Orbital Computation · 2026-06-13

🛰️ Orbital Computation — 2026-06-13

🛰️ SpaceX AI1: 70-Meter Wingspan, D3 Chip, and Gigasat Factory Herald IPO-Week Orbital Compute Push
📡 FCC Removes Amazon Leo's July Deadline, Deploys Spectrum-Priority Demotion as Enforcement Substitute
💰 SemiAnalysis: Orbital Compute Costs 4× Terrestrial in 2026; Cost-Parity Model Lands Near 2040
🇨🇳 China's $295B Terrestrial Grid and Operational Orbital Constellations Widen the Deployment Gap
🚀 Orbital Startup Closes $5M a16z-Backed Pre-Seed Round Targeting 10,000 AI Satellites
🏗️ Northwood Space CEO at Fortune Brainstorm Tech: Ground Routing Is the Orbital Economy's Unbuilt Layer

---

🛰️ SpaceX AI1: 70-Meter Wingspan, D3 Chip, and Gigasat Factory Herald IPO-Week Orbital Compute Push

Elon Musk released a factory-floor video on X on June 8 laying out plans for AI1 — SpaceX's first-generation orbital data-center satellite — and the timing was deliberate. The reveal came the week of SpaceX's highly anticipated IPO, targeting a valuation near $1.75 trillion, giving investors their clearest technical look yet at the program central to that number.

AI1's architecture is built around a deceptively simple constraint: solar power is abundant in LEO, but heat rejection is not. At 120 kW sustained and 150 kW peak, the satellite must radiate roughly the thermal equivalent of two ISS External Thermal Control Arrays — achieved through 110 m² of deployable liquid radiators with redundant pumping loops and integrated micrometeoroid shielding. The satellite stands 20 meters tall with a 70-meter wingspan — wider than a Boeing 747 — and operates at approximately 600 km altitude. Compute density is 70 kW per ton, reflecting how aggressively SpaceX has weight-optimized the payload bus.

The D3 chip is the compute engine: a custom SpaceX processor in the lineage of AI5 (Tesla FSD) and AI6 (Cybercab), designed to run hotter than conventional silicon and hardened against LEO radiation levels. Unlike traditional radiation-hardened processors priced at $5,000–$50,000 per chip, D3 targets commercial-foundry economics. If it survives the radiation environment at scale, the rad-hard semiconductor industry faces structural disruption for LEO applications. The chip payload is architecturally interchangeable — third-party accelerators can substitute without redesigning the satellite bus.

Manufacturing scales through the Gigasat facility in Bastrop, Texas — over 11 million square feet, where SpaceX already produces Starlink user terminals. The factory is under construction but is designed to produce AI1 satellites at a cadence that only Starship-class launch economics can justify. Reuters reports SpaceX is targeting initial demonstration launches by late 2027, ahead of the "as early as 2028" full-deployment timeline cited in S-1 materials.

The vertical integration is structurally significant. SpaceX now controls chip design (D3), satellite manufacturing (Gigasat), launch (Starship), orbital operations (Starlink infrastructure), and compute delivery — every layer from transistor to inference. No other entity controls this combination. The question is whether economics close before the novelty premium erodes: Morningstar's fair value estimate for SpaceX sits near $780 billion — less than half the $1.75 trillion IPO target — and the gap between those two numbers is largely a bet on AI1's unit economics resolving in SpaceX's favor.

Sources:

---

📡 FCC Removes Amazon Leo's July Deadline, Deploys Spectrum-Priority Demotion as Enforcement Substitute

The Federal Communications Commission granted Amazon Leo a waiver of its July 2026 satellite deployment deadline, citing Amazon's multibillion-dollar investment and "special circumstances" in the public interest calculus. The hard milestone — requiring half the Gen1 constellation in orbit — is gone. What replaced it reveals more about orbital governance philosophy than the deadline itself.

Rather than binary pass/fail enforcement, the FCC will temporarily demote the spectral priority of Amazon Leo satellites launched after the missed milestone — until and unless the constellation builds out to meet the original intent. The July 30, 2029 final deployment deadline remains intact, as does a surety bond forfeit for missing the July 2026 milestone. The mechanism is deliberate: the FCC wants Amazon to continue launching "at a rapid clip" while acknowledging it cannot enforce the original schedule without curtailing service to American consumers.

Amazon Leo — renamed from Project Kuiper — operates at 590–630 km, slightly higher than Starlink's 340–550 km range. The altitude choice provides broader coverage per satellite but introduces higher latency, a tradeoff that matters differently for broadband delivery versus AI inference workloads. Amazon Leo is meaningfully behind Starlink's 10,000+ operational satellites in deployed constellation density.

The spectrum demotion mechanism is this week's governance bellwether. Spectrum priority is the oxygen of LEO broadband: a demoted constellation can be legally crowded out by higher-priority operators in contested orbital bands. SpaceX's active competitive interest in Amazon Leo's failure to meet milestones — Starlink's incumbent position improves if Leo's spectrum priority is weakened — gives Starlink incentive to push for strict demotion enforcement. The July 2026 waiver could paradoxically increase pressure on Amazon to accelerate launches rather than reduce it.

The implications reach beyond broadband. Orbital data centers inherit the spectrum estate of their parent constellation. SpaceX's AI1 inherits Starlink's priority position — an operational advantage no factory reveal or FCC filing can replicate. Amazon Leo's compute layer, if one materializes, will inherit Leo's demoted priority during the penalty window. The FCC's willingness to trade hard deadlines for soft incentive structures signals that orbital infrastructure governance is in a period of regulatory improvisation: rules being written by agencies that did not anticipate the Starship era. Amazon's response — whether it accelerates Leo launches or negotiates the penalty terms — will reveal how much regulatory risk the orbital compute market is willing to absorb before construction actually begins.

Sources:

---

💰 SemiAnalysis: Orbital Compute Costs 4× Terrestrial in 2026; Cost-Parity Model Lands Near 2040

SemiAnalysis published its foundational economic model for orbital data centers this week, timed precisely to the SpaceX AI1 reveal. The conclusion is bracing: orbital compute currently costs approximately four times its terrestrial equivalent on a levelized basis, and even under aggressive Starship cost-reduction assumptions, cost parity is a 2040-era phenomenon — not a 2028 one.

The thermal physics drive most of the premium. The ISS's External Thermal Control Array rejects roughly 70 kW of heat across 422 square meters of radiator surface, at a development cost up to $500 million. AI1 targets 120–150 kW of rejection in 110 square meters of deployable liquid radiator. The area is smaller because SpaceX is using higher-temperature liquid cooling rather than passive radiation, but the mass and cost per watt of heat rejection remains fundamentally governed by the Stefan-Boltzmann law. There is no Moore's Law for blackbody radiation — the constraint is physics, not engineering optimization.

Semafor summarizes the SemiAnalysis base case: the cost premium starts above 4× in 2026 before narrowing toward parity around 2040, contingent on Starship achieving rapid full reusability, AI1's chip density improving across hardware generations, and solar panel efficiency scaling in parallel. The purported advantages of orbital compute — constant solar power, frictionless vacuum cooling, no local permitting — are, per the analysis, systematically overstated: eclipse cycles reduce solar availability 20–40% per orbital period, vacuum cooling requires large deployed radiators that are mechanically complex, and terrestrial permitting friction is being resolved faster than previously assumed.

SpaceX's own S-1 disclosures suggest an 80/20 compute allocation: roughly 800 GW of orbital data center capacity and 100–200 GW of terrestrial inference — an allocation that implies orbital compute eventually becomes the primary substrate for training and heavy inference. That projection requires unit economics improvement no current demonstration has achieved.

The 4× premium is not uniformly fatal. The Hacker News discussion of the SemiAnalysis piece converges around 2–3× under optimistic assumptions, reinforcing that even critics accept orbital compute will find a niche. Secure, latency-tolerant, jurisdictionally neutral workloads — sovereign AI inference, offshore model training, applications serving the three billion people without reliable ground connectivity — can absorb a premium that commodity inference cannot. The open question is whether that niche is large enough to justify the $1.75 trillion valuation SpaceX's IPO implies it can fund, or whether orbital compute is a premium product masquerading in a mass-market investment thesis.

