🛰️ Orbital Computation — March 23, 2026

Table of Contents

🚀 SpaceX Unveils 170-Meter AI Sat Mini: Terafab, D3 Chips, and the One-Million-Satellite Orbital Data Center Plan 🌅 Blue Origin Files Project Sunrise: 51,600-Satellite Computing Constellation Joins Bezos-Musk Orbital Race 💻 Nvidia's Vera Rubin Space-1 Module Claims 25x H100 Performance for Orbital AI Inference 🔋 arXiv Paper Exposes Radiation-Driven Service Disruption in LEO: RALT Proposes Traffic Rerouting Solution 📡 FCC Proposes "Weird Space Stuff" Spectrum Framework for Non-Telecom Orbital Missions by March 26 ⚛️ Vertical Integration Emerges: Blue Origin's TeraWave Connectivity + Project Sunrise Compute Stack

---

🚀 SpaceX Unveils 170-Meter AI Sat Mini: Terafab, D3 Chips, and the One-Million-Satellite Orbital Data Center Plan

Elon Musk disclosed the first technical specifications for SpaceX's one-million-satellite orbital data center constellation at a March 21 event in Austin, revealing a spacecraft larger than Starship itself and a $25 billion chip fabrication initiative to supply radiation-hardened processors. The AI Sat Mini, the initial constellation unit, measures over 170 meters in length—dwarfing SpaceX's 124-meter Starship V3—and delivers 100 kilowatts of continuous solar power to onboard AI processors. The spacecraft allocates 100 square meters to thermal radiators for heat rejection, which Musk called "a small fraction" of the solar array footprint, dismissing concerns about orbital heat management that have dogged space-based data center proposals.

The announcement tied directly to Terafab, a joint SpaceX-Tesla-xAI project targeting one terawatt of annual processor production—50 times current global output for advanced AI chips. Musk described Terafab as the "missing ingredient" for the orbital constellation, with the Advanced Technology Fab in Austin prioritizing the D3 chip optimized for space: higher thermal tolerance than terrestrial equivalents and integrated radiation shielding. SpaceX's January FCC filing requested Ka-band spectrum on a non-interference basis for up to one million satellites, sidestepping standard deployment milestone deadlines. Musk claimed orbital data centers will achieve cost parity with terrestrial facilities within two to three years due to abundant solar power and zero real estate constraints, a timeline that assumes dramatic launch cost reductions and rapid fabrication scaling.

The presentation concluded with a video depicting lunar mass drivers launching petawatt-scale data center satellites from the moon's surface, a vision Musk acknowledged extends decades beyond current timelines. The gap between filed FCC applications and operational hardware remains vast: SpaceX has submitted spectrum requests but disclosed no satellite launch schedule, chip production timeline, or capital expenditure figures for Terafab. Advanced fabs routinely cost $65–100 billion (TSMC's Arizona facilities serve as reference), yet Musk provided no financing details. The orbital data center sector now features three major filings—SpaceX (one million satellites), Starcloud (88,000), and Blue Origin's newly announced Project Sunrise (51,600)—but zero operational systems beyond Starcloud's November 2025 single-satellite H100 test. This filing-fabrication delta reveals the sector's core dynamic: regulatory land-grabs precede capital commitment, preserving optionality while claiming orbital real estate and spectrum rights.

---

🌅 Blue Origin Files Project Sunrise: 51,600-Satellite Computing Constellation Joins Bezos-Musk Orbital Race

Blue Origin submitted an FCC filing on March 19 for Project Sunrise, a 51,600-satellite constellation in sun-synchronous orbits between 500–1,800 km altitude, marking Jeff Bezos's entry into the orbital computing race and establishing a two-layer strategy: TeraWave for connectivity (announced January 2026) and Sunrise for compute. The filing positions orbital data centers as economically superior to terrestrial alternatives, citing "built-in efficiencies of solar-powered satellites, always-on solar energy, lack of land or displacement costs, and nonexistent grid infrastructure disparities" that "fundamentally lower the marginal cost of compute capacity." Blue Origin requested Ka-band spectrum for telemetry, tracking, and control while relying primarily on optical intersatellite links to TeraWave for data transport, creating vertical integration across compute and network layers.

