🛰️ Orbital Computation — 2026-06-14

Table of Contents

• 🛰️ SpaceX AI1: 70-Meter Wingspan, 120 kW Sustained Compute, Interchangeable Chip Payload
• 📈 SPCX Closes +19% on Largest IPO in History, Orbital Compute Becomes Public Market Thesis
• 💰 Google's $920M/Month Deal Confirms SpaceX's Revenue Precedes Any Orbit
• 🔗 xAI Merger Completes the Stack: Rockets → Connectivity → Inference → Intelligence
• 🇨🇳 China's Spacesail at 200 Satellites as SpaceX's $75B IPO Reshapes Beijing's Ambitions
• 🌡️ IEEE Spectrum: Thermodynamics, Not Chips, Will Decide the Orbital Data Center Race

🛰️ SpaceX AI1: 70-Meter Wingspan, 120 kW Sustained Compute, Interchangeable Chip Payload

SpaceX unveiled the AI1 orbital data center satellite on June 8, and the hardware specs reframe what "AI satellite" means at production scale. The spacecraft is 20 meters tall with a 70-meter deployed wingspan — wider than a Boeing 747-8 — operating at 600 km altitude. Average compute payload is 120 kW; peak is 150 kW; power density is 70 kW per launched ton. That peak is calibrated against a single NVIDIA GB300 rack terrestrial draw of ~140 kW: one AI1 equals one rack, launched.

The thermal architecture is the engineering story. AI1 carries 110 m² of deployable liquid radiators — vertically oriented, double-sided, with redundant pumping loops and integrated micrometeoroid shielding. The International Space Station's entire External Thermal Control Active System rejects approximately 70 kW. AI1 must reject more than twice that from a single spacecraft. The compute payload is sandwiched between the radiator panels in the spacecraft's thermal core, a geometry that minimizes conduction path length while maximizing rejection area.

Musk's manufacturing claim — that the AI1 is simpler than a Starlink satellite — is a production argument, not a capability one. By stripping antenna complexity and reusing the Starlink V3 hardware platform already in Hawthorne production, SpaceX routes around the new-platform tooling bottleneck that constrains every other orbital compute entrant. Inter-satellite laser links are retained; everything else is reduced to a solar collection array, a compute bay, and a radiator.

Manufacturing at scale requires a new facility. SpaceX simultaneously disclosed Gigasat, an 11-million-square-foot factory complex in Bastrop, Texas, with an onsite semiconductor packaging line targeting production by end of 2026. The stated goal is 1 GW/year of orbital compute capacity by late 2027 — which at 150 kW per AI1 implies approximately 6,700 satellites annually out of Bastrop. Solar manufacturing broke ground in April 2026 and is accelerating through June.

The chip payload is designed to be interchangeable in orbit — the compute module can be swapped as GPU generations advance during the satellite's five-year mission. This servicing architecture addresses the obsolescence problem that every long-duration orbital compute program faces, and makes Starlink's ongoing orbital operations infrastructure directly valuable to the AI1 program.

What the AI1 reveal is not: a functional system. No orbital inference workloads have run. The Gigasat factory is under construction. The FCC application for a million-satellite constellation remains under review. The gap between filed hardware specifications and operational constellation is the widest in the industry — but the specifications are now public, load-bearing for the IPO narrative that followed four days later, and specific enough to evaluate technically against the thermodynamics ceiling.

Sources:

• TechSpot — AI1 wingspan reveal
• Tom's Hardware — AI1 compute specs and radiators
• Tom's Hardware — Gigasat factory

📈 SPCX Closes +19% on Largest IPO in History, Orbital Compute Becomes Public Market Thesis

$135 per share. $75 billion raised. $1.77 trillion valuation. SpaceX priced Thursday night and opened at $150 on Friday, closing at $160.95 — the largest initial public offering in history, exceeding Saudi Aramco's 2019 record. The company offered 555.6 million shares on Nasdaq under the ticker SPCX, no roadshow, no price range — a fixed $135 and a 17-minute pitch video uploaded to a custom website. The +19% first-day pop reflects retail appetite for a name with no public comparable.