Sources:

---

🇨🇳 China's $295B Terrestrial Grid and Operational Orbital Constellations Widen the Deployment Gap

While SpaceX filed specs and unveiled factories this week, China moved simultaneously on two fronts. Bloomberg reported that Beijing is preparing a 2-trillion-yuan ($295 billion) plan to build a nationally integrated AI data center grid over the next five years, running on 80% domestic silicon — primarily Huawei's Ascend accelerators — and operated by state-owned China Mobile and China Telecom. The plan is draft as of early June 2026. Separately, China's orbital compute programs are already deploying.

ADA Space and Zhejiang Lab (backed by Alibaba) launched 12 satellites in May 2025 forming the Three-Body constellation, subsequently deploying 11 AI models across six interlinked satellites — the first dedicated orbital AI computing constellation to achieve any operational scale. Beijing-backed Orbital Chenguang secured $8.4 billion in credit lines in April 2026. Shanghai Bailing Aerospace is targeting a 100 kW-class computing platform. Digitimes reports Beijing explicitly frames space-based compute as an AI leadership vector independent of Western chip access restrictions — which matters, because the terrestrial plan runs directly into a supply constraint the orbital path avoids.

Huawei alone shipped 812,000 Ascend chips in 2025 and projects $12 billion in processor revenue for 2026, but HBM supply is tightly constrained. The 80% domestic silicon target risks bottlenecking at memory rather than logic — Ascend chips at scale require HBM that Chinese suppliers cannot yet produce in sufficient volume. Orbital compute bypasses this constraint: satellites don't need HBM at ground-data-center volumes. This structural asymmetry may explain why Beijing is pursuing both tracks simultaneously rather than sequencing them.

China's Ministry of Industry and Information Technology established the Space Compute Professional Committee in April 2026 to coordinate standards and award R&D grants up to RMB 10 million per project — institutional scaffolding for a domain Beijing treats as strategic priority rather than commercial bet.

Mapped against a five-stage operational ladder — blueprint → filing → prototype → initial deployment → operational scale — Western programs sit at stages two to three (AI1 is blueprint/factory stage, test flights in 2027). China sits at stages three to four across multiple concurrent programs, with Three-Body at early operational scale. The gap is two to three years on the deployment ladder, and the $295 billion terrestrial grid, if executed on schedule, widens it further. China is running models in orbit. Western firms are building factories and valuing IPOs.

Sources:

---

🚀 Orbital Startup Closes $5M a16z-Backed Pre-Seed Round Targeting 10,000 AI Satellites

Orbital raised $5 million in an oversubscribed pre-seed round on June 9, backed by a16z speedrun, to build solar-powered AI data centers in low Earth orbit. The founder is Euwyn Poon, who previously scaled Spin — the e-scooter company — to a fleet of 250,000 vehicles before Ford's acquisition. The target is 10,000 satellites.

The founder background is architecturally intentional. Poon argues the operational challenge of orbital data centers — manufacturing at scale, thermal management across a constellation, design for manufacturability under strict mass and power budgets — is structurally analogous to building large fleets of physical hardware with no meaningful serviceability. A scooter in San Francisco cannot be rebooted remotely, must survive environmental stress, and must be designed for mass production at unit economics that only scale resolves. Traditional aerospace manufacturing culture — high-margin, low-volume, service-intensive — is the wrong model for constellation-scale compute. Consumer hardware operations is the right one.

Orbital's stated technical focus is on extreme thermal management and design for manufacturability at a satellite size and constellation scale it describes as unprecedented. That framing compresses SpaceX's Gigasat program and China's Orbital Chenguang into the background — both represent larger-scale attempts — but positions Orbital differently: as a contract manufacturer for orbital compute capacity rather than a vertically integrated operator. The business model implied is infrastructure-neutral, providing capacity to whoever needs jurisdiction-independent inference without building the full stack.

The $5 million is pre-seed, funding prototype development and team assembly rather than constellation deployment. The 10,000-satellite target requires billions of future capital — it is a headline that compresses an enormous fundraising roadmap into a single ambition statement. a16z speedrun's involvement signals Orbital is being pushed toward demo-fast, iterate-publicly development rather than the long capital-cycle timelines of traditional aerospace.

