The European orbital compute thesis

Above the atmosphere, solar is roughly 8× more productive, heat sheds by radiation, and vacuum is free insulation. The binding constraint on the ground — power, cooling, density — is not the binding constraint in space. The binding constraint in space is energy delivered to compute, and that's a better problem. LEO is ~150× faster than GEO for space-data round-trips, and the regulatory window (ITU 2027, EU Space Law) is being written right now.

Between late 2025 and early 2026, orbital compute went from a slide in an ESA briefing to a product with customers and a launch manifest: Starcloud trained an LLM in space (Dec 2025), Axiom launched the first two ODC nodes to the ISS (Jan 2026), Google "Project Suncatcher" announced an 81-satellite TPUs-in-LEO constellation (early 2027). Three curves are converging — launch cost (10× down in 5 years), AI demand (×2 every 9 months), and the regulatory window. The architectures are being designed right now. Europe has ~18 months to pick a lane.

The natural investor question: OHB has a 40-year track record, Airbus Defence and Space launches sovereign constellations, EnduroSat has the smallsat supply chain locked. They could do orbital compute. They won't. Orbital compute is 100% of our mission and ~1% of theirs; they run 5–10 year mission cycles; the opportunity is an 18-month window. The actual unfair advantage is the AI-native methodology — the AI4CE PhD research applies deep reinforcement learning to mission-architecture design, a capability a 30-year engineering culture cannot replicate.

We get the market because the primes wait for proven demand. The 18-month window is for venture-funded startups to define the architecture — which orbit altitudes, which SLA profiles, which regulatory lane, which anchor workload. Whoever locks the architecture by mid-2027 sets the standards that incumbents will have to comply with. We are racing for the standard, not the compute market. The compute market follows the standard. The primes are not our competitors; they are our potential integrators or acquirers.

TAM. Global data centre market: ~$1.1T by 2035 (Precedence Research, 2026). The orbital-DC addressable slice — LEO + sovereign + AI workloads — is a fraction of that, but a meaningful one.

SAM. Orbital data centre market: ~$35B by 2030 (Markets-and-Markets + Yahoo Finance analyst consensus, 2026). AI data centre slice ~$200–500B by 2032 (Grand View Research, 2026).

SOM. Our slice — European sovereign-AI in our orbit / SLA profile — is ~$2–5B by 2032. That's the lane: one sovereign hyperscaler or public-sector AI programme underwriting the first-constellation use case, in European jurisdiction, on European-controlled infrastructure.

Sources: Precedence Research (2026); Grand View Research (2026); Markets-and-Markets + Yahoo Finance analyst consensus (2026). Full citation list in /story and in the US22 design review (01-space-AI/01-research/site-design-review-2026-06-20.md).

The AI-native methodology isn't a slide — it's a peer-reviewed research programme. The 6 publications that ground the AI4CE work (in 02-ai4ce/04-publications/ on the project repo):

• "AI4CE: Building the Infrastructure for AI-supported CubeSat Generation" (Ceglarek, 2024). The methodology paper — the infra + data + reward-shaping that lets deep RL generate space-mission architectures.
• "Bottom-up AI Support for Conceptual Design" (Ceglarek et al., 2022). The original 2022 framing of AI-assisted design as a co-pilot, not a black box.
• "AI4CE: Automated Space Mission Design Concepts" (Ceglarek et al.). The systems-engineering case for AI-generated designs as a deliverable artefact.
• "AI4CE: Improvements on Generating Space Systems" (Ceglarek et al.). The follow-on: where the methodology broke, what we patched, and what the next iteration unlocks.
• "AI4CE: Towards Benchmarking Deep Reinforcement Learning against non-AI System Generation Methods" . The benchmark paper — the head-to-head comparison of RL-generated architectures vs. classical human-engineered ones, on shared metrics.
• "AI4CE BOTTOM-UP AI SUPPORT FOR CONCEPTUAL DESIGN" (Ceglarek et al., conference version). The peer-reviewed conference paper.

The methodology isn't a slide. It's a 6-paper programme, and the orbital-compute work inherits it. An investor who wants to see the AI-native methodology work can read the papers — that's a credibility signal that survives technical due diligence.