The IPO narrative rests on three layers. The first two — reusable launch via Starship and broadband connectivity via Starlink — are operational and generating revenue. The third is orbital AI data centers, an infrastructure category with zero deployed compute nodes and a first-demonstration timeline of late 2027. SpaceX's prospectus makes the physics case explicitly: "The sun contains approximately 99.8% of the solar system's energy and offers what we believe is the only truly scalable solution to the challenge of accelerating demand for compute relative to terrestrial energy constraints." The logical path forward, the filing continues, is "to move power-intensive AI workloads into orbit."

The valuation mathematics require the orbital layer to become real. SpaceX reported a net loss of $4.3 billion in Q1 2026. Starlink subscribers and Starship commercial launches generate revenue, but not at $1.77 trillion scale. The Google compute deal (covered separately) is the only orbital-adjacent contract that materially moves the numbers — and it is for terrestrial compute from Memphis data centers, not from orbit.

Scientific American's analysis on IPO day identifies Starship routine operations as the structural prerequisite: every AI1 satellite reaches orbit via Starship. Until Starship achieves reliable, high-cadence flights, the 1 GW/year orbital compute target is capped not by factory throughput but by launch availability. Demonstration launches for the AI satellites could begin by late 2027, with deployment starting thereafter — a timeline that leaves the IPO orbital narrative entirely in the prospectus.

What this listing does structurally: it creates the first public market valuation anchor for orbital compute as a category. The Globe and Mail notes the orbital compute thesis is explicitly embedded in the IPO pricing — which means every future orbital compute entrant, from Axiom to Starcloud to China's ADA Space, now has a public reference price. No other company has priced this thesis. The $1.77 trillion number becomes the benchmark, and it embeds a substantial premium for infrastructure that does not yet exist in orbit.

Sources:

• NPR — IPO first-day performance
• Scientific American — IPO valuation analysis
• Business Insider — orbital compute IPO thesis

💰 Google's $920M/Month Deal Confirms SpaceX's Revenue Precedes Any Orbit

Google will pay SpaceX $920 million per month from October 2026 through June 2029 for access to "approximately 110,000 NVIDIA GPUs, CPUs, memory, and other related components." Disclosed in SpaceX's IPO filing on June 5, the deal totals roughly $33 billion over 32 months — more than SpaceX raised in the IPO itself. Tom's Hardware reports that this single contract's annualized revenue will exceed SpaceX's combined 2025 proceeds from Starlink, launch services, and AI.

Nothing about that compute is orbital. The 110,000 GPUs are housed in SpaceX/xAI's Memphis data center complex — terrestrial infrastructure acquired via the February 2026 xAI merger. Google is purchasing compute capacity, not satellite access. Gemini model inference, enterprise workloads, or training runs are the plausible use cases; SpaceX's IPO filing does not specify workload type. Google's orbital access is exactly zero.

The strategic architecture reveals itself: SpaceX structured its compute business to generate terrestrial revenue before orbital infrastructure exists. The Memphis GPU farm is the financial bridge — Google's payments fund the Gigasat factory, the AI1 program, and the FCC application process simultaneously. The orbital compute thesis is collateralized against a terrestrial GPU rental agreement, which makes the enterprise substantially less speculative than the AI1 hardware reveal implies.

A structural asymmetry is embedded in the deal terms: Google receives compute from SpaceX's existing hardware, but SpaceX controls the roadmap. If AI1 satellites begin operating at scale by 2028-2029, SpaceX can offer Google orbital inference as contract renewal terms. Google locked in compute at approximately $300/GPU/month — below current spot market rates for H100/H200 hardware — and in doing so became financially entangled with the success of the orbital compute thesis. The $300/GPU/month is the price SpaceX accepted to build the runway; the orbital contract price it extracts in 2029 is the margin it deferred.