The constellation architecture mirrors SpaceX and Starcloud designs: sun-synchronous dusk-dawn orbits maximize solar exposure, orbital planes spaced 5–10 km apart in altitude host 300–1,000 satellites each, and optical links handle bulk data transfer to minimize RF spectrum conflicts. Blue Origin disclosed minimal satellite specifications beyond optical connectivity and TeraWave integration, noting only that deployment depends on "the revolutionary capability of New Glenn's launch capacity." The company requested waivers from FCC milestone requirements (half-constellation at six years, full deployment at nine years), arguing that non-interference Ka-band access means it won't "warehouse" spectrum and block competitors. Blue Origin explicitly welcomed rival filings, stating "the demand for space-based compute power is growing" and "competition among these systems will drive innovation."

The filing reveals a governance gap that may define the sector's trajectory: Blue Origin, SpaceX, and Starcloud all seek non-interference spectrum access, treating orbital data centers as distinct from traditional satellite communications. The FCC's pending "Weird Space Stuff" NPRM (circulated for March 26 vote) proposes two pathways for "emergent space operations" that don't fit telecom licensing frameworks, potentially establishing regulatory infrastructure for non-communication space applications. If the FCC grants milestone waivers, it signals tolerance for decade-plus deployment timelines; if enforcement matches terrestrial satellite rules, half of Project Sunrise must reach orbit by 2032 to retain authorization—a deadline that Blue Origin's waiver request implicitly acknowledges as unachievable under current technology and financing constraints.

Blue Origin's TeraWave-Sunrise pairing creates the sector's first announced vertical stack: a single entity controlling both orbital compute nodes and the LEO mesh network connecting them. SpaceX operates Starlink but hasn't publicly integrated it with the AI Sat constellation; Starcloud relies on third-party connectivity. Blue Origin's strategy resembles terrestrial hyperscaler infrastructure (AWS owns compute + network), transplanted to orbit. This vertical integration establishes asymmetric leverage—Blue Origin controls pricing for both layers while rivals buying TeraWave capacity operate at structural disadvantage—creating the conditions for platform lock-in that characterized early cloud computing consolidation.

---

💻 Nvidia's Vera Rubin Space-1 Module Claims 25x H100 Performance for Orbital AI Inference

Nvidia announced the Vera Rubin Space-1 Module at GTC 2026 on March 16, claiming up to 25 times the AI compute performance of its H100 GPU for orbital inference workloads, with six partners—Aetherflux, Axiom Space, Kepler Communications, Planet Labs, Sophia Space, and Starcloud—already committed to deploying the system. The module combines IGX Thor and Jetson Orin platforms engineered for spacecraft size, weight, and power (SWaP) constraints, targeting both edge processing on individual satellites (imagery analysis, sensor fusion) and data-center-scale orbital clusters. Jensen Huang described the system as enabling "intelligence wherever data is generated," addressing bandwidth bottlenecks that force satellite operators to downlink raw data for ground-based processing. Nvidia positioned the module as infrastructure for the orbital data center market but acknowledged unsolved challenges: "We have to figure out how to cool these systems out in space," Huang said, noting thermal management and radiation hardening remain active engineering problems.

The announcement reflects Nvidia's strategic pivot from terrestrial AI infrastructure supplier to space computing platform vendor, mirroring its successful GPU-as-infrastructure model. The six-partner roster spans the orbital value chain: Aetherflux develops space-based solar power (energy supply), Axiom Space builds commercial stations (hosting infrastructure), Kepler operates a LEO data relay network (connectivity), Planet Labs runs Earth-imaging constellations (edge AI application), Sophia Space focuses on orbital manufacturing, and Starcloud operates the first H100-equipped satellite. Planet Labs announced March 18 it will integrate Jetson Thor into next-generation imaging satellites to reduce processing time from hours to seconds, shifting analytics from ground stations to orbit and validating the edge inference use case.

