THE POST
Data centers in space looking pretty appealing right now
ENGAGEMENT
- likes235,977
- reposts14,308
- replies16,037
- quotes2,920
- bookmarks4,918
- views10,612,197
AS RETURNED BY THE X API · FETCHED 2026-11-05 06:17 UTC
WHAT IT SAYS
Musk states, without detail, that data centres in space look appealing right now. The post sits next to his earlier argument that power, transformers, cooling, networking and batteries are the AI bottleneck on Earth.
CONTEXT
Orbital compute is no longer only a slide. Fierce Network summarises Starcloud's Starcloud-1 (an NVIDIA H100 flown in November 2025), a 2027 Starcloud-2 cluster target, Google's Project Suncatcher prototypes for early 2027, and SpaceX talk of sun-synchronous compute satellites. McKinsey's May 2026 interview with Starcloud CEO Philip Johnston puts Starcloud 3 at roughly 200 kW around 2028-2030 and sketches a long-run constellation on the order of 20 GW of inference capacity.
The economics are still the open question. Fierce cites aerospace engineer Andrew McCalip's model that 1 GW of orbital solar compute would cost about $51.1 billion against $15.9 billion on Earth, before compute hardware. Johnston tells McKinsey space only undercuts terrestrial cost if launch falls toward roughly $500/kg, and that dawn-dusk sun-synchronous orbits are used to avoid large batteries.
On the ground, the pressure that makes orbit look appealing is documented this week. Power Engineering reports Omdia projecting up to 35 GW of behind-the-meter data-centre power by 2030, with BESS investment possibly above $50 billion a year by then, as grid queues stretch and developers buy gas turbines and batteries to get watts sooner.
WHAT IT MEANS FOR K
If orbital solar compute becomes cheap enough to scale, it adds harnessed power off Earth's grids. That would raise effective K for AI workloads, but only after launch costs and radiator mass clear the economic bar. Today it is a plan, not new watts.
CLAIMS
- Data centres in space look more attractive than terrestrial sites right now.
- Orbit can sidestep land and grid bottlenecks with continuous solar in the right orbit.
EVIDENCE
- Starcloud-1 flew a data-centre-grade GPU in Nov 2025; Starcloud-2 targeted for 2027; Google Suncatcher prototypes for early 2027 (Fierce Network).
- McKinsey/Johnston: ~200 kW Starcloud 3 around 2028-2030; competitiveness hinges on launch near $500/kg.
- McCalip model via Fierce: ~$51.1B vs $15.9B for 1 GW orbital vs terrestrial solar compute infrastructure.
- Power Engineering/Omdia: up to 35 GW behind-the-meter data-centre power by 2030 as grid waits bite.
REASONING
- Terrestrial power delays are real and make orbital options more interesting.
- No commercial-scale orbital data centre is delivering GW today; the post is a preference, not a commissioning.
- Verdict is speculative until launch economics and flown capacity close the cost gap.
SEE ON THIS SITE
TAGS
- COMPUTE
- SPACE
- AI POWER
- SOLAR
- DATA CENTRES