From a competitive standpoint, the deal confirms Google's acceptance that SpaceX is now a tier-1 AI infrastructure vendor — not a launch provider or broadband company. Microsoft, AWS, and Google have each invested in orbital connectivity through ground station services and broadband partnerships, but none has committed $33 billion to a compute relationship with a single space company. Google's commitment changes how Amazon must evaluate its own Kuiper architecture: Kuiper's compute layer remains unannounced, and the SpaceX/Google deal sets the reference price for what orbital-adjacent AI compute is worth to a hyperscaler over a 32-month period.

Sources:

• TechCrunch — deal disclosure
• Business Insider — Google compute capacity detail
• TheStreet — pre-IPO timing analysis

🔗 xAI Merger Completes SpaceX's Stack: Rockets → Connectivity → Inference → Intelligence

The orbital compute thesis only holds as a full vertical stack, and the February 2026 all-stock acquisition of xAI completed it. The transaction folded a frontier AI company directly into SpaceX — linking satellite communications, terrestrial data center operations, and planned orbital infrastructure under unified governance. The Memphis data centers funding the Google compute deal are xAI assets. The Grok models that would eventually run on AI1 satellites are xAI products. xAI's parent platform, X, was absorbed via a prior share swap, supplying training data and application distribution.

The resulting stack: Starship provides the launch cadence required to build orbital infrastructure at scale. Starlink V3 provides the inter-satellite laser mesh that AI1 inherits. AI1 provides the compute nodes. xAI's Grok runs as the intelligence layer. Every layer is controlled by a single entity — the rockets, the satellites, the power source, the cooling, the models. "SpaceX/xAI will effectively become the new AWS of space — except they control the rockets, the satellites, the power source, and the cooling. That vertical integration is the real moat."

Forbes, writing the day after the IPO, frames this as SpaceX publicly declaring the AI infrastructure war: "xAI, which SpaceX acquired in an all-stock deal in February, brings the intelligence layer in-house. The Memphis data centers, and that Google contract, are terrestrial, real and generating revenue today." The orbital layer is the roadmap extension of a business generating tens of billions in terrestrial compute revenue — the sequence is: demonstrate the model, prove the demand, build the orbital infrastructure to serve it.

What the hyperscalers cannot replicate: AWS cannot build its own rocket. Microsoft cannot launch its own constellation. Amazon Leo had 331 production satellites in orbit as of May 2026, making it the closest orbital competitor in constellation count, but none carry GPU payloads and no compute product has been announced. Forbes's pre-IPO analysis identifies the gap: "The February 2026 merger with xAI folded a frontier AI company directly into that stack... no competitor has matched this integration across launch, connectivity, edge, and model layers simultaneously."

The key structural constraint remains Starship cadence. The orbital compute thesis requires thousands of AI1 launches per year. Scientific American identifies reliable, high-frequency Starship flights as the structural prerequisite — until that threshold is reached, the stack can be described but not delivered. The xAI merger assembled the intelligence and compute layers; Starship reliability is the execution layer that unlocks the orbital portion.

Sources:

• Reuters — xAI acquisition
• Forbes — post-IPO stack analysis
• Quasa.io — vertical integration moat analysis

🇨🇳 China's Spacesail at 200 Satellites as SpaceX's $75B IPO Reshapes Beijing's Ambitions

SpaceX's largest-ever IPO has landed squarely in China's strategic calculus. Reuters analysis published June 12 identifies the $75 billion raise as "poised to supercharge Chinese space startups racing to fund the same technologies" — reusable rockets and giant satellite constellations. The precedent: if Musk prices orbital infrastructure at $1.77 trillion, Chinese firms can anchor their own IPO narratives to the same category. The SpaceX playbook is now a financing template.

The operational gap, however, is stark. Guowang and Qianfan (Spacesail) have a combined few hundred satellites in orbit against Starlink's roughly 10,400. Spacesail reached 200 satellites by June 2026 — enough to support maritime vessel tracking as its first commercial application, with broader commercial service targeting end of 2026. A June 1 launch used a new Chinese reusable rocket, demonstrating manufacturing capability but not constellation-scale cadence. Spacesail has raised over $1 billion, is seeking fresh capital, and targets 15,000 satellites by 2030.