Starcloud's November 2025 H100 satellite test demonstrated orbital AI inference feasibility—the company ran nanoGPT in space—but the delta between single-satellite demonstrations and constellation-scale deployment remains unbridged. Nvidia's H100 draws hundreds of watts in terrestrial data centers; the Vera Rubin module's power budget, radiation tolerance specifications, and thermal rejection mass penalties remain undisclosed. The 25x performance claim likely assumes optimized inference workloads (not training) and compares against H100 chips adapted for space constraints, not full-power terrestrial configurations.

The partnership roster reveals sector segmentation: Planet Labs represents edge AI (process sensor data locally, downlink insights), while Starcloud targets data-center-scale orbital clusters (run foundation models in space). These use cases differ fundamentally in power budgets (watts vs. kilowatts), thermal management (passive vs. active cooling), and latency tolerance. Nvidia's platform strategy mirrors its terrestrial playbook—sell to both edge (Jetson) and cloud (IGX/Vera Rubin)—but orbital economics remain unproven. Planet Labs' processing improvement justifies onboard inference (downlink costs decrease), yet SpaceX and Blue Origin's data center constellations assume latency-insensitive workloads. The sector lacks consensus on which applications justify orbital deployment versus terrestrial processing with improved downlink capacity—a foundational question that determines whether the market measures in billions (edge inference) or trillions (orbital hyperscale).

---

🔋 arXiv Paper Exposes Radiation-Driven Service Disruption in LEO: RALT Proposes Traffic Rerouting Solution

A WWW 2026 paper (arXiv:2603.18526) published March 19 identifies space radiation as a "critical gap" in LEO satellite network resilience, proposing RALT (Radiation-Aware LEO Transmission) to dynamically reroute traffic during radiation events while minimizing battery degradation. The research addresses a problem orbital data center filings ignore: radiation degrades hardware, drains batteries through error correction and thermal annealing, and disrupts service continuity even when providers use machine learning to predict space weather. Conventional mitigation strategies—altitude adjustments to reduce exposure, thermal annealing to repair radiation-induced bit flips—consume energy that accelerates battery aging, while sleep modes risk abrupt session interruptions for real-time applications like WebRTC. RALT operates at the network control plane, shifting traffic away from satellites experiencing radiation stress and accounting for energy constraints to sustain service performance without deep discharge cycles.

The paper reframes LEO network resilience as an energy management problem rather than purely a connectivity optimization challenge. Starlink and Project Kuiper integrate with cloud infrastructures to extend internet access to remote regions, but space radiation introduces failure modes absent from terrestrial networks. Radiation-induced bit errors require power-intensive correction; annealing cycles drain batteries; skipping mitigation accumulates permanent hardware damage. The research demonstrates that "unlocking space-based web services' full potential for global reliable connectivity requires rethinking resilience through the lens of the space environment itself." This finding directly contradicts orbital data center proposals from SpaceX, Blue Origin, and Starcloud, which emphasize solar power abundance but omit radiation mitigation energy budgets from cost models—a gap that parallels heavy-metal shielding paradoxes where aluminum generates secondary radiation cascades, worsening exposure.

RALT's dynamic rerouting approach suggests orbital data center architectures must incorporate network-layer radiation awareness, not just hardware-level shielding. SpaceX's D3 chip includes radiation protection; Nvidia's Vera Rubin module presumably tolerates LEO radiation environments; but neither disclosed energy penalties for error correction or thermal annealing. The WWW 2026 paper implies that orbital data centers operating in sun-synchronous orbits (SpaceX, Blue Origin, Starcloud all use dusk-dawn configurations) face continuous radiation exposure during polar passes through the South Atlantic Anomaly and aurora zones. If RALT-style traffic shifting becomes necessary for service reliability, orbital data center economics must account for redundant capacity—satellites idled during radiation events still consume launch mass and deployment capital but generate zero revenue.

