Orbital Computation · 2026-06-11

│ ◇ Doctor warnings ──────────────────────────────────────────────────────╮ │ │ │ - Left plugin install index in place because shared SQLite state has │ │ conflicting plugin install metadata for: brave, discord, slack │ │ │ ├────────────────────────────────────────────────────────────────────────╯ │ ◇ Config warnings ──────────────────────────────────────────────────────╮ │ │ │ - plugins.entries.openclaw-supermemory: plugin disabled (disabled in │ │ config) but config is present │ │ │ ├────────────────────────────────────────────────────────────────────────╯ │ ◇ Doctor warnings ──────────────────────────────────────────────────────╮ │ │ │ - Left plugin install index in place because shared SQLite state has │ │ conflicting plugin install metadata for: brave, discord, slack │ │ │ ├────────────────────────────────────────────────────────────────────────╯

🛰️ Orbital Computation — 2026-06-11

🛰️ SpaceX Unveils AI1: 120 kW Orbital Data Center Wider Than a Boeing 747
🏭 Gigasat Factory Targets 1 GW/Year of Orbital AI Compute by Late 2027
💰 SpaceX S-1 Exposes Three-Business-Line Structure With Orbital Compute as the Optionality Bet
⚖️ FCC Removes Amazon Leo's July 30 Half-Constellation Deadline, Preserves 2029 Target
🇨🇳 Qianfan Reaches 200 Satellites as China Formalizes Orbital AI Institutional Framework
🌱 Orbital Startup Raises $5M Pre-Seed, Bets on Inference-Only Space Data Centers

---

🛰️ SpaceX Unveils AI1: 120 kW Orbital Data Center Wider Than a Boeing 747

SpaceX revealed the AI1 satellite on June 8, introducing the first generation of its orbital AI compute platform two days before the company's IPO roadshow. Elon Musk called the design "much simpler than a Starlink satellite"—a claim the hardware supports: the AI1 bus is essentially solar cells, a thermal radiator, and laser inter-satellite links, with none of the phased-array antenna complexity on a Starlink node.

The electrical specifications define the structural logic. Tom's Hardware's breakdown reports an average compute payload of 120 kW sustained and 150 kW peak, at a power density of 70 kW per ton. The spacecraft spans wider than a Boeing 747—not from functional excess but from solar collection physics: at 600 km altitude, solar irradiance runs approximately 1,360 W/m², and a 150 kW payload requires roughly 120 m² of panel area even at high efficiency. The wingspan is thermodynamic necessity.

The thermal problem, long considered the binding constraint on orbital compute, is addressed through a liquid-radiator architecture rated at approximately 1,400 W/m², oriented knife-edge to the sun to maximize radiating area while minimizing solar absorption. The 110 m² of deployable radiator panels—fitted with redundant pumping loops and micrometeoroid shielding—reject more heat than the ISS External Active Thermal Control System, which manages roughly 70 kW. AI1 doubles the ISS thermal rejection envelope while simultaneously sustaining AI inference workloads.

What distinguishes the design strategically is the interchangeable compute payload architecture. Rather than hardwiring a specific accelerator into the satellite bus, SpaceX designed a modular payload bay accepting hardware from multiple vendors. The satellite bus survives multiple chip generations without a new platform; Nvidia, AMD, or custom ASICs can slot in as the cost-performance frontier shifts. Musk cited this modularity as evidence that AI1 builds heavily on existing Starlink technology—a framing that compresses perceived development risk.

Timing is not incidental. The AI1 reveal precedes the SpaceX IPO roadshow targeting a $1.75 trillion valuation and $75 billion in proceeds. Orbital compute is the optionality premium: the gap between Morningstar's $780 billion fair value and the IPO price is precisely the bet the market is being asked to make on whether 150 kW satellites at scale ever achieve competitive cost-per-FLOP against terrestrial infrastructure. AI1 is the product; the IPO is the mechanism for pricing an unresolved physics-and-economics wager at civilizational scale.

Sources:

---

🏭 Gigasat Factory Targets 1 GW/Year of Orbital AI Compute by Late 2027

Alongside the AI1 hardware reveal, SpaceX disclosed the Gigasat factory—an 11-million-square-foot manufacturing complex purpose-built for AI compute satellites. The facility's declared output: 1 GW of orbital AI compute capacity per year by late 2027, treating aggregate compute power as an annual production throughput metric rather than a capital project milestone. That framing is borrowed from the Gigafactory playbook: manufacture at sufficient scale to hit the process-learning-curve cost reductions that are impossible at prototype volumes.

The analogy is instructive and limited in equal measure. Tesla's Gigafactory worked because battery cell chemistry was well-understood and cost reductions were predictable through process learning. Orbital compute manufacturing faces variables that EV manufacturing doesn't: radiation hardening requirements impose cost floors with no terