Starcloud
Pre-revenue R&D-stage. Intended model: build and operate solar-powered AI datacenters in orbit (dedicated solar arrays + vacuum radiative cooling) and sell GPU compute capacity, initially via a named partner (Crusoe 'Crusoe Cloud' on a Starcloud satellite). No commercial revenue as of mid-2026; a bitcoin-mining payload is planned on Starcloud-2 as an interim monetization test.
The thesis on this name
State of the AI Cloud
Orbital AI-datacenter pure-play that flew the first data-center-class GPU (an Nvidia H100) in low-Earth orbit; PRIVATE, pre-revenue, the fastest YC company to reach unicorn status.
Earnings, margins, COGS & capex
Pre-revenue, R&D-stage orbital-datacenter startup. No commercial revenue, no audited financials, no public price. Value is entirely in a demonstrated technical milestone (first data-center-class GPU operated in LEO, Nov 2025) plus an aspirational roadmap (Starcloud-2/-3, an FCC filing for up to 88,000 satellites, a pitched 5GW satellite). All 'cost-competitive' orbital economics are company claims contingent on launch costs falling to ~$200-500/kg — a level that does not exist today. Funding-round valuations are the only hard financial marks: ~$1.1B post-Series-A (Mar 2026), ~$200M raised.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~90¢ is cost of goods and ~10¢ operating expense, leaving ~0¢ of operating profit.
Revenue trend
Margins
N/A — pre-revenue; claimed cost advantage is an unverified projection
N/A — R&D burn, no revenue
N/A
COGS structure
Not disclosed. Cost structure is launch + hardware dominated: satellite manufacture, Nvidia GPUs (H100 flown; Blackwell targeted), deployable radiators, and per-kg launch cost on SpaceX Falcon 9 (~$1,500/kg customer price today) trending — per the company's thesis — toward ~$500/kg only if Starship matures at scale. That launch-cost decline is a forward assumption, not a fact.
Capex
Not disclosed in dollar terms. Program roadmap: Starcloud-1 (~60kg, ~325km, Falcon 9, Nov 2025, one H100 — demonstrated); Starcloud-2 (late 2026, multiple GPUs incl. an Nvidia Blackwell, an AWS server blade, a bitcoin-mining unit, large deployable radiator — announced); Starcloud-3 (~200kW, ~3-ton, via SpaceX Starship, ~2028-29 — aspirational); a pitched 5GW datacenter satellite with a ~4 sq km solar array (concept only).
Latest earnings
N/A (private, no consensus, no revenue)
No financial guidance. Roadmap guidance only: Starcloud-2 late 2026, limited GPU capacity via Crusoe from 2027, Starcloud-3 ~2028-29; up to 88,000 satellites is an FCC filing (capacity ambition), not deployed hardware. All aspirational.
- Post-money valuation
- ~$1.1B (Series A, Mar 2026)
- Series A raise
- $170M (led by Benchmark + EQT Ventures)
- Total raised
- ~$200M (incl. ~$34M seed 2024)
- GPUs operated in orbit
- 1 (Nvidia H100, Starcloud-1) — Blackwell targeted next
- Commercial revenue
- $0 (pre-revenue)
- FCC-filed constellation
- up to 88,000 satellites (filing, not capacity)
Growth drivers
- First-mover technical lead — only firm to operate a data-center-class GPU (H100) in orbit and run/train real models (Google Gemma inference; a tiny NanoGPT trained on Shakespeare), Nov-Dec 2025
- Named commercial partner Crusoe ('Crusoe Cloud' on a Starcloud satellite, first cloud-in-space; limited GPU capacity from early 2027)
- Deep Nvidia relationship (Inception — Nvidia reportedly receives radiation soft-error telemetry to inform space-grade silicon)
- Blue-chip backers (Benchmark, EQT, a16z, In-Q-Tel) and unicorn status enabling a rapid Starcloud-2 build
- Launch-cost collapse optionality — if Starship reaches ~$200-500/kg, the free-solar / free-cooling thesis could become genuinely disruptive (contingent, not demonstrated)
Bull & bear
The category-defining orbital-compute pure-play and the only name in the theme with a hard, verified demonstration — first data-center-class GPU in orbit, first LLM run and first model trained in space — with blue-chip backers, a named customer, and genuine disruptive optionality if Starship collapses launch costs.