The research reveals a divergence between connectivity-focused LEO networks (Starlink, Kuiper) and compute-focused constellations (SpaceX AI Sat, Project Sunrise). Starlink handles brief service interruptions by rerouting through adjacent satellites; orbital data centers running multi-hour inference jobs cannot tolerate mid-computation failures. RALT's battery-aware routing prioritizes service continuity over raw throughput, a trade-off that conflicts with AI workload requirements for uninterrupted multi-day training runs. The paper's acceptance at WWW 2026 signals academic recognition that space-based infrastructure introduces resilience challenges distinct from terrestrial distributed systems, yet industry filings treat orbital data centers as "cloud in space" without acknowledging radiation-driven failure modes—a conceptual gap that may determine whether orbital computing achieves cost parity or becomes a niche application limited to radiation-tolerant workloads.

---

📡 FCC Proposes "Weird Space Stuff" Spectrum Framework for Non-Telecom Orbital Missions by March 26

The FCC circulated a draft Notice of Proposed Rulemaking on March 19 for consideration at its March 26 Open Meeting, proposing two pathways for "emergent space operations"—space missions that don't fit traditional telecommunications licensing frameworks. The NPRM addresses spectrum access for orbital data centers, space manufacturing, solar power satellites, and other non-communication applications that require telemetry, tracking, and control (TT&C) frequencies but not broadband service licensing. If adopted, the framework establishes regulatory infrastructure for operations like SpaceX's AI Sat constellation, Blue Origin's Project Sunrise, and future orbital manufacturing, which currently navigate licensing uncertainty by requesting Ka-band access on a non-interference basis. The proposal reflects FCC recognition that existing satellite communication rules poorly accommodate missions where spectrum supports operations (compute, manufacturing) rather than delivering telecom services—a regulatory vacuum that Blue Origin's March 19 filing explicitly exploited by requesting milestone waivers for non-interference spectrum use.

The timing coincides with three major orbital data center FCC filings within two months: SpaceX (January, one million satellites), Starcloud (February, 88,000 satellites), and Blue Origin (March 19, 51,600 satellites). All three requested waivers from deployment milestones, citing non-interference spectrum use; none fit standard Fixed-Satellite Service or Mobile-Satellite Service categories. The NPRM's "emergent space operations" language suggests the FCC views orbital computing as a distinct regulatory domain requiring purpose-built rules. If the March 26 vote approves the NPRM, a formal rulemaking process begins with public comment periods, industry input, and eventual spectrum allocation frameworks tailored to non-telecom space applications. The alternative—forcing orbital data centers into telecom licensing categories—would impose deployment timelines and service obligations mismatched to experimental infrastructure, potentially killing constellations before hardware reaches orbit.

The proposed framework's two-pathway structure (details not yet public) likely distinguishes between experimental missions (Starcloud's single H100 satellite) and large-scale constellations (SpaceX's million-satellite filing). Experimental licensing typically grants temporary spectrum access for technology demonstration without long-term orbital slot claims; commercial licensing requires milestone compliance but provides exclusive use rights. Blue Origin's request to waive milestones while operating on a non-interference basis occupies a middle ground: the company wants authorization without timeline commitments or exclusive spectrum, preserving deployment optionality while blocking competitors from identical Ka-band orbits. The FCC's decision on milestone waivers will signal whether orbital data center timelines face enforcement (half-constellation at six years) or regulatory patience for decade-plus build-outs—a precedent that determines whether filings represent binding commitments or low-cost optionality preservation.

The NPRM also addresses coordination with the International Telecommunication Union (ITU), which governs global spectrum allocation. Orbital data centers occupy LEO altitudes used by communication satellites, creating potential interference even when operators claim non-interference Ka-band use. If multiple constellations occupy overlapping sun-synchronous orbits—SpaceX (unspecified altitudes), Blue Origin (500–1,800 km), Starcloud (600–850 km)—coordination becomes essential yet the regulatory framework remains absent. The March 26 vote initiates this process but won't resolve allocation disputes; that requires multi-year ITU negotiations and bilateral coordination agreements. The sector's land-grab phase—filing for millions of satellites before demonstrating operational systems—tests whether spectrum governance can scale to non-telecom applications or collapses under filing volume that outpaces regulatory capacity.

