1. Introduction
Our last issue argued that the binding constraint on frontier technology is no longer chips — it is watts. This one follows that thread to its most radical conclusion. If the problem with AI compute is power, cooling, water, and land, then the most audacious response on the table is to move the compute somewhere those constraints do not exist: into orbit.
That sounds like science fiction. It is now a funded, in-orbit reality. In November 2025, Starcloud became the first company to operate an NVIDIA H100 — a mainstream terrestrial data centre GPU — in space, and went on to train the first AI model in orbit. [1] Within months it had raised a $170 million Series A at a $1.1 billion valuation, then a $250 million extension in August 2026 that valued it at $2.3 billion. [2] Google, SpaceX, and Blue Origin have all announced programmes of their own. The idea has crossed from novelty to a genuine, if early, contest over the next layer of digital infrastructure.
The investor question is not whether orbital compute is technically possible — it has been demonstrated. It is whether the economics close, when, and where in the value chain a defensible position can be built.
2. The Capital Story
Money is arriving faster than the technology is maturing, which is itself the signal worth reading.
Starcloud is the clearest case. Founded in January 2024, it reached unicorn status faster than any company in Y Combinator's history, and by August 2026 had raised roughly $450 million in total. [2][3] Its August extension was led by Manhattan West, with NVIDIA contributing approximately $25 million and Cisco Investments also participating — strategic validation from the company whose chips the sector depends on. [3] Notably, In-Q-Tel, the venture arm of the US intelligence community, has also backed the company, signalling sovereign and defence interest in the capability. [4] Its stated ambition is vast: an 88,000-satellite constellation delivering some 20 gigawatts of orbital compute, for which it has already filed with the FCC. [3]
The capital is also broadening beyond the headline name into the enabling value chain — satellite manufacturing (EnduroSat raised $104 million), imagery and data infrastructure, and launch. [5] High-profile backers including Benchmark, EQT, Sequoia and a16z scout funds lend credibility that pulls further capital in. [5]
But the defining constraint is not money. It is launch. Starcloud's CEO has been explicit that securing rocket capacity, not raising capital, is the binding problem: "we're going to need to book an enormous amount of launch." [6] With SpaceX's Falcon 9 scheduled to phase out by 2028 in favour of the still-unproven Starship, launch availability — not compute design — is becoming the industry's central bottleneck. [3]
3. The Bull Case: Why Orbit, and Why Now
The argument for orbital compute rests on the same physical constraints that our firm power piece identified on the ground.
Power. A solar panel in the right orbit produces up to eight times more energy per year than the same panel on Earth, with near-continuous sunlight and no need for batteries. [7] For AI training — enormously power-hungry and, crucially, latency-tolerant — that is a structural advantage.
Cooling and water. Terrestrial data centres consume billions of gallons of water and spend 30–40% of their energy budget on cooling. [8] In orbit, waste heat radiates into the vacuum and water consumption falls to zero. [9]
Land, permitting, and grid. Orbital facilities need no land-use permits, no water rights, and no multi-year grid interconnection agreements — precisely the frictions throttling terrestrial buildout, where US interconnection queues now stretch past five years. [9]
This is why the largest technology companies are moving. Google's Project Suncatcher plans two prototype TPU satellites with Planet Labs by early 2027, envisioning distributed clusters of 81 satellites. [10] SpaceX and xAI have filed to operate very large satellite fleets for orbital computing. [3] Jeff Bezos has predicted gigawatt-scale data centres in space within 10 to 20 years, arguing they will eventually beat the cost of terrestrial ones. [11] The best-suited workloads are clear: AI training, batch inference, and scientific simulation — compute-heavy tasks where throughput matters more than instant response. [9]
Picture 2. The Bull Case vs The Bear Case
4. The Bear Case: The Thermodynamics Problem
A credible investment view has to take the objections seriously, and they are substantial.