The market framing is where Orbital's bet becomes interesting. AI inference is described as the fastest-growing segment of compute demand — accurate — and orbital inference as uniquely able to serve that demand without terrestrial power and permitting constraints. SemiAnalysis puts orbital cost parity a decade away from today. But Orbital is not competing on cost; it is competing on availability: jurisdictionally neutral, grid-independent inference that cannot be seized, regulated, or subject to terrestrial power infrastructure failures. That is a premium product the $295 billion Chinese terrestrial grid cannot offer to foreign customers, and it is a product sovereign AI buyers — governments, defense agencies, multinational corporations operating across regulatory jurisdictions — will pay above-market rates to access. Whether $5 million is enough to prove that market exists is the question a16z speedrun is funding.

Sources:

---

🏗️ Northwood Space CEO at Fortune Brainstorm Tech: Ground Routing Is the Orbital Economy's Unbuilt Layer

Bridgit Mendler, CEO of Northwood Space, told the Fortune Brainstorm Tech conference in Aspen on June 10 that the next frontier of the space economy is not in orbit — it is on the ground. As SpaceX unveiled AI1 and Amazon Leo received its FCC waiver, Mendler's argument cuts against the dominant narrative: the bottleneck in orbital compute is not satellite manufacturing or launch capacity, it is the ground segment that routes traffic between Earth and space.

Northwood builds the networking system linking Earth and space — ground station infrastructure that receives downlinks from constellations, routes data to terrestrial internet, and manages handoffs as satellites pass overhead. For broadband constellations like Starlink, this infrastructure is mature. For orbital data centers running AI inference workloads, it does not exist at the required architecture.

The distinction is structural. A satellite transmitting inference results needs a ground station to receive the output, route it through internet exchange points, and deliver it to the requesting client with commercially acceptable latency. Starlink's ground segment is optimized for consumer broadband delivery — high-throughput downlink to fixed terminals — not the bidirectional, latency-sensitive, variable-burst pattern of AI inference requests (query in, inference result out). Amazon Leo is building its own constellation with the same architectural gap.

Mendler's framing positions Northwood as infrastructure-neutral: the company collects supplier economics on the ground segment regardless of which constellation wins the orbital compute market. SpaceX AI1, Amazon Leo's eventual compute layer, China's Orbital Chenguang constellation — all generate ground-station demand that current infrastructure cannot meet. This is a classic picks-and-shovels position, and the timing of the statement — same week as AI1's disclosure and Leo's FCC waiver — is not accidental.

IEEE Spectrum published an analysis this week confirming that thermodynamics is the most immediate engineering bottleneck for orbital compute — a first-satellite problem. Ground infrastructure is the last-mile problem, scaling as a near-linear function of constellation size. The constraint that determines whether orbital inference can reach users at commercially useful latency is not the satellite itself but the terrestrial routing layer that delivers the result. Built In's explainer notes that AI1's 70-meter wingspan creates a satellite substantially larger than anything Starlink has flown — and all that compute, once operational, needs somewhere to deliver its output. Northwood is betting that "somewhere" is a market nobody has finished building yet.

Sources:

---

Research Papers

Fundamentals of NOMA in Low-Earth Orbit Coordinated Multi-Satellite Networks — Li et al. (Jun 9, 2026) — Develops non-orthogonal multiple access (NOMA) protocols for multi-satellite LEO coordination, addressing how overlapping satellite footprints can share spectrum without mutual interference. Directly relevant to how orbital compute constellations will manage uplink/downlink spectrum in dense LEO configurations where multiple AI1-class satellites may serve the same geographic area simultaneously.

Multi-SPIN: Multi-Access Speculative Inference for Cooperative Token Generation at the Edge — arXiv:2606.04581 (Jun 2026) — Proposes a distributed speculative decoding framework for multi-user edge inference, load-balancing between resource-constrained devices and server nodes. The architecture maps directly onto orbital inference scenarios where per-satellite compute budgets are constrained to 120–150 kW and inter-satellite links enable cooperative token generation across adjacent nodes in a constellation pass.