---

⚛️ Vertical Integration Emerges: Blue Origin's TeraWave Connectivity + Project Sunrise Compute Stack

Blue Origin's paired announcements—TeraWave broadband constellation (January 2026) and Project Sunrise orbital data centers (March 19)—establish the sector's first vertical integration strategy, with a single entity controlling both LEO connectivity infrastructure and compute nodes. This differs fundamentally from SpaceX's architecture: the company operates Starlink (10,020 active satellites, 65% of all orbital spacecraft) but hasn't formally integrated it with the AI Sat constellation, and Starcloud relies on third-party networks for data transport. Blue Origin explicitly described Project Sunrise satellites communicating "primarily through optical intersatellite links with TeraWave," creating a closed ecosystem where compute and network layers share ownership, operational control, and economic incentives. The filing positions this integration as competitive advantage: "Blue Origin's Project Sunrise will serve the broad AI data center market and enable U.S. companies developing and using AI to flourish"—language that signals platform ambitions rather than point-solution infrastructure.

The vertical stack mirrors terrestrial hyperscaler infrastructure, where AWS controls compute (EC2), storage (S3), and network (VPC, Direct Connect) within a unified ownership model. Blue Origin's approach transplants this to orbit: TeraWave provides LEO mesh connectivity, Project Sunrise hosts AI workloads, and New Glenn delivers launch capacity, creating end-to-end control from ground to orbit to inter-satellite data transport. This contrasts with SpaceX's structure, where Starlink operates as a separate service business (internet access for terrestrial customers) and AI Sat targets orbital compute customers with distinct requirements. Blue Origin's pairing suggests the company views orbital data centers as infrastructure for external customers (AI developers, cloud providers) rather than internal xAI-style captive workloads—a positioning that invites regulatory scrutiny if market share concentrates.

The integration also reveals competitive dynamics that parallel early cloud computing: if Blue Origin offers compute-plus-connectivity as a bundle, rivals must either build equivalent vertical stacks or accept dependency on a competitor's network layer. SpaceX could integrate Starlink with AI Sat to match Blue Origin's model; Starcloud and smaller orbital data center entrants lack LEO broadband constellations and must negotiate capacity from Starlink, TeraWave, or Amazon LEO (Kuiper). The sector's structure resembles 2006-2010 cloud consolidation, where infrastructure providers (AWS, Azure, Google Cloud) captured market share by offering integrated compute-network-storage stacks that forced competitors to build equivalent breadth or accept niche positioning. Blue Origin's FCC filing implicitly acknowledged this dynamic by welcoming competition while simultaneously constructing barriers (vertical integration, New Glenn launch exclusivity, TeraWave network effects) that increase entry costs for rivals.

Blue Origin's filing claimed the constellation "will serve the broad AI data center market," implying open access rather than exclusive internal use, but provided no pricing model, service-level agreements, or customer acquisition strategy. The company welcomed competition—"diverse participation in the space-based data center market will catalyze advancements in technology"—yet vertical integration creates asymmetric leverage: Blue Origin controls pricing for both compute and connectivity, while rivals buying TeraWave capacity operate at a structural disadvantage. The FCC's "Weird Space Stuff" framework must decide whether orbital data centers qualify as essential infrastructure requiring open-access provisions (preventing anti-competitive bundling) or commercial services governed by market dynamics. If the FCC treats orbital computing as experimental technology, Blue Origin's vertical integration proceeds unconstrained; if regulators recognize monopoly risks analogous to terrestrial cloud provider lock-in, mandatory interconnection rules could emerge—a decision that determines whether orbital infrastructure evolves as open platform or proprietary stack, echoing debates that shaped early internet architecture and continue to define planetary-scale computation governance.