The hardest is heat. In space there is no air or water to carry warmth away, so every watt of waste heat must be radiated. Using the Stefan-Boltzmann law, IEEE Spectrum modelled that a single 700-watt H100 GPU held at 60°C would require roughly 1.4 square metres of radiator. [12] The World Economic Forum estimated that even a one-megawatt facility — a thousand times smaller than a terrestrial hyperscale campus — would need around 1,600 square metres of radiator, roughly the size of a hockey rink. [13] Radiator mass drives launch cost, and rockets and chips "want opposite things." [13] One Oxford researcher put it bluntly: a large orbital data centre would need an ISS-sized structure, and without Starship the concept is not yet physically feasible at scale. [14]
The other objections compound it. Ionizing radiation degrades computing hardware over time. [12] In-orbit repair is effectively impossible, so a five-year compute card must survive inside a 20–25-year platform. [14] And ground-to-orbit bandwidth caps how much data can move up and down. [9] Even relative optimists concede the sector depends on Starship reaching its cost and cadence targets — a milestone still ahead. [15]
The honest read is that the demonstrations are real but small, and the gap between a 60-kilogram satellite running one GPU and a commercially meaningful facility is measured in orders of magnitude.
Picture 3: When Does Orbital Compute Get Real?
5. What This Means for Allocators
Orbital compute is a genuine frontier, not a fantasy — but it is early, and the investability filter matters more here than in any sector we have covered this year.
The headline equity is largely inaccessible or already repriced: Starcloud at $2.3 billion, Google and SpaceX pursuing this inside trillion-dollar balance sheets. The more defensible frontier-allocator positions sit in the enabling layers, where value accrues regardless of which orbital operator ultimately wins — satellite manufacturing at scale, radiator and thermal-management engineering (the actual bottleneck), radiation-hardened compute, and the launch capacity that everyone is now competing for. As with defense, space, and firm power, the picks-and-shovels layer is where the accessible opportunity lives.
Timing discipline is essential. Analysts broadly expect the first meaningful applications — defence and satellite-data processing — to reach scale around 2028, with AI-training overflow dependent on Starship economics landing closer to 2030. [16] This is patient capital: a thesis to build a position in early, not one to underwrite for a two-year return.
For the Gulf specifically, orbital compute is a natural extension of the sovereign-capability build-out already underway — and it is a live theme rather than a distant one. At Future Investments Circle's Dubai event on September 7th, our Intelligent Machines panel took up exactly this question, with Dr. Shareef Al Romaithi, founder and CEO of Abu Dhabi's Madari Space, among the voices. Madari is building sovereign data-storage and edge-AI compute infrastructure in low Earth orbit, backed by the Mohammed bin Rashid Innovation Fund and based in one of the UAE's dedicated space economic zones, with a first pilot mission targeted for 2026. [17] A former Etihad captain and NASA HERA analog-astronaut alumnus, Al Romaithi frames the opportunity in sovereignty and sustainability terms: he argues the world's data centres now emit more carbon than the entire aviation industry, and that placing compute in orbit lets governments process and store critical data in real time, powered by the sun, without the terrestrial power and cooling burden. [17][18] It is the sovereign-infrastructure version of the same thesis Starcloud is pursuing commercially — and a reminder that the UAE is positioning itself early in a frontier most of the market still treats as science fiction.
Dr. Shareef Al Romaithi, CEO of MADARI Space, at the Future Investments Circle Dubai, September 7
7. Conclusion
Orbital compute compresses every theme we have written about this year into one audacious proposition. It is a space story, an AI story, and above all an energy story — an attempt to escape the terrestrial power and cooling limits that now define the economics of intelligence.
The demonstrations are real. The capital is real. The physics is genuinely hard, and the timeline runs to the 2030s. That combination — proven concept, brutal engineering, long horizon — is the signature of a frontier at its earliest investable moment. The allocators who study it now will be reading the map years before the returns arrive.
Compute is leaving the planet. Slowly, expensively, and not yet profitably — but it has left.
Sources
[2] TechCrunch — Starcloud raises $250 million for orbital data centers as launch options dry up (August 2026)
[4] Tech Funding News — Starcloud becomes unicorn with $170M raise led by Benchmark and EQT Ventures (March 2026)
[13] World Economic Forum — Why cooling is the real obstacle to space-based data centres (June 2026)