Orbital Computation · 2026-06-18

Now I have enough material. Let me write the full report.

🛰️ Orbital Computation Watcher — 2026-06-18

🚀 SpaceX AI1 Orbital Data Center Debuts at $75B IPO: 150kW Per Satellite, xAI Integration, Prototype in 2027
🏭 Gigasat in Bastrop: SpaceX's 11M-Square-Foot Factory Must Manufacture AI Satellites at Aerospace Scales Never Attempted
🌍 NASA JPL's NAVI-Orbital Runs Gemma 3 on Loft Orbital YAM-9 — First Autonomous VLM Demonstrated in LEO
🇨🇳 ADA Space "Star Compute": China's Distributed 2,800-Satellite Mesh Bets Against SpaceX's Centralized AI1 Architecture
💰 Bezos Calls 2-3 Year Timeline "Probably Too Ambitious" as Blue Origin Files 51,600-Satellite Project Sunrise
🌡️ IEEE Spectrum: Thermal Radiator Degradation Forces 40% Mass Penalty — Thermodynamics as Orbital AI's Hard Ceiling

---

🚀 SpaceX AI1 Orbital Data Center Debuts at $75B IPO: 150kW Per Satellite, xAI Integration, Prototype in 2027

SpaceX CEO Elon Musk revealed the AI1 orbital data center satellite to X on June 9, four days before the company's June 13 Nasdaq debut under ticker SPCX — the largest IPO in Wall Street history at $135 per share, raising approximately $75 billion. The sequencing was deliberate: AI1 is the asset SpaceX sold investors on, and the specifications are designed to make the argument concrete in ways a prospectus alone cannot.

Each AI1 spacecraft delivers 150 kW peak / 120 kW sustained compute, spans 70 meters tip-to-tip — wider than a Boeing 747 — and stands 20 meters tall when deployed. Community analysis mapping specs to terrestrial equivalents finds that 150 kW is roughly one NVIDIA GB300 rack, approximately 72 GPUs, packaged inside a structure larger than any satellite SpaceX has previously launched. That arithmetic surfaces the core tension: compute density per dollar launched is currently far below ground-based alternatives, and the economics only close if Starship drives launch costs toward negligible levels while thousands of units operate simultaneously in constellation.

The xAI acquisition — SpaceX absorbed Musk's AI startup earlier in 2026 — was positioned as operationally necessary for the orbital compute thesis. xAI integration creates a vertical stack where SpaceX controls the launch vehicle, satellite bus, compute payload, and the Grok inference model itself — every layer of value extraction in one corporate envelope. Data Center Dynamics reports that SpaceX has already struck major compute-rental deals with Anthropic and Google for its terrestrial data centers — the commercial template it intends to replicate in orbit, with the IPO providing capital to execute the transition.

The gap between filed applications and operational systems remains the decisive metric. Prototype AI1 units target early 2027. Space.com's analysis notes that "orbital data centers are mostly notional, and not actually demonstrated by operating tech — at least yet" — a qualifier that applies to SpaceX's roadmap as much as competitors'. The $75 billion raised provides capital to close that gap, but the 1 GW by late 2027 target requires manufacturing scale that does not yet exist. What SpaceX revealed June 9 is not a product; it is a prospectus for an infrastructure layer whose timeline will be decided by Starship cadence, radiator physics, and industrial execution across a greenfield factory site.

Sources:

---

🏭 Gigasat in Bastrop: SpaceX's 11M-Square-Foot Factory Must Manufacture AI Satellites at Aerospace Scales Never Attempted

SpaceX's Gigasat facility is the industrial counterpart to the AI1 reveal: a 1,000-acre site in Bastrop County, Texas, with over 11 million square feet of planned building space. TechRadar's coverage framed the target precisely: 1 GW per year of space AI compute by late 2027. At 120 kW sustained per satellite, hitting that target requires manufacturing and launching approximately 8,333 AI1 units annually. That number is the central implausibility in SpaceX's published roadmap.

Each AI1 satellite carries a 70-meter wingspan. Building 8,333 per year represents a scale of aerospace production with no historical precedent — for reference, the entire global commercial satellite industry launches hundreds of spacecraft annually across all operators combined. USA Today reports that the Gigasat facility is expected to be operational by end of 2027, meaning SpaceX has approximately 18 months to construct, tool, and ramp an 11-million-square-foot satellite factory from a greenfield site. The Tesla Gigafactory analogy is intentional — the name signals SpaceX views orbital data center manufacturing as an equivalent industrial ramp problem to electric vehicle production, and it is betting that the same playbook transfers.

The vertical integration logic is explicit. SpaceX controls the launch vehicle (Starship, Boca Chica), the satellite bus (proven Starlink architecture), the compute payload (AI1 with NVIDIA hardware), the connectivity backhaul (Starlink inter-satellite links), and now the manufacturing facility. TSPAsemiconductor's analysis noted that this full-stack control resembles Apple's vertical integration model — value captured across every layer rather than at a single point in the supply chain — but unlike Apple, SpaceX's product requires launching physical mass to orbit on a cadence that has never been demonstrated.

The binding constraint is Starship launch cadence. Quasa's analysis points out that manufacturing 8,000+ AI1 satellites annually is irrelevant without launches at equivalent scale. The orbital data center thesis is downstream of the reusable heavy-lift thesis — Gigasat's 1 GW target is a bet that Starship achieves the cadence necessary to make the factory economically meaningful rather than a monument to ambition. That dependency is the most material execution risk embedded in SpaceX's $75 billion IPO, and it is the one risk the AI1 reveal was conspicuously quiet about.

Sources:

---