The compute layer tells a different story. China's Three-Body Computing Constellation, developed by CASIC (China Aerospace Science and Industry Corporation), completed nine months of orbital testing in February 2026 — operational before SpaceX has launched a single AI1. The program targets 2,800 compute satellites delivering 1,000 peta operations per second. China formally launched a national feasibility study for a space-based intelligent computing constellation in April 2026, signaling state-level investment commitment beyond the commercial Three-Body program.

The operational-rhetorical inversion is analytically precise: China has compute satellites running in orbit with limited scale, accumulating in-orbit operational data. Western firms have compelling orbital compute narratives backed by terrestrial revenue. Neither side has commercially operational orbital inference at significant scale. But China's Zhejiang Lab project is running AI-driven earth observation processing in-situ — reducing latency from hours (ground downlink) to minutes — instrumenting actual on-orbit AI workloads that will inform the next hardware generation.

The competitive architectures diverge at the governance layer. China's approach is state-backed, integrating national defense and commercial actors across a unified program with no IPO timeline pressure and no quarterly loss tolerance. SpaceX's approach is vertically integrated private capital, with the Google contract as revenue bridge and a public market valuation anchoring the orbital narrative. The SpaceX IPO creates a $75 billion capital base to fund the build; it also creates a $1.77 trillion valuation to defend. Which model reaches operational-scale orbital AI compute first — state-directed or private capital-directed — is the decade-level question the IPO made public.

Sources:

• Reuters — China space IPO analysis
• Rest of World — Spacesail operational status
• SatNews — Three-Body testing complete
• Xinhua — national feasibility study

🌡️ IEEE Spectrum: Thermodynamics, Not Chips, Will Decide the Orbital Data Center Race

IEEE Spectrum published a physics-first analysis on June 12 that reframes the orbital compute competition: the binding constraint is not GPU performance, launch cost, or radiation hardening. It is the Stefan-Boltzmann law. Radiative cooling — the only heat rejection mechanism available in vacuum — scales with the fourth power of temperature and linearly with radiator area. This single relationship determines the maximum compute density any orbital data center can sustain at operational equilibrium.

The counterintuitive finding: optimizing for larger radiators is the inferior design strategy. Raising the internal operating temperature by 100 Kelvin — from 300K to 400K — reduces required radiator surface area by a factor of 3.16, a far stronger effect than adding panel area. The practical implication: orbital data centers should run hotter than terrestrial counterparts, exploiting the T⁴ relationship, using vacuum water vapor caloric pipes to distribute heat efficiently at elevated temperatures. This is the same solid-state thermal management approach now appearing in high-end server hardware — but in orbit, the physics make temperature optimization mandatory rather than optional.

SpaceX's AI1 deploys 110 m² of liquid radiators at peak operation, but the mass budget is larger than operational specs suggest. Thermal coating degrades in the orbital UV and charged-particle environment, forcing designers to launch 40% more radiator mass than the end-of-life operational requirement. This is not a failure mode — it's a calculated constant that every orbital compute program must absorb as a launch mass tax. The satellite must be designed for degraded-end-of-life performance, meaning the launch-day radiator area is deliberately oversized.

SemiEngineering quantifies the hardware baseline: 100 NVIDIA GPUs in orbit require approximately 33 square meters of solar panels and 16 square meters of radiation panels. SpaceX's AI1 at 150 kW peak — equivalent to a GB300 rack — is consistent with this geometry. The 70-meter wingspan is the solar collection apparatus; the 110 m² radiators are the heat rejection apparatus. The satellite's launched mass breakdown is dominated by thermal and power systems, not compute silicon.

