---
id: "concept-space-data-centers"
type: "concept"
source_timestamps: ["00:27:53", "00:46:25", "00:47:00"]
tags: ["space-tech", "artificial-intelligence", "energy-infrastructure"]
related: ["claim-ai-energy-bottleneck", "entity-aetherflux", "concept-first-principles-thinking", "claim-space-solar-viability", "question-space-data-economics", "prereq-ai-energy-intensity"]
definition: "Hosting AI compute infrastructure in low Earth orbit to utilize continuous solar energy and bypass terrestrial power-grid limitations."
sources: ["robinhood"]
sourceVaultSlug: "cardone-bhatt-robinhood-aetherflux-2026Jun25"
originDay: 10
---
# Space-Based AI Compute (Aetherflux Thesis)

## Definition

Hosting AI training and inference infrastructure in **low Earth orbit (LEO)** so it can be powered directly by near-continuous solar illumination, with processed data — not raw power — beamed back to Earth via optical (laser) links.

## The thesis

[[entity-baiju-bhatt]]'s current venture, [[entity-aetherflux]], aims to solve the energy bottleneck facing AI (see [[claim-ai-energy-bottleneck]]). Training and running advanced AI models requires exponential increases in electrical power, which terrestrial grids cannot supply quickly due to permitting, transmission build-out, and environmental constraints.

Bhatt's proposed architecture:

1. Place data-center satellites in orbits — particularly **sun-synchronous orbits** — that receive near-continuous solar illumination (see [[claim-space-solar-viability]]).
2. Perform compute (training and inference) in orbit using onboard GPUs / accelerators.
3. Transmit only the **processed data** — trained weights, inference results — back to Earth using high-bandwidth **optical/laser** communications.
4. This bypasses the highly inefficient process of beaming raw electrical power down to Earth, instead transmitting high-value, low-volume bits.

## Why it is a first-principles bet

The argument is grounded in [[concept-first-principles-thinking]]: judging the venture against the cost of legacy terrestrial data centers misses the point. Costed from raw inputs — silicon, aluminum, launch mass per kilowatt, photon-to-bit efficiency — the architecture can be viable even if analogous systems currently appear cheaper on Earth.

## External corroboration

- Payload Space describes Aetherflux's *Galactic Brain* as processor-hosting satellites powered by unceasing solar energy, with proprietary laser systems shuttling data to and from orbital GPUs.
- JLL cites Aetherflux among companies actively deploying first-generation orbital data centers.
- Sener and NTU Singapore independently argue that orbital data centers are technically viable at small scale, with **radiative cooling** handling thermal loads.

## Open questions

Economic viability at scale is genuinely contested — see [[question-space-data-economics]]. Engineering critiques cite launch cost, radiation-hardening, micrometeoroid/debris exposure, and the impossibility of on-site repair. JLL and Sener frame orbital data centers as **complementary** infrastructure for asynchronous, energy-heavy workloads (AI training, simulation, Earth-observation processing) — not as a wholesale replacement for terrestrial compute.

## Prerequisite knowledge

See [[prereq-ai-energy-intensity]].


## Related across days
- [[claim-ai-energy-bottleneck]]
- [[claim-space-solar-viability]]
- [[question-space-data-economics]]