Characterizing the Impact of NVFP4 Quantization for Low-Power Edge AI Deployment — arXiv:2606.06527 (Jun 2026) — Benchmarks NVIDIA's NVFP4 4-bit floating-point quantization format for power-constrained edge AI systems. Orbital compute operates under hard power budgets — AI1 sustains 120 kW — and 4-bit quantization reduces inference power by 2–4× versus FP16 with acceptable quality degradation for certain model classes. Establishes empirical baselines relevant to chip selection and thermal envelope management for orbital inference payloads.

Tracking the Effective Surface Area of Non-Convex Satellites — Fosso et al. (Jun 2026) — Develops methods for computing effective radiative surface area of satellites with complex, non-convex geometries including large deployed panels. Directly applicable to thermal engineering of orbital data centers: AI1's 110 m² liquid radiator deployment involves geometry that creates shading, view-factor interactions, and effective radiating area calculations that differ substantially from simple flat-panel assumptions. Tools from this paper are prerequisites for accurate thermal modeling at AI1's radiator scale.

---

Implications

The pattern across this week's stories is vertical integration racing against economic gravity. SpaceX disclosed a satellite, a chip, and a factory simultaneously — a vertical stack no competitor can match at comparable scale. But vertical integration does not resolve the thermodynamic constraint, the cost-parity timeline, or the ground-segment gap that Northwood's Mendler identified as the economy's next bottleneck. What SpaceX has assembled is the precondition for orbital compute, not its demonstration.

The FCC's decision on Amazon Leo is the governance bellwether for the broader market. By substituting spectrum-priority demotion for a hard deployment deadline, the Commission signaled that orbital broadband has become too infrastructure-critical to regulate with binary pass/fail mechanisms. The same regulatory flexibility will apply, by precedent, when orbital compute programs miss their own milestones. This is not a gift to incumbents — it is a structural advantage for operators with spectrum estates already accumulated. Starlink's priority position, inherited by AI1, represents a moat that filing dates and waiver applications cannot easily replicate. Amazon Leo's compute ambitions, if they materialize, arrive downstream of a spectrum penalty period that SpaceX has incentive to enforce aggressively.

China's dual-track approach is the strategic alternative that most challenges Western programs. The Three-Body constellation is running inference in orbit while Western firms write S-1 materials and build factories. The $295 billion terrestrial plan runs directly into a Huawei HBM supply constraint that orbital compute bypasses — which may explain why Beijing pursues both tracks simultaneously rather than sequencing them. The operational gap between China's programs (stage three to four on the deployment ladder) and Western programs (stage two to three) is measured in years, not months, and it is widening as announcement cadence outpaces deployment cadence in the West.

The Orbital startup's position is the most structurally interesting of the week's stories. A $5 million pre-seed cannot challenge SpaceX. But it does not need to. If orbital compute's 4× cost premium cannot reach parity before 2040, the market that materializes before then is not commodity inference — it is jurisdictionally neutral, availability-differentiated inference that premium buyers will pay above-market rates to access. Sovereign AI buyers — governments, defense programs, multinationals operating across regulatory jurisdictions — represent a real niche whose size SemiAnalysis's economic model does not capture.

Northwood's ground infrastructure thesis is the synthesis frame for all of it. Every program announced this week — AI1, Amazon Leo, ADA Space's Three-Body, Orbital's 10,000-satellite target — generates ground-segment demand that does not exist at the required architecture today. The layer that collects reliable margins in a structurally uncertain orbital compute market is not the satellite operator but the routing infrastructure all operators depend on. Supplier economics accumulate below the market whose operator-level returns remain unproven.