- Sole operator of a state-of-the-art GPU in orbit (H100, Starcloud-1) — turned the concept from PowerPoint into a flown, verified demo
- Fastest YC company to unicorn; $170M Series A at ~$1.1B (Benchmark + EQT), ~$200M raised — credible capital and momentum
- Deep Nvidia tie (Inception + radiation telemetry) and a named commercial partner in Crusoe (first cloud-in-space, capacity from 2027)
- Near-continuous orbital solar (claimed up to ~8x terrestrial productivity) + vacuum radiative cooling is a physically sound energy edge
- If Starship reaches ~$200-500/kg at high cadence, the free-solar / free-cooling economics could genuinely disrupt AI compute
- Optionality on a reported ~$2.2B follow-on round with SpaceX strategic interest (unconfirmed)
A >$1B valuation on a pre-revenue company whose demonstrated system is a refrigerator-scale single-GPU satellite; the entire economic case rests on ~$200-500/kg launch costs that do not exist and on MW-scale radiative cooling that is unsolved — and it is PRIVATE with no public buy.
- Pre-revenue at >$1B — the gap between demonstrated (one H100 run) and priced-in (5GW / 88,000 sats) is enormous
- Launch-cost-gated: break-even needs ~$500/kg vs Falcon 9's ~$1,500/kg today; the curve depends on unproven Starship scale (V3 flew only its first suborbital test, June 2026)
- Rejecting gigawatts of GPU heat via radiators in vacuum (~1,200 m² per MW) is unsolved at scale — the physics wall
- 88,000-satellite plan is an FCC FILING, not capacity; the 5GW / ~4km-panel satellite and $0.05/kWh figure are aspirational company projections
- A bitcoin-mining payload on Starcloud-2 signals monetization uncertainty for the AI-compute thesis
- Better-capitalized rivals (Google Suncatcher, Blue Origin) could crowd it out; regulatory/orbital-debris risk on a mega-constellation is material
- PRIVATE and pre-revenue — no direct public-equity access; strictly venture-exposure / watchlist, never a conviction long
What it is worth
Private venture mark anchored to the Series A (~$1.1B post-money, $170M raise led by Benchmark + EQT, Mar 2026); ~$200M total raised. No revenue and no audited financials, so no multiple applies — the mark is a venture bet on optionality, not a cash-flow valuation. Reverse read: >$1B is priced against a demonstrated single-H100-in-orbit demo and a 5GW / 88,000-satellite ambition that is years and multiple unsolved-physics / launch-cost problems away. Strictly venture-exposure / watchlist; PRIVATE — no public-market price.
A sharp markdown or down-round if a demo slips, a launch partner or Crusoe deal slides, the AI-capex/venture bid cools, or a far better-capitalized rival (Google, Blue Origin) reframes the category — stranding a pre-revenue balance sheet against a capital-intensive roadmap. As venture equity in a pre-revenue, physics-gated concept, permanent impairment is a real outcome.
~$1.1B (the last priced Series A mark holding): the company keeps executing demos (Starcloud-2 late 2026, Crusoe capacity 2027) but the fundamental launch-cost and MW-cooling problems stay unsolved, so it remains a well-funded R&D-stage venture bet rather than a scaling business.
~$2.2B+ (the reported follow-on mark holds) or higher into a hypothetical ~2027-2028 round at a $3-5B valuation IF Starcloud-2 hits performance targets, launch costs credibly trend toward the thesis, and the AI-compute-in-space narrative keeps a venture bid. IPO not on the roadmap before ~2029.
Pre-revenue. All 'cost-competitive' orbital economics (~10x cheaper energy, ~$0.05/kWh, ~$500/kg break-even) are UNVERIFIED COMPANY PROJECTIONS resting on unbuilt Starship launch costs — not demonstrated results. The reported ~$2.2B follow-on round and SpaceX strategic interest are press reports, not a confirmed priced round. Not a public price; not ownable.
