What is Starcloud and how does it compare to SpaceX's orbital data center plans?

Starcloud is an independent Seattle-area startup building orbital data centers, separate from SpaceX. It launched Starcloud-1 in November 2025 with an NVIDIA H100 GPU and successfully processed AI workloads in orbit (including training a small language model). In March 2026 it raised $170M Series A at a $1.1B valuation. In May 2026 it signed a contract with Starlink for laser terminals across 25+ Starcloud satellites. SpaceX's own orbital data center plan is much bigger and farther out: in January 2026 SpaceX filed with the FCC for up to 1 million data center satellites, with first deployment targeted for 2028. Starcloud is in orbit and operating; SpaceX is planning.

Why would anyone want a data center in orbit?

Three reasons that AI infrastructure operators are seriously evaluating: (1) Solar power is continuous and free at orbital altitudes — no grid constraints, no fuel costs, no regional capacity shortages. (2) Radiative cooling into space is effectively unlimited — terrestrial AI data centers in 2026 are bottlenecked on water and cooling permits in places like Louisiana and Arizona. (3) Permitting is simpler — orbital data centers don't need state-level utility deals or 7GW gas plant approvals like Meta's Hyperion campus in Louisiana. The downsides are real: launch costs, radiation, single-point-of-failure satellites, debris risk, and latency back to Earth users. For pure AI training (long-running, latency-insensitive workloads), orbital data centers may pencil out economically by 2028-2030.

Is AWS or Microsoft launching orbital data centers?

Not directly as of June 2026. AWS and Microsoft are doubling down on terrestrial AI data centers — Microsoft's Stargate ($100B+), AWS's Trainium clusters, and Anthropic's Project Rainier on AWS. AWS Ground Station already provides satellite communication infrastructure, but no Bezos-led orbital compute. Both hyperscalers are watching Starcloud and SpaceX closely, but their bet for 2026-2028 is on terrestrial scale (Meta's 7GW Hyperion in Louisiana, Google's Council Bluffs expansion, AWS's South Carolina build-out) rather than orbital experimentation. Orbital data centers will likely be category-3 or 4 capacity through 2028, supplementing rather than replacing terrestrial compute.

Should AI startups care about orbital data centers in 2026?

Practically, no. Starcloud-1 is one satellite with one H100, useful for proof-of-concept AI workloads but not a production training cluster. SpaceX's full constellation is years away. For 2026-2027, AI startups should plan around terrestrial compute via AWS Bedrock, Azure, GCP, CoreWeave, Lambda Labs, and the Anthropic/OpenAI direct APIs. Where orbital data centers may become relevant earlier: long-running training jobs that don't need low-latency inference (foundation model pre-training), and government/defense workloads where physical isolation has value. Watch Starcloud-2 in October 2026 — projected 100x the power of Starcloud-1 — for the first commercially meaningful orbital AI compute.

Quick Answer

Starcloud vs SpaceX vs AWS Orbital Data Centers June 2026

Published: June 20, 2026

Starcloud vs SpaceX vs AWS Orbital Data Centers June 2026

Orbital AI data centers went from sci-fi to FCC filings in 18 months. Starcloud has a satellite in orbit running AI workloads. SpaceX has filed for up to 1 million data center satellites. AWS and Microsoft are still betting on terrestrial scale. Here is how the orbital AI compute race stacks up in June 2026.

Last verified: June 20, 2026.

TL;DR

Starcloud is the leader in operating orbital data centers. Starcloud-1 launched November 2025 with an NVIDIA H100, $1.1B valuation in March 2026.
SpaceX filed for up to 1 million data center satellites in January 2026; first deployment targeted for 2028. Liquid cooling confirmed by Musk in June 2026.
AWS, Microsoft, Google are doubling down on terrestrial AI data centers (Stargate, Hyperion, Project Rainier) — no orbital plays as of June 2026.
The economics work for AI training, not inference: solar power, radiative cooling, simpler permitting.
For AI startups today: plan terrestrial. Orbital is interesting starting late 2026 (Starcloud-2) and meaningful from 2028+.

Direct comparison

Dimension	Starcloud	SpaceX Orbital	AWS / Microsoft / Google (Terrestrial)
Status June 2026	One satellite in orbit (Starcloud-1, Nov 2025)	FCC filing (Jan 2026), no deployments	Operating at exascale
Funding	$170M Series A, $1.1B valuation (Mar 2026)	SpaceX balance sheet + Starlink	Trillions cumulatively
First production target	Starcloud-2 (Oct 2026, 100x Starcloud-1 power)	2028	Operating now
Compute per node	1× NVIDIA H100 (Starcloud-1)	Liquid-cooled GPU clusters (planned)	Tens of thousands of H100/H200/B200 per cluster
Constellation size (planned)	88,000 satellites (FCC filing)	Up to 1,000,000 satellites (FCC filing)	Hundreds of terrestrial data centers
Power source	Solar (continuous in orbit)	Solar (continuous in orbit)	Grid + gas + nuclear + renewable
Cooling	Radiative to space	Liquid + radiative (Musk: “no water”)	Water + air + immersion + liquid
Connectivity	Starlink laser terminals (May 2026 deal)	Starlink native	Fiber backbone
Best for	Training jobs, isolated compute	Future massive-scale AI training	Inference, training, everything today
When relevant for AI builders	2027+	2029+	Today

Starcloud’s trajectory in detail

Starcloud is the most concrete orbital data center company in 2026.