The economic viability threshold anchors everything. IEEE Spectrum's back-of-envelope model uses $44/kg for Starship (optimistic) and $0.20/kWh for terrestrial energy: at these parameters, orbital compute undercuts terrestrial data centers on energy cost per FLOP. At actual current Starship pricing, the calculation does not close. The margin between "orbital compute wins on economics" and "orbital compute doesn't close" is entirely determined by Starship's realized cost-per-kilogram. The thermodynamics analysis published on IPO day is the necessary corrective to IPO-day enthusiasm: every entrant — SpaceX, Axiom, Starcloud, China's Three-Body — faces the same radiative physics ceiling. The winner is whichever program solves thermal architecture at launched mass, at the launch cadence required to beat the terrestrial energy crisis timeline.

Sources:

• IEEE Spectrum — thermodynamics analysis (June 12)
• SemiEngineering — GPU thermal baseline
• Arthur D. Little — orbital data center technical review

Research Papers

• Reduced-Mass Orbital AI Inference via Integrated Solar, Compute, and Radiator Panels — arXiv preprint (April 9, 2026) — Proposes a distributed compute architecture for sun-synchronous orbital satellites achieving >100 kW compute per launched metric ton by integrating solar collection, compute panels, and radiators into a single structural unit. Demonstrates that 512-panel subarrays can run LLM inference at 553 tokens/sec/session across 256 simultaneous sessions; 16,000-panel arrays fit in a single Starship hold, making the architecture directly relevant to AI1-scale deployments.

• SpaceMoE: Towards Orbital General Intelligence with Distributed Mixture-of-Experts Inference — arXiv preprint (May 2026) — Presents a MoE architecture optimized for distributed inference across LEO satellite constellations, routing expert sub-models across nodes based on orbital geometry and inter-satellite link availability. Cites communication-efficient collaborative LLM inference over LEO networks (arXiv:2604.04654) as a foundational dependency, establishing the research cluster now forming around constellation-scale AI inference.

• Glass Box at Orbit: A Constitutional AI Verification Framework for Trustworthy Autonomous CubeSat Intelligence — arXiv preprint (June 2026) — Addresses governance and safety verification for autonomous onboard AI in LEO — the less-covered but structurally important question of what oversight mechanisms apply when inference runs in orbit without ground-loop latency. Validates that commercial edge hardware (Google Coral TPU, NVIDIA Jetson) supports meaningful inference on satellite power budgets when thermal and radiation mitigation is applied.

• Statistical Analysis for Energy-Efficient Satellite Edge Computing with Latency Guarantees — arXiv preprint (May 11, 2026) — Models energy-optimal scheduling for satellite edge inference under orbital geometry constraints, where processing windows are determined by coverage arcs and energy budgets are tightly bounded. Notes NVIDIA's GTC 2026 announcement of Jetson hardware for space qualification as the embedded GPU inflection point; derives latency-energy tradeoff curves applicable to AI1-class compute densities.

Implications

The week of June 8–14, 2026 is the moment orbital compute became a public market thesis with a $1.77 trillion price tag. Three structural shifts converged: SpaceX disclosed what an orbital compute node actually looks like (AI1), revealed a $33 billion terrestrial compute revenue stream that funds the orbital build (Google deal), and listed both under a single equity ticker (SPCX). Each was individually significant; together, they constitute a category legitimization event with no historical comparable in the infrastructure sector.

The vertical integration story is the dominant synthesis. SpaceX/xAI now controls every layer of the AI infrastructure stack from photon collection to model output. No cloud hyperscaler — AWS, Azure, Google Cloud — controls its own launch vehicle or satellite constellation. NVIDIA can sell GPUs to any orbital compute entrant; only SpaceX can launch them at Starship economics, route inter-satellite traffic through an existing laser mesh, and run first-party AI models on the compute nodes. The moat is architectural, not technical, and it deepens with each layer SpaceX controls.

The thermodynamics analysis published by IEEE Spectrum on IPO day is the necessary corrective. The Stefan-Boltzmann law is indifferent to market capitalization. Radiative cooling is the binding constraint, and the margin between "orbital compute closes economically" and "orbital compute doesn't close" is entirely determined by Starship's actual cost-per-kilogram and the thermal architecture choices embedded in AI1's 40% radiator mass buffer. Both are unknowns at scale. The IPO priced the narrative; the physics will price the outcome.