SWOT
Strengths
- Only company to operate a state-of-the-art GPU (Nvidia H100) in orbit and run/train real models on it — concept converted from slideware to a flown demo (Starcloud-1, Nov 2-3 2025)
- Fastest YC company to reach unicorn status; blue-chip cap table (Benchmark, EQT, a16z, In-Q-Tel)
- Deep Nvidia relationship (Inception — privileged radiation soft-error telemetry) and a named commercial customer (Crusoe)
- Genuinely strong core pillar — near-continuous orbital solar (claimed up to ~8x more productive than terrestrial) + vacuum heat-sink is physically sound
- First-mover data + demonstrated schedule execution (from seed to flown H100 demo in <2 years)
Weaknesses
- Pre-revenue at a >$1B valuation — the demonstrated system is a ~60kg 'refrigerator-scale' single-GPU satellite, not a datacenter
- Entire economic case hinges on ~$200-500/kg launch costs that do not exist today (Falcon 9 ~$1,500/kg customer price) and depend on unproven Starship maturity at scale
- Radiative-only MW-scale cooling in vacuum (~1,200 m² of radiator per MW) is unsolved and the hardest, most under-hyped problem
- Radiation-hardening at fleet scale, on-orbit servicing, and downlink bandwidth all unsolved at datacenter scale
- A planned bitcoin-mining payload signals monetization uncertainty for the core AI-compute thesis
Opportunities
- Launch-cost collapse via a matured, high-cadence Starship would unlock the free-solar / free-cooling economics
- Edge-compute-for-satellites and secure/sovereign national-security compute are nearer-term niches than 'replace terrestrial hyperscale'
- A reported ~$2.2B follow-on round (with SpaceX strategic interest, unconfirmed) would fund the Starcloud-2/-3 buildout
- Category validation from hyperscalers (Google Project Suncatcher) legitimizes the theme and could pull customers/capital in
- First-cloud-in-space branding via Crusoe if 2027 limited-capacity offering lands on schedule
Threats
- Far better-capitalized entrants — Google (Project Suncatcher), Blue Origin (Project Sunrise ~51,600-sat filing) — could crowd out the venture pure-plays entirely
- The whole thesis is a single-variable bet on Starship $/kg — a launch-cost curve Starcloud does not control
- FCC spectrum + orbital-debris review, Kessler-syndrome / light-pollution concerns, and ITAR/export friction on an 88,000-sat plan add timeline risk
- Valuation prices an outcome (5GW / 88,000 sats) enormously ahead of what is demonstrated (one H100 run) — a growth or funding stumble re-rates hard
- AI-capex / venture-sentiment drawdown could freeze the mega-financing this capital-intensive roadmap requires
Moats, dependencies & bottlenecks
Moats
Real, verified demo lead today, but a single-satellite demo — hyperscalers with unlimited capital (Google) can close the gap fast.
Privileged soft-error data informs space-grade silicon; ecosystem-level, not exclusive — Nvidia sells to everyone.
Benchmark/EQT/a16z/In-Q-Tel and ~$200M raised fund the next builds, but the roadmap needs far more capital.
Marquee logo and 2027 capacity intent, but demo-scale and not yet revenue-proven.
Sound energy pillar, but the cooling half is unsolved at MW scale — the moat is contingent on physics + launch-cost problems it hasn't cracked.
Dependencies
launch provider + the $/kg curve the whole economic case rests on PRIVATE; break-even needs ~$500/kg vs ~$1,500/kg Falcon 9 today. Starship V3 flew only its first suborbital test (June 2026); rapid reuse + ~180 flights/yr undemonstrated.
GPU supplier (H100 flown, Blackwell targeted) + ecosystem backer Compute is Nvidia silicon; supply/price exposure. NVDA is NOT confirmed as a Series-A cap-table lead — the honest linkage is it sells the chips.
First customer / go-to-market (Crusoe Cloud on a Starcloud satellite) Demo-scale, 2027 limited capacity; revenue not yet proven and slippage risk is real.
Spectrum + orbital-debris approval for an 88,000-sat filing Mega-constellation review, Kessler-syndrome / light-pollution concerns, ITAR/export friction — all timeline risk.
