SovereignIntelligencein Space

Every country needs to own the production of their own intelligence. Jensen Huang, World Governments Summit, 2024
A briefing for NVIDIASymphony Space, November 2026
The problem

In orbit, heat is the limit.
And it’s welded to the GPU.

≈1,600 m²
of radiator to reject 1 MW of heat in orbit, about an ice hockey rink
The compute it cools: about eight dense AI racks, ≈6 m²

Today’s orbital data centers bolt GPUs, radiators and the spacecraft into one machine.

When the GPU ages, everything retires together, radiators included. Meanwhile NVIDIA ships a new architecture every year.

RubinYear 1
Rubin UltraYear 2
FeynmanYear 3
NextYear 4
NextYear 5
NextYear 6
A monolithic satellite launched with Rubinis still flying Rubin when it retires, about five years later
The opportunity

Orbit just became a data center market.

“Space computing, the final frontier, has arrived.”

Jensen Huang, GTC keynote, March 2026, unveiling the Space-1 Vera Rubin Module

SpaceX, Google, Blue Origin, Starcloud and Axiom are all racing to put compute in orbit. Almost no one has a plan to keep it current.

25×

More AI compute than an H100 for space inference, from the Space-1 Vera Rubin ModuleNVIDIA, March 2026

$39B

In-orbit data center market forecast by 2035, up from about $1.8B in 2029BIS Research

$13.4B

Added for Golden Dome-related space and missile defense. Its sensing layer needs processing in orbit.FY2026 Defense Appropriations Act

Decouple the GPU from the spacecraft.

The platform lasts 15 years

Power, data, thermal and command, run by Symphony as a service.

Compute swaps every generation

Standard docking, power, data and thermal interfaces. Keep the radiators.

Radiators scale on demand

Tiles staged at our on-orbit depot. Partner servicers add and replace them.

Rubin Rubin Rubin Rubin Tile depot Platform year 0 Compute generation Rubin
What it means for NVIDIA

Every generation becomes an orbital sale.

Ground data centersRefreshed on NVIDIA’s cadence
Monolithic orbitRefresh means a new spacecraft
Spacecraft 1, one GPU generation
Spacecraft 2, relaunched whole
Spacecraft 3, relaunched whole
Symphony platformOnly the compute relaunches
One platform, 15-year life
Gen 1
Gen 2
Gen 3
Gen 4
Gen 5
Year 0Year 5Year 10Year 15Illustrative
Orbit buys on a server cycle

Compute is the only part that relaunches, so orbital fleets can refresh like ground fleets do.

Merchant silicon stays the standard

Google is taking TPUs to orbit with Project Suncatcher. An open slot lets the best GPU win every refresh.

Sovereign AI, above every border

Allied governments own compute in orbit without building a platform or a launch program.

Defense compute at the sensor

Golden Dome tracking is processed in orbit, with sensors and GPUs upgraded on separate clocks.

The future of compute has an orbit.

Intelligence payloadsGPUs and sensors in standard slots, swapped every generation
The platformPower, data, thermal and command for 15 years
Arrays that growPower and radiator surfaces added on orbit as load rises
Compute stops
being a satellite.
It becomes a rack.

Depots, servicers and transfer vehicles turn orbit into a data center supply chain: launch compute when it’s new, keep everything else running.

Why Symphony Space

We’ve done this with governments, and in orbit.

Merry Walker Co-founder and CEO

Former U.S. diplomat: led emerging-tech policy, then International S&T Lead at White House OSTP. Brought France into the Artemis Accords, worked on the first AI Executive Order, and negotiated technology access with foreign governments. Mechanical engineer.

Nicolas Maubert Co-founder and CTO

25+ years at CNES and ESA across 97+ missions. Chief Engineer for Galileo design and validation at ESA-ESTEC.

29LOIs and agreements signed, $776K, including 16 hosting agreements
2028Launch slot secured for April, Cape Canaveral. Adagio-Mini follows in December.
100+Missions across a team averaging 26 years of experience

Put every NVIDIA generation in orbit with us.

1
Fly

Reserve a compute slot on Adagio-Mini, December 2028. Qualify Jetson, IGX Thor or Space-1 Vera Rubin on an open, swappable interface.

2
Standardize

Co-author the open compute-to-thermal interface. Every NVIDIA space module plugs into any compliant platform.

3
Invest

Join our Seed round as a strategic investor. Go to market together on sovereign AI in orbit.

Merry Walker
Co-founder and CEO, Symphony Space
hello@symphony-space.com
symphony-space.com
Use the arrow keys or swipe