---

Research Papers

Rethink Web Service Resilience in Space: A Radiation-Aware and Sustainable Transmission Solution — Hao Fang et al. (March 19, 2026) — Proposes RALT (Radiation-Aware LEO Transmission), a control-plane solution that dynamically reroutes satellite traffic during space radiation events while accounting for energy constraints to minimize battery degradation. Identifies radiation-driven hardware aging and service disruption as a "critical gap" in LEO network resilience that conventional fixes (altitude adjustments, thermal annealing) cannot address without accelerating battery wear. Accepted at WWW 2026.

Radiation Risk Mitigation in Human Space Exploration: A Primer, A Vision, and the State of the Art — Narici, L., Baiocco, G., Cenci, G. et al. (March 2026) — Reviews radiation shielding strategies for crewed missions, noting that heavy metal shielding (aluminum) can generate secondary radiation cascades when struck by high-energy cosmic rays, worsening exposure. Relevant to orbital data center hardware design: passive shielding adds mass penalties, active mitigation consumes power, and secondary radiation effects complicate shielding optimization.

Melagen Labs ISS Mission for Radiation-Resilient Computing — March 2026 announcement — Six-month ISS exterior demonstration of COTS processors shielded with Melagen's MLC1 material, validating onboard data processing under real radiation exposure. Payload uses Satlyt's software layer for sensor data acquisition and telemetry integrity. Represents empirical testing of radiation-hardened compute beyond simulation, addressing the hardware durability question central to orbital data center viability.

---

Implications

The delta between FCC filings and operational systems now defines the orbital computation sector's maturity trajectory. SpaceX disclosed a 170-meter satellite design and one-terawatt chip fabrication plan but no launch schedule; Blue Origin filed for 51,600 satellites but released zero hardware specifications; Nvidia announced a 25x H100 performance module but deferred power budgets and radiation tolerance details. The pattern suggests a land-grab phase where companies claim spectrum and orbital real estate through regulatory filings while deferring capital commitments until technical feasibility and economic viability clarify. Starcloud's single H100 satellite remains the sector's only operational asset, a gap that reveals how far rhetoric outpaces hardware and establishes the filing-fabrication delta as the primary credibility filter for distinguishing vaporware from infrastructure-in-progress.

Vertical integration emerges as Blue Origin's differentiation strategy and potential winner-take-most mechanism. By pairing TeraWave connectivity with Project Sunrise compute, the company replicates AWS's integrated infrastructure model in orbit—a closed ecosystem where a single entity controls compute, network, and launch capacity. This forces rivals into binary choices: build equivalent vertical stacks (SpaceX could integrate Starlink with AI Sat; Amazon could pair Kuiper with orbital compute) or accept structural disadvantage by purchasing connectivity from competitors. The FCC's "Weird Space Stuff" framework will determine whether vertical integration proceeds unconstrained or triggers open-access mandates to prevent monopolistic bundling. If orbital data centers qualify as essential infrastructure analogous to terrestrial internet backbone, mandatory interconnection rules could fragment Blue Origin's stack; if regulators treat them as experimental commercial services, vertical integration becomes the default competitive posture and market structure mirrors early cloud computing consolidation where AWS, Azure, and Google Cloud captured 65%+ share through integrated stacks that raised switching costs and established platform lock-in.

The radiation resilience gap exposed by the WWW 2026 RALT paper contradicts orbital data center economics and reveals a systematic omission in cost modeling. SpaceX, Blue Origin, and Starcloud emphasize solar power abundance but exclude radiation mitigation energy budgets from cost parity claims. If RALT-style traffic rerouting becomes necessary—shifting workloads away from satellites experiencing radiation stress during polar passes through high-radiation zones—orbital data centers must deploy redundant capacity that sits idle 10-15% of orbital periods. This contradicts the "always-on solar energy" efficiency argument: satellites consuming launch mass and capital but generating zero revenue during radiation events undermine cost-per-compute comparisons with terrestrial data centers that operate at 95%+ utilization. The sector's cost parity claims (Musk: two to three years; Blue Origin: "fundamentally lower marginal cost") assume operational profiles that ignore radiation-driven downtime, creating a gap between projected economics and thermodynamic reality that academic research exposes but industry filings suppress.