November 2025: Launched Starcloud-1 with an NVIDIA H100. Successfully processed AI workloads in orbit, including training a small language model.
March 2026: Closed $170M Series A. Post-money valuation $1.1B.
May 2026: Signed contract with SpaceX Starlink for over 50 Starlink Mini Laser terminals across 25+ Starcloud satellites. Enables continuous high-bandwidth low-latency connectivity between Starcloud nodes and Starlink, bypassing ground station bottlenecks.
June 2026: Operating Starcloud-1, preparing for Starcloud-2.
October 2026 (planned): Starcloud-2 launch. Projected 100x the power of Starcloud-1.
Long term: FCC filing for up to 88,000 satellites in the full constellation.

SpaceX’s plan

SpaceX is going much bigger but later.

January 2026: FCC filing for up to 1,000,000 data center satellites.
June 2026: Musk confirms liquid cooling for orbital data centers — interesting because it suggests a closed-loop cooling system, not radiative-only.
2028: First production data center satellites expected.

The difference is scale and timeline. Starcloud is shipping in 2026. SpaceX is planning for 2028. If both execute, SpaceX could dwarf Starcloud by an order of magnitude — but Starcloud has the operating experience now.

AWS / Microsoft / Google: terrestrial scale-up

Hyperscalers are not chasing orbit. They’re scaling terrestrial.

Microsoft Stargate: $100B+ multi-site terrestrial AI campus build-out.
Meta Hyperion (Louisiana): 2GW initial, up to 5GW potential, online summer 2028. Total Meta AI infra commitment >$200B.
Meta Prometheus (Ohio): 1GW supercluster online in 2026 — Meta’s first gigawatt-scale facility.
AWS Project Rainier (Anthropic): Custom Trainium clusters for Anthropic training, multi-state.
Google: Continued TPU buildout at Council Bluffs and other sites.

The hyperscaler bet is “we can permit, build, and power terrestrial AI compute faster than orbit can become economic.” That bet is probably right through 2028. After 2028, the calculus changes if Starcloud and SpaceX execute.

When orbital wins (and when it doesn’t)

When orbital wins

Long-running training jobs. Foundation model pre-training is latency-insensitive. Months-long training runs don’t care about Earth-orbit round-trip.
Power-bottlenecked workloads. Terrestrial AI data centers in 2026 are constrained by grid permits and gas plant approval. Orbital is “just launch more.”
Cooling-bottlenecked workloads. Radiative cooling into space is effectively unlimited.
Sovereign / isolated workloads. Physical isolation has real value for government, defense, and certain regulated industries.

When terrestrial wins

Inference. Sub-100ms latency matters for chat, code completion, agentic AI. Orbit adds round-trip cost.
Training that needs frequent data refresh. RAG-style training and continual learning prefer terrestrial bandwidth.
Cost in 2026-2027. Launch costs are still real money. Starcloud’s $170M raise is one-satellite economics; terrestrial Hyperion is multi-hundred-billion-dollar per-cluster economics that scale much better today.

The competitive dynamics

The interesting tension: Starcloud is faster to market, SpaceX is bigger to market, hyperscalers don’t need to play yet. If Starcloud-2 ships in October 2026 and demonstrates real production AI workloads in orbit, expect:

AWS or Google to announce an orbital strategy in 2027.
More orbital data center startups to raise — there are already several Y Combinator companies in this space.
A reckoning on terrestrial AI compute permitting — if orbital becomes viable for training, the political case for Louisiana and Arizona gas plants weakens.

What this means for AI builders today

2026 planning: terrestrial only. AWS, Azure, GCP, CoreWeave, Lambda, direct vendor APIs.
2027-2028 watch list: Starcloud-2 results, SpaceX FCC progress, hyperscaler reactions.
2028+ planning: evaluate orbital for long-training-run workloads, sovereign deployments, and capacity arbitrage when terrestrial is constrained.

Sources

Starcloud announcements (Nov 2025, Mar 2026, May 2026)
SpaceX FCC filings (Jan 2026)
PCMag: “This orbital data center startup is integrating Starlink lasers” (May 2026)
PCMag: “SpaceX’s orbiting data centers will use liquid cooling” (June 2026)
GeekWire: “Starcloud hits $1.1B valuation” (March 2026)
Forbes: “Data centers in space coming soon, but with down-to-earth hurdles” (June 15, 2026)
NVIDIA blog: Starcloud collaboration

Published June 20, 2026 by andrew.ooo. See related coverage on Anthropic/SpaceX Colossus 1 compute deal and the broader AI infrastructure race.