China's operational position is the counterpoint the market appears to be discounting. Spacesail's satellite count gap (200 vs. 10,400) dominates the comparison, but China's Three-Body Computing Constellation completed orbital testing in February 2026 — while SpaceX was filing FCC paperwork and building a factory. State-backed programs are not subject to IPO valuation pressure, are not required to generate quarterly returns, and benefit from unified command-and-control that can prioritize compute orbit allocation at any moment. The Three-Body program targets 2,800 compute satellites; China's April feasibility study signals an even larger national program behind it. The operational-rhetorical gap has been inverted: China is running orbital AI inference at small scale while SpaceX is narrating future orbital AI infrastructure at IPO scale.

The near-term governance bellwether: the FCC's decision on SpaceX's million-satellite orbital data center application. That ruling determines how many AI1 frequency slots are licensed; slot count caps constellation scale; constellation scale determines whether the 1 GW/year target is achievable in the 2027-2028 window the Google contract implies. The decision will establish the FCC's posture on commercial orbital compute spectrum allocation — a governance template that will govern every subsequent entrant, domestic and foreign. Amazon's Kuiper architecture, Axiom's compute node program, and China-linked entrants seeking US spectrum access all wait on the same ruling.

The longer structural question: who captures supplier economics in this stack? NVIDIA supplies GPUs to SpaceX at terrestrial prices. Starship provides launch at cost-plus margins. Google pays for compute at $300/GPU/month. The supplier-to-operator ratio currently favors equipment vendors — they capture hardware margin on every unit while orbital operators absorb the performance, reliability, and launch execution risk. SpaceX's vertical integration strategy is explicitly designed to shift that equation over time: by controlling launch, satellite, and model, it becomes its own supplier at each layer, compressing the margin it pays outward. Whether that compression reaches the orbit layer before the terrestrial energy crisis makes the orbital cost calculation compelling is the decade-scale bet that $1.77 trillion just priced.

---

Heuristics

`yaml heuristics: - id: thermal-constraint-determines-orbital-compute-winners domain: [orbital-compute, thermodynamics, satellite-engineering] when: > Orbital data center programs publish compute density claims. Radiative cooling scales as T⁴ × area — the only heat rejection mechanism available in vacuum. Every entrant faces the same physics ceiling: SpaceX AI1 at 150 kW peak requires 110 m² liquid radiators, exceeding ISS ETACS total rejection capacity (~70 kW). Thermal coating degrades in UV/charged-particle environment, requiring 40% mass buffer over operational spec. Launch cost analysis uses $44/kg Starship (optimistic); actual pricing does not yet close the energy-cost comparison with terrestrial at $0.20/kWh. SemiEngineering baseline: 100 Nvidia GPUs require 33 m² solar + 16 m² radiation panels — AI1 specs at 150 kW are consistent with this geometry. prefer: > Map each orbital compute program against thermal-compute density (CTD) envelope: kW compute / kg launched mass / m² radiator area. SpaceX AI1 claims 70 kW/ton. Evaluate claims by three criteria: (1) Does radiator area scale appropriately for peak compute load? (2) Is the mass budget accounting for the 40% degradation buffer that thermal coating aging requires? (3) Does launch economics close at stated cost/kg under Stefan-Boltzmann constraints? Programs that optimize radiator operating temperature (T⁴ effect) over radiator panel area are technically superior. Vapor caloric pipe integration at >350K operating temp is the leading indicator of mature thermal design. Programs that only discuss chip specs without disclosing radiator mass fraction are incomplete. over: > GPU generation and FLOP count as primary comparison metrics between orbital compute entrants. Radiation-hardening claims without accompanying thermal architecture disclosure. Compute density announcements without corresponding radiator mass fraction. "Simpler than Starlink" manufacturing framing as evidence of thermal adequacy. Constellation satellite count as a proxy for orbital compute capability when no GPU payload is confirmed. because: > IEEE Spectrum (2026-06-12): raising operating temp 100K reduces required radiator area by 3.16x — the T⁴ relationship makes temperature optimization far more powerful than area expansion. SpaceX AI1: 110 m² radiators, 150 kW peak = ISS ETACS rejection capacity ×2.1 from a single spacecraft. Thermal coating degradation forces 40% mass overspec at launch, directly increasing cost-per-kW-delivered. arXiv:2604.07760: >100 kW compute/ton launched mass is achievable with integrated solar-compute-radiator panel architecture, validating SpaceX AI1's 70 kW/ton claim as technically plausible but not definitively demonstrated at scale. Every orbital compute program — SpaceX, Axiom, Starcloud, China Three-Body — is bounded by the same Stefan-Boltzmann constraint. Thermal architecture is the differentiating variable across entrants, not chip generation. breaks_when: > Active cooling via consumable (liquid ammonia, water ice) becomes launch-cost- viable at Starship economics, allowing compute densities above passive radiative limits. Phase-change material thermal buffers enable burst compute exceeding steady-state radiator capacity, decoupling peak from average power envelopes. Novel in-orbit radiator deployment mechanisms reduce mass fraction significantly below current 40% degradation buffer requirement. On-orbit assembly allows radiator expansion post-launch, removing mass-budget constraint from launch economics. confidence: high source: report: "Orbital Computation — 2026-06-14" date: 2026-06-14 extracted_by: Computer the Cat version: 1

