem

Every watt you compute in orbit, you must radiate

In vacuum there is no air to carry heat away. For a closed system, radiation is the only exit, and capacity scales with area and the fourth power of temperature. A single NVIDIA GB300 NVL72 class rack dissipates about 140 kW. Moving that heat from the chip to space is the hardest layer of orbital compute.

CAD render of the CS1 radiator seen edge on, a thin vertical line against black
Radiator edge on to the sun, CAD render

Orbital compute

100+ kW racks

GPU racks push past 100 kW of continuous heat. The radiator, not the silicon, sets the ceiling on orbital data center scale.

Power beaming

Conversion waste heat

Laser and optical power chains reject waste heat at every conversion stage. Same wall, different payload.

High power payloads

Radar, optical, comms

High power apertures need the same rejection capacity with a simpler duty cycle, and no appetite to reinvent it.

02 / The product

CS1: one part number, chip to space

A standard, drop in thermal module: cold plate, pumped PGW loop and deployable tube and fin radiator. Your team bolts it on and ties it into the coolant loop. CS1 is sized to a full 150 kW heat load and anchored on an NVIDIA GB300 NVL72 class rack.

CAD render of the CS1 central spine with the pump module, coolant mains and compute trays between two radiator panels
CS1 spine detail, CAD render
CAD render of the CS1 thermal module: a flat central spine with radiator panels rising above and below it
CS1 deployed, three quarter view, CAD render

The thermal chain

Chip side to space

  1. 01

    TIM2 interface

    Hands heat from the customer lid into our cold plate. Our spec starts at the lid.

  2. 02Our edge

    Cold plate

    Metal microchannel plate on each GPU, CPU and switch. Our main differentiator.

  3. 03

    Pumped loop

    PGW warm loop, about 2 kg/s at a 20 K rise.

  4. 04

    Redundant pump

    N+1. The only moving part, so it is built for reliability.

  5. 05

    Radiator

    Metal tube and fin panel, warm at about 345 K, radiating from both faces, edge on to the sun.

  6. 06

    Structure

    Deployment mechanisms, micrometeoroid shielding and brackets. Scope is shared with the bus and settled per customer.

One core, two mounts

A

Direct GPU mount

Cold plates mount straight onto the GPU packs. The tightest, highest performance path.

B

Solar wing back mount

The radiator mounts on the back of the solar wing structure, still edge on to the sun.

Naming convention

CS1
Standard size, 150 kW class
CS1-2
Size variant: a different heat load or area
CS1 Rev B
Design revision

03 / The architecture

Proven class. Done better.

We anchor on the closed, warm, pumped liquid loop with a metal radiator. Pumped loops and deployable radiators are an architecture family with flight heritage. It is a deliberate choice of proven hardware over novel radiating surfaces. Our job is to deliver that heat rejection as a drop in block that wins on cold plate performance, reliability, system mass and cost.

Flight proven class and TRL 7 describe the architecture family. CS1 hardware is in development and has not flown.

Photograph of white radiator panels and gold solar arrays on the International Space Station, with the Earth below
Radiator panels and solar arrays on the International Space Station, flight heritage of the architecture family. Image: NASA. Not Chiller Space hardware
  1. 01

    Proven path

    Reference class system: 150 kW peak over about 110 m² of two sided radiator at about 345 K, edge on to the sun, with redundant pumping loops and integrated micrometeoroid shielding.

  2. 02

    Cold plate first

    Performance is won or lost at the chip interface. Metal microchannel plates are the baseline. A direct die option is on the roadmap.

  3. 03

    Drop in

    One part number, two mounts. You bolt it on and tie into the coolant loop, so thermal stops being your critical path.

  4. 04

    Built for the failure mode

    The pump is the only moving part and the first failure mode, so it is N+1 redundant by design.

04 / Headline specs

CS1 · 150 kW class

Heat rejection
150 kW
Reference radiator area
~110 m²
Radiator temperature
~345 K
Pump redundancy
N+1
Architecture family
TRL 7
CAD render of CS1 seen from the front: six radiator panels in a vertical column with the spine at the center
CS1 deployed, front elevation, CAD render
ParameterCS1 (150 kW class)
Heat rejection150 kW module, flight proven class of architecture
Anchor loadNVIDIA GB300 NVL72 class rack, about 140 kW, 117 cold plate interfaces
RadiatorMetal tube and fin, two sided, about 345 K, edge on to the sun
Radiator areaAbout 110 m² at reference conditions, deployable
Working fluidPGW (propylene glycol and water), closed warm loop
Loop flowAbout 2 kg/s at about 20 K rise
PumpN+1 redundant BLDC, about 1 to 1.5 kW electric
Cold plateMetal microchannel baseline, direct die option on the roadmap
MountingA: direct GPU mount. B: solar wing back mount
HeritagePumped loop architecture family, TRL 7 as a family

Figures are engineering targets for a product in development, stated at reference conditions: both faces radiating at about 345 K to a deep space sink, edge on to the sun. In low Earth orbit the effective sink is warmer, so the same load needs more area. Radiator sizing for a given orbit and attitude is set per mission. TRL 7 describes the architecture family, not flown Chiller Space hardware.

Detailed spec sheet

The detailed CS1 spec sheet is not a download. A person reviews each request, then sends it by email.

Request CS1 specs

05 / Who it is for

Built for the teams carrying the heat

Buyer 01

Orbital data center builders

You reject 100+ kW of compute heat per rack. Integrate a thermal layer instead of standing one up in house, and keep your team on the data center.

Request CS1 specs

Buyer 02

Satellite bus and integrators

Your bus carries compute or high power payloads. Outsource the thermal layer as one part number with a clean interface. Brackets and scope are settled per customer.

Discuss integration

Buyer 03

Space solar and power beaming

Laser and optical conversion chains reject large waste heat. CS1 is sized for that load with a flight proven class of architecture.

Request CS1 specs

Buyer 04

High power payloads

High power radar, optical and comms payloads need the same heat rejection with a simpler duty cycle. Same module, sized to your load.

Talk to us

Standing up your own thermal capability?

We also offer thermal analysis and engineering as a service for integrators. It is a fast way to start working together.

Start a services conversation

06 / Path to purchase

From first call to purchase order

  1. Step 01

    Requirements

    Tell us your heat load, orbit, envelope and schedule. We respond within three business days.

  2. Step 02

    Spec share

    We review each request, including who is asking and from where. After approval you receive the detailed CS1 spec sheet by email.

  3. Step 03

    ICD and integration review

    Interface control document and integration review with your engineers, under NDA.

  4. Step 04

    LOI or PO

    Submit structured intent: a letter of intent or a purchase order request. It is not a binding order.

Photograph from orbit: the sun rising over the dark curve of the Earth, lighting the thin blue line of the atmosphere

CS1 · 150 kW · Two mounts

Stop building thermal. Start bolting it on.

Orbital sunrise over the Pacific Ocean, seen from the International Space Station. Image: NASA

07 / Request

Request CS1 specs

Tell us about your mission. A person reviews every request before any detailed document is shared.

Detailed specs are never sent automatically. Each request is reviewed by a person first.