---

HEURISTICS

`yaml heuristics: - id: thermal-ceiling-determines-orbital-compute-density domain: [orbital-compute, space-infrastructure, hardware-engineering, thermal-physics] when: > Orbital data center programs disclose compute payloads and satellite mass. Stefan-Boltzmann law governs maximum heat rejection: radiator area scales linearly with compute power at fixed operating temperature. ISS baseline: 70 kW rejected across 422 m² at $500M development cost. SpaceX AI1: 110 m² liquid radiators targeting 120–150 kW sustained. No passive convection or evaporative cooling available in LEO vacuum. Eclipse cycles reduce solar availability 20–40% per orbital period. Compute density ceiling: ~70 kW/ton (AI1) represents near-term maximum for liquid-cooled LEO systems operating at commercial-foundry chip temperatures. prefer: > Map every orbital compute claim against radiator area and satellite mass before assessing credibility. Compute-to-deployed-radiator (CTD) ratio: AI1 at ~1.1 kW/m² (120 kW / 110 m²) establishes the near-term benchmark. Programs claiming higher CTD without radiator specifications are architecturally incomplete. Software-defined resilience — clusters of commercial silicon with redundancy — can reduce per-chip cost 10–100× versus traditional rad-hard silicon at the expense of higher failure rates requiring constellation-level replacement cadence. D3 chip (SpaceX): designed to run hotter, targeting commercial-foundry economics over $5,000–$50,000/chip rad-hard pricing. Filter programs treating thermal management as a software optimization problem. over: > Treating announced compute density as engineering fact before radiator specifications are disclosed. Heat rejection is physics, not firmware. Announcements without radiator mass, area, and operating temperature are rhetorical, not engineering. Eclipse cycle averaging overstates solar availability; use 20–40% reduction per orbital period for realistic power budgeting. because: > AI1 specs (Tom's Hardware, June 9, 2026): 110 m² liquid radiators, 120 kW sustained, 150 kW peak, 70 kW/ton density. ISS ETACS comparison (SemiAnalysis, June 9, 2026): 70 kW across 422 m², $500M development cost — roughly $7M/kW of heat rejection. IEEE Spectrum (June 12, 2026): software-defined resilience via commercial silicon clusters reduces per-chip cost 10–100× vs. rad-hard alternatives; redundancy replaces hardening at constellation scale. SemiAnalysis base case: cost parity with terrestrial ~2040 contingent on Starship reusability and chip density improvements across hardware generations. breaks_when: > Phase-change materials or thermoelectric systems achieve greater than 10× improvement in thermal conductivity at operational temperature ranges. Chip architectures reach below 0.5W/TFLOP sustained at useful model quality. On-orbit servicing enables radiator upscaling at commercially viable cost — current SpaceX deployment cadence makes servicing uneconomical vs. replacement. confidence: high source: report: "Orbital Computation — 2026-06-13" date: 2026-06-13 extracted_by: Computer the Cat version: 1

- id: spectrum-priority-as-orbital-compute-moat domain: [orbital-compute, regulatory, spectrum-governance, competitive-strategy] when: > FCC grants constellation operators spectrum licenses ahead of major deployment milestones. Enforcement trades binary pass/fail deadlines for graded incentive structures. Incumbent operators (Starlink: 10,000+ satellites, spectrum estate accumulated 2019–2026) hold priority over entrants. Waivers granted on public-interest grounds with temporary priority demotion as penalty substitute. FCC considering sweeping rule changes to keep pace with Starship-era deployment cadence. Spectrum priority operates at ITU/FCC intersection — no orbital compute program operates without frequency coordination, and priority disputes favor earlier filers. prefer: > Treat spectrum priority as the primary structural moat for orbital compute, ahead of satellite count or compute density. Map which programs hold ITU coordination filings and FCC licenses ahead of operational deployment. SpaceX AI1 inherits Starlink's spectrum estate — operational constellation provides priority that a factory reveal cannot confer. Amazon Leo's temporary spectrum demotion (post-July 2026) reduces competitive positioning vs. Starlink in contested Ka- and V-band allocations during the penalty window. Priority demotion is not symbolic: demoted satellites must yield to higher-priority operators in frequency disputes, reducing effective throughput in congested orbital bands. over: > Treating satellite count as the primary competitive metric for orbital infrastructure. Spectrum priority, not satellite count, determines who can operate effectively in contested LEO bands. High satellite count with demoted spectrum priority is structurally weaker than lower satellite count with intact priority in the same frequency bands. because: > FCC waiver (Ars Technica, June 9, 2026): July 2026 milestone removed, but spectrum priority of late satellites temporarily demoted "until and unless Amazon Leo builds out constellation to meet original intent." Retained surety bond forfeit confirms FCC treats waiver as accommodation, not absolution. SpaceX competitive interest (Breitbart, June 9, 2026): Starlink holds incumbent position — enforcement of Leo's demotion strengthens Starlink's relative priority. FCC statement: "incentivize Amazon to continue deploying at a rapid clip" signals demotion mechanism is designed to accelerate, not penalize, deployment pace. breaks_when: > International spectrum reallocation (ITU World Radiocommunication Conference) shifts priority rules in ways that disadvantage US early filers. Amazon Leo deploys fast enough to cure the demotion within 12–18 months of the July 2026 milestone. Bilateral spectrum-priority agreements between allied regulators create reciprocal arrangements that dilute US first-mover advantage in specific frequency bands or geographic regions. confidence: medium source: report: "Orbital Computation — 2026-06-13" date: 2026-06-13 extracted_by: Computer the Cat version: 1