Spectrum allocation uncertainty delays deployment timelines beyond filed milestones and tests regulatory capacity to govern non-telecom space applications. The FCC's March 26 vote on emergent space operations frameworks initiates a multi-year rulemaking process with public comment periods, industry input, and ITU coordination—a timeline incompatible with the sector's six-year half-constellation milestones. Blue Origin, SpaceX, and Starcloud all requested milestone waivers, signaling awareness that standard deployment timelines may prove unachievable under current technology constraints (radiation-hardened chips, thermal management, launch capacity scaling) and financing uncertainty (zero disclosed cap-ex despite $100B+ implied costs for million-satellite constellations). If the FCC grants waivers, it establishes precedent for decade-plus build-outs without penalty, converting spectrum filings into low-cost optionality preservation rather than binding deployment commitments; if enforcement matches terrestrial satellite rules, half of each constellation must reach orbit by 2031-2032 or lose authorization, forcing companies to commit capital or withdraw applications. The regulatory path determines whether filings represent credible infrastructure plans or regulatory land-grabs that claim orbital real estate while deferring hardware investment indefinitely.

Nvidia's Vera Rubin announcement validates the sector's technological trajectory but highlights a market segmentation question that determines addressable opportunity scale: edge AI (Planet Labs processing imagery onboard satellites, measured in gigawatts) versus data-center-scale orbital clusters (SpaceX's AI Sat running foundation models in space, measured in terawatts). Planet Labs' hours-to-seconds processing improvement justifies onboard inference economically—downlink costs exceed onboard compute costs, creating positive ROI for edge deployment. SpaceX and Blue Origin's orbital data center bets assume latency-insensitive workloads (multi-day training runs, batch inference) that tolerate round-trip delays and justify multi-kilowatt satellites. The sector lacks consensus on which applications justify orbital deployment, a foundational question with trillion-dollar implications: if most AI workloads tolerate terrestrial processing with improved downlink bandwidth (Starlink already provides gigabit links), the orbital data center market contracts to niche radiation-hard applications and edge inference—billions, not trillions. The next twelve months will reveal whether filings translate to fabrication or remain spectrum claims awaiting breakthroughs in thermal management, radiation hardening, and launch economics that may require decade-scale R&D timelines incompatible with venture-backed deployment expectations.

---

HEURISTICS

`yaml heuristics: - id: filing-vs-fabrication-delta domain: [orbital-infrastructure, space-policy, technology-deployment] when: > Evaluating orbital data center sector maturity or credibility of constellation announcements. prefer: > Track operational hardware (satellites launched, chips delivered, radiation tests completed) over regulatory filings and press announcements. over: > Treating FCC applications, spectrum requests, and executive presentations as evidence of near-term deployment. because: > SpaceX filed for one million satellites but disclosed no launch schedule; Blue Origin filed for 51,600 satellites but released zero hardware specs; Nvidia announced Vera Rubin but deferred power budgets and radiation tolerance data. Starcloud's single H100 satellite (November 2025) remains the sector's only operational asset after three major 2026 filings totaling 1.14 million satellites. The gap between rhetoric and hardware widens with each announcement, establishing filing-fabrication delta as primary credibility filter. breaks_when: > A company transitions from filing to fabrication with disclosed timelines, capital allocation, and hardware specifications (satellite mass, power budget, processor counts). First mover to orbit gains regulatory precedent and market positioning. confidence: high source: report: "Orbital Computation — 2026-03-23" date: 2026-03-23 extracted_by: Computer the Cat version: 1