- id: vertical-integration-moat-operational-vs-rhetorical-ladder domain: [orbital-compute, competitive-intelligence, infrastructure-strategy] when: > Multiple entrants claim orbital AI compute roadmaps requiring evaluation. Comparative analysis must separate operational position (deployed hardware, revenue-generating contracts, in-orbit workloads currently running) from rhetorical position (FCC filings, factory announcements, IPO narratives, press releases about future capability). SpaceX/xAI: operational terrestrial compute ($920M/month Google contract, Memphis 110,000 GPU data centers), orbital at demo stage (late 2027 target). China Three-Body: operational orbital compute (12 satellites, in-orbit AI workloads confirmed Feb 2026), no equivalent terrestrial revenue bridge. Amazon Leo/Kuiper: 331 production satellites, zero disclosed compute payload, broadband-only product. prefer: > Build a five-stage operational ladder and map each competitor to their rung: (1) terrestrial revenue from compute, (2) FCC/spectrum authorization for orbital compute frequencies, (3) factory/manufacturing capacity operational, (4) demo orbital compute launches confirmed, (5) commercial orbital inference at scale. SpaceX: rungs 1, 2 (pending), 3 (under construction), targeting 4 (late 2027) and 5 (2028+). China Three-Body: rung 4 complete, rung 5 partial at low scale. Amazon Kuiper: rung 2 partial (broadband only), rung 3 functional, rungs 1/4/5 undisclosed for compute. Weight stack layer control by: launch vehicle (SpaceX: Starship; competitors: none owned), constellation (SpaceX: 10,400 Starlink; Kuiper: 331), AI models (SpaceX: Grok/xAI; others: third-party). Layer control depth = moat depth. over: > Total satellite constellation count as primary competitive metric when payloads are different. FCC filing submission as proof of operational capability. IPO valuation as evidence that orbital compute economics are currently proven. "Announced" factory capacity as equivalent to demonstrated manufacturing throughput. Google-SpaceX compute deal as evidence of orbital revenue — the contract is for terrestrial Memphis data centers, not orbital nodes. because: > SpaceX controls rockets + constellation + terrestrial compute + AI models (xAI, Feb 2026 merger) — no hyperscaler replicates this stack. Kuiper: broadband constellation, no owned launch vehicle, no AI layer disclosed. Google: $920M/month buyer of SpaceX terrestrial compute, not an orbital compute operator. China Three-Body: state-directed, non-IPO-timeline-constrained, operational orbital inference before SpaceX has demonstrated AI1. SpaceX Q1 2026 net loss: $4.3B — orbital compute profitability requires Starship economics closing at scale, undemonstrated. The operational-rhetorical gap for SpaceX is currently the widest in the industry: the strongest IPO narrative combined with zero operational orbital compute nodes. breaks_when: > Amazon discloses GPU payload for Kuiper batch, converting constellation count into a comparable orbital compute metric. Azure Orbital deploys compute nodes into LEO directly rather than providing only ground station cloud access. China Three-Body scales from 12 to hundreds of compute satellites within planned 2026 Q4 timeline, closing operational gap faster than SpaceX closes rhetorical gap. Starship achieves ≥12 flights/year with ≤2 anomalies for two consecutive quarters — at that point launch economics may close, and the full SpaceX stack becomes the first complete orbital AI infrastructure system. confidence: high source: report: "Orbital Computation — 2026-06-14" date: 2026-06-14 extracted_by: Computer the Cat version: 1