Funding for the capital-intensive Starcloud-2/-3 and constellation buildout Deeply cash-consumptive R&D; the roadmap needs continued mega-financing, sensitive to AI-capex sentiment. Reported ~$2.2B round unconfirmed.
Advantages
- Only firm to operate a data-center-class GPU (H100) in orbit — verified first-mover
- First LLM run and first model trained in space (Gemma inference; NanoGPT training), Nov-Dec 2025
- Fastest YC company to unicorn; blue-chip backers and a named customer
- Deep Nvidia relationship with privileged radiation telemetry
- Physically sound orbital-solar energy pillar
Weaknesses
- Pre-revenue at a >$1B valuation — refrigerator-scale demo, not a datacenter
- Economics contingent on launch costs that don't exist and cooling physics that's unsolved
- 88,000-sat plan is a filing, not capacity; 5GW satellite is a concept
- Bitcoin-mining payload signals monetization uncertainty
- PRIVATE — no public buy; strictly venture-exposure
Bottlenecks
- Launch cost per kg — the linchpin; ~$500/kg break-even vs ~$1,500/kg Falcon 9 today, gated on unproven Starship scale
- MW-scale radiative-only cooling in vacuum (~1,200 m² of radiator per MW) — the physics wall, unsolved
- Radiation-hardening at multi-year fleet scale (single-chip lab results only so far)
- Downlink bandwidth + on-orbit servicing/repair (hardware failures are strandings)
- Capital intensity vs a pre-revenue balance sheet — the roadmap outruns the money raised
Top signals & trends
Top signals
Hard, verified demonstration — converts the concept from slideware to a flown demo, though at refrigerator scale.
Blue-chip capital (Benchmark + EQT) and momentum, but a rich mark on a pre-revenue base.
First named commercial customer / cloud-in-space — but demo scale and not yet revenue.
A capacity AMBITION / regulatory filing, not deployed hardware; also invites debris/spectrum scrutiny.
Would fund the next builds, but NOT confirmed closed — treat as reported, not fact.
Signals monetization uncertainty for the core AI-compute thesis.
The gap between demonstrated and priced-in is enormous; valuation prices an R&D-stage outcome.
Trends
The terrestrial constraint is the whole 'why space' pull — orbital solar + vacuum heat-sink is the pitch.
The linchpin — necessary for the economics, but ~$200-500/kg is a forward assumption, not today's ~$1,500/kg.
Legitimizes the category but also threatens to crowd out the venture pure-plays.
FCC spectrum + debris review, Kessler/light-pollution concerns, ITAR friction on an 88,000-sat plan.
Enables the mega-financing a capital-intensive roadmap needs — but reverses fast in a sentiment drawdown.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Starship for the thesis) Launch provider today; Starship's future $/kg is the curve the whole economic case depends on. PRIVATE.
GPU supplier (H100 flown, Blackwell targeted) + Inception/ecosystem backer; receives radiation soft-error telemetry.
An AWS server blade is a planned Starcloud-2 payload (announced).
First named commercial partner — 'Crusoe Cloud' on a Starcloud satellite (late 2026); limited GPU capacity in space from early 2027. Demo scale, not yet revenue.
Hyperscaler moonshot — TPUs in LEO with optical links, published research (arXiv 2511.19468), 2-satellite Planet Labs demo targeted early 2027. Unlimited capital + in-house silicon; the strongest 'category is real' validator and the biggest crowd-out threat. The only clean US-listed optionality on the theme.
FCC filing (~Mar 2026) for ~51,600 orbital-DC satellites + Bezos feasibility framing. Deep-pocketed with in-house heavy-lift (New Glenn); vision-stage, nothing built. PRIVATE — not part of Amazon.
Space-station prime extending into Orbital Data Center nodes (edge-compute-for-satellites); ISS-hosted AxDCU-1 (fall 2025) + early LEO nodes (Jan 2026). More grounded near-term niche, but a program line inside a capital-hungry parent.
Early-stage modular self-cooling 'TILE' compute units; ~$22M raised, a 2027 Apex demo flight targeted. Most architecturally novel, least proven.
Ruggedizes commodity servers (HPE Edgeline) for LEO via a SatFrame bus; least-funded, no orbital flight yet.