• 120 satellites at 600 km LEO. Continuous coverage above 50° latitude for mid-latitude customer sites. Walker-style orbital plane distribution.
• ~100 TFLOPS per satellite (radiation-hardened, inference-optimised; training runs use multi-sat sharding). 50–200 TFLOPS range depending on workload.
• 5 EU ground stations (Svalbard, Frankfurt, Madrid, Athens, Helsinki). EU jurisdiction, EU-controlled operations, EU-compliant data residency.
• Optical inter-sat links + RF ground link. Downlink 0.1–10 Gbps per satellite; uplink 0.01–1 Gbps.

Total constellation compute: ~12,000 TFLOPS (120 × 100). A meaningful tier of inference capacity sitting in EU jurisdiction, latency-close to the data. Move the sliders on /explore to see the coverage / latency / cost curves shift.

We are at the post-research, pre-architecture stage. The deep-dive is done. The competitive landscape is mapped (11 verified competitor profiles in our research repo). The regulatory path is sketched. The engineering trade-offs are understood. What we need next:

• Capital — €2–5M seed to lock the architecture, file the orbital-spectrum applications, and stand up the simulation stack (the "digital twin" of the four-tier orbit constellation). We are not raising for hardware yet.
• Anchor customer(s) — one European hyperscaler or sovereign-AI programme to underwrite the first-constellation use case. The anchor defines the orbit and the SLA, not the other way around.
• EU / ESA partnership — for the regulatory, launch, and spectrum-permission work. The technical thesis is European; the regulatory thesis is also European; they should ship together.

What €2–5M buys

An investor who can do the math in 10 seconds is an investor who can write a check. Illustrative breakdown (€3M mid-point):

• €1.2M FTE — 4 engineers × 2 years. Mission design + RL ops + regulatory work + EU partnerships. No satellite hardware engineers yet.
• €0.6M spectrum filing + regulatory — ITU filings, EU Space Law readiness, Bundesnetzagentur authorisation, GDPR / AI-Act compliance review.
• €0.4M simulation stack — the "digital twin" of the 4-tier orbit constellation. The training environment for the AI-native mission design loop.
• €0.2M legal + corporate — entity setup, IP assignment, contract templates, accounting.
• €0.6M reserve — 24+ months of runway. Hardware spend is the Series A conversation, not this one.

Total ≈ €3M (illustrative, JP to validate against actual 2026 European rates). The seed funds the architecture lock; the architecture lock justifies the Series A.

Comparable funding rounds

• Starcloud: $170M Series A (March 2026) at a $1.1B valuation. The headline US orbital-AI Series A.
• US orbital-AI competitors (Axiom Space, Kepler Communications, Project Suncatcher, Sophia Space) have raised $100M+ each.

We are raising €2–5M seed — an order of magnitude below the Series A's we've seen in this space. The seed funds the architecture lock that justifies the Series A math.

What we're not raising for

No satellite hardware yet. €2M raised = 0 satellites = 24+ months of pure-software runway. The first flight is post-architecture-lock, post-funding. Every euro goes into the design loop, the regulatory lane, and the anchor customer. Hardware spend is the Series A conversation.

We claim orbital compute is the lane. We claim the 18-month window is real. We claim the AI-native methodology is the unfair advantage. None of those claims is a signed paid contract. Four honest unknowns:

• The anchor customer. We have exploratory conversations with several European hyperscalers and sovereign-AI programmes. We can't name them publicly yet. The SLA-defining conversation happens post-funding; a Letter of Intent from any one of them would 10x the Series A math.
• The launch provider. Trade study pending architecture lock: SpaceX Falcon 9, Rocket Lab Neutron, Arianespace Vega-C. Cost-per-launch is the single biggest CapEx line. The trade study is a 4-week conversation with the 3 vendors.
• The radiation-hardening cost curve. At the inference-optimised compute density we need (100 TFLOPS/sat), the rad-hard component cost vs COTS-with-shielding trade-off is a 12-month vendor conversation. Today's pricing is roughly 10× ground compute per TFLOP. The Series A depends on that closing.
• The first-flight milestone. The first flight is post-architecture-lock, post-funding, post-anchor LOI. Today's site is the methodology + the roadmap; the satellite is the next chapter. The seed gets us to the Series A; the Series A funds the satellite.

The pre-seed is the architecture lock. The seed is the first flight. The Series A is the constellation. Three capital rounds, three milestones, three sets of investors. The ask on this page is the first one.