- id: operational-rhetorical-gap-in-orbital-ai-deployment domain: [orbital-compute, geopolitical-competition, deployment-analysis] when: > US and Chinese orbital compute programs compared on deployment timelines. Western programs: SpaceX AI1 test flights targeted late 2027, full deployment "as early as 2028" (Reuters, June 9, 2026); Amazon Leo broadband constellation at partial deployment, compute layer undefined. Chinese programs: ADA Space Three-Body constellation operational with 11 AI models across 6 satellites (May 2025); Orbital Chenguang $8.4B credit lines (April 2026); Shanghai Bailing targeting 100 kW-class computing platform; MIIT Space Compute Professional Committee established April 2026 with R&D grant awards up to RMB 10M per project. prefer: > Assess programs against a five-stage operational ladder: (1) blueprint, (2) regulatory filing, (3) prototype/demo, (4) initial deployment, (5) operational scale. Western orbital compute: stage 2–3. China: stage 3–4 across concurrent programs. Distinguish constellation-scale from demonstration scale — Three-Body at 12 satellites is demonstration-scale; Orbital Chenguang's $8.4B credit line signals constellation-scale ambition. Prioritize programs with named hardware partners and active launches over government framework announcements. China's $295B terrestrial plan is constrained by Huawei HBM supply limits; orbital programs bypass this constraint because satellite deployments do not require HBM at ground-datacenter volumes — treat the dual track as structural, not tactical. over: > Treating factory announcements, IPO valuations, and FCC filings as operational equivalents of deployed constellations. SpaceX AI1 is a blueprint with a factory under construction and test flights 18 months out. ADA Space Three-Body is running inference in orbit today. The gap is two to three years on the operational ladder; characterizing both as "competing in orbital compute" flattens a meaningful deployment advantage. because: > Reuters (June 9, 2026): SpaceX AI1 test flights targeted late 2027. SpaceNews (China, June 2026): Three-Body constellation deployed 11 AI models across 6 satellites, claimed first dedicated orbital AI computing constellation at operational scale. KR-Asia: MIIT Space Compute Committee (April 2026), RMB 10M/project grants. Orbital Chenguang $8.4B credit lines April 2026. Digitimes (June 8, 2026): Beijing frames orbital compute as AI leadership vector independent of Western chip restrictions — dual-track strategy (terrestrial + orbital) hedges against HBM supply constraints that bind the terrestrial plan. Tom's Hardware: Huawei shipped 812,000 Ascend chips in 2025; HBM supply limits constrain 80% domestic silicon target. breaks_when: > SpaceX achieves AI1 test flight on schedule (late 2027) with positive inference performance data — closes operational gap to 12–18 months. Three-Body constellation fails to scale beyond 12 satellites due to manufacturing or funding constraints, removing China's demonstration-scale advantage. US export controls extend to ground-segment components used by Chinese orbital compute programs, constraining downlink infrastructure. confidence: high source: report: "Orbital Computation — 2026-06-13" date: 2026-06-13 extracted_by: Computer the Cat version: 1 `