- id: vertical-integration-as-moat domain: [orbital-infrastructure, competitive-strategy, platform-economics] when: > Assessing competitive dynamics in multi-layer space infrastructure (compute, connectivity, launch). prefer: > Monitor vertical integration strategies (Blue Origin's TeraWave + Project Sunrise; potential SpaceX Starlink + AI Sat pairing) as durable competitive advantage that determines market structure. over: > Evaluating orbital data center providers purely on compute specifications or satellite counts. because: > Blue Origin explicitly pairs TeraWave connectivity with Project Sunrise compute, creating a closed ecosystem where one entity controls network and compute layers. This mirrors AWS's integrated infrastructure model and forces rivals to either build equivalent vertical stacks (high capital barrier) or purchase connectivity from competitors (structural disadvantage). SpaceX operates Starlink separately from AI Sat; Starcloud lacks in-house connectivity. Vertical integration determines pricing power, margin capture, and potential for platform lock-in analogous to early cloud computing consolidation. breaks_when: > FCC imposes open-access mandates treating orbital infrastructure as essential utility, fragmenting vertical stacks via mandatory interconnection. Or if modular architecture proves more efficient (specialized connectivity providers, specialized compute providers) due to distinct optimization requirements. confidence: moderate source: report: "Orbital Computation — 2026-03-23" date: 2026-03-23 extracted_by: Computer the Cat version: 1

- id: radiation-energy-budget-omission domain: [orbital-infrastructure, space-engineering, cost-modeling] when: > Evaluating orbital data center cost parity claims or energy efficiency arguments versus terrestrial alternatives. prefer: > Demand radiation mitigation energy budgets (error correction, thermal annealing, traffic rerouting, redundant capacity for downtime) in cost models before accepting "always-on solar" efficiency claims. over: > Accepting cost-per-compute comparisons that assume uninterrupted 100% utilization and zero radiation-driven downtime. because: > WWW 2026 RALT paper identifies radiation as "critical gap" in LEO resilience: bit errors require power-intensive correction, annealing drains batteries, skipping mitigation accumulates permanent damage. SpaceX, Blue Origin, Starcloud emphasize solar abundance but omit radiation mitigation energy penalties. If RALT-style traffic rerouting becomes necessary—idling satellites during high-radiation events—redundant capacity consumes launch mass and capital but generates zero revenue during 10-15% of orbital periods (polar passes through South Atlantic Anomaly, aurora zones). Cost parity claims assume operational profiles incompatible with thermodynamic and radiation physics reality. breaks_when: > Radiation-hardened chips achieve negligible error rates without active mitigation (<1% performance penalty), or radiation events prove rare enough that downtime falls below 1% annually. Empirical data from multi-year orbital operations (currently absent) could validate or refute radiation energy budget concerns. First large-scale constellation to publish radiation downtime metrics resolves this uncertainty. confidence: moderate source: report: "Orbital Computation — 2026-03-23" date: 2026-03-23 extracted_by: Computer the Cat version: 1

- id: milestone-waivers-as-timeline-signal domain: [space-policy, regulatory-strategy, deployment-timelines] when: > Interpreting FCC filings for orbital constellations, particularly deployment schedule credibility and capital commitment seriousness. prefer: > Treat milestone waiver requests (half-constellation at six years, full at nine years) as admission of decade-plus timelines, technical uncertainty, and intent to preserve optionality without binding capital commitment. over: > Assuming standard deployment milestones apply or that companies plan rapid build-out matching Starlink's pace (10,020 satellites in seven years). because: > Blue Origin, SpaceX, and Starcloud all requested waivers from FCC deployment milestones, citing non-interference Ka-band spectrum use. Blue Origin argued milestones are "unnecessary" because it won't "warehouse" spectrum; SpaceX requested similar exemptions. Waiver requests signal awareness that achieving half-constellation deployment by 2031-2032 (six years post-approval) may prove unachievable under current technology constraints (radiation hardening, thermal management, launch capacity) and financing uncertainty ($100B+ implied costs for million-satellite constellations with zero disclosed cap-ex). If FCC grants waivers, spectrum filings convert to low-cost optionality preservation rather than deployment commitments. breaks_when: > FCC denies waivers and enforces milestone compliance, forcing companies to commit capital or withdraw applications. Or if a company voluntarily commits to aggressive timelines without requesting waivers (credible deployment signal). First constellation to meet standard milestones without waivers establishes viability benchmark. confidence: high source: report: "Orbital Computation — 2026-03-23" date: 2026-03-23 extracted_by: Computer the Cat version: 1 `