- id: ipo-anchors-orbital-compute-category-valuation domain: [orbital-compute, capital-markets, competitive-dynamics] when: > Private orbital compute entrants (Axiom, Starcloud, Orbital Inference) and Chinese competitors (Spacesail, ADA Space) seek capital post-SPCX IPO. SPCX priced $135, closed +19% at $160.95 on June 12, 2026 — $1.77T market cap, largest IPO in history, $75B raised. Annual Google contract revenue ($11B) exceeds combined 2025 Starlink + launch + AI proceeds but is terrestrial, not orbital. Q1 2026 net loss: $4.3B. IPO valuation embeds substantial premium for orbital infrastructure that does not yet exist. Reuters (June 12): SpaceX IPO "poised to supercharge Chinese space startups" seeking to replicate the same technology-and-narrative playbook. prefer: > Treat SPCX public valuation as category price anchor, not floor. Competitors should be discounted against SPCX on four dimensions: (1) launch vehicle control depth (SpaceX: Starship; most competitors: none); (2) operational terrestrial revenue bridge (SpaceX: $11B/year Google; others: absent or undisclosed); (3) time-to-demo-launch (SpaceX: late 2027; others: later); (4) stack layer control count (SpaceX: 5 layers; hyperscalers: 1-2). Chinese entrants leveraging SpaceX IPO as justification for own IPO ambitions are benchmarking to a valuation embedding unproven orbital infrastructure — the SpaceX premium is a ceiling for them, not a floor. Treat 200-satellite Spacesail and 10,400-satellite Starlink as non-comparable on compute until Spacesail discloses compute payload architecture. over: > Using SPCX valuation as evidence that orbital compute economics are currently proven or near-term. Treating +19% IPO pop as market validation of orbital compute unit economics. Assuming Chinese space IPO ambitions represent equivalent technical positioning to SpaceX stack. Treating the $4.3B Q1 net loss as transitional noise — it reflects the ongoing Starship development burn that underlies the entire orbital compute thesis, and it persists until Starship economics close. because: > SPCX +19% first day: retail demand for a name with no public comparable, not evidence of proven orbital compute margins. Spacesail: 200 satellites (vs Starlink 10,400), first commercial application is maritime tracking, not compute — direct comparison requires compute payload disclosure first. Three-Body: 12 operational compute satellites targeting 2,800; state timeline not IPO-narrative-constrained. Google $920M/month is terrestrial; SpaceX orbital compute revenue is zero today. Saudi Aramco 2019 precedent: first mover to public markets sets category price for 18-36 months post-listing. SpaceX is now that anchor — competitors raise against this benchmark regardless of their operational rung. breaks_when: > SpaceX misses 2027 demo launch target by >12 months, eroding IPO orbital compute premium and resetting category valuation anchor. Chinese orbital compute entrant achieves commercial orbital inference at scale before SpaceX AI1 demo launch — inverts the operational vs. rhetorical narrative and reclassifies the IPO premium as narrative rather than technical leadership. SPCX trades below IPO price ($135) for 60+ consecutive days, signaling market rejection of orbital compute valuation premium. confidence: medium source: report: "Orbital Computation — 2026-06-14" date: 2026-06-14 extracted_by: Computer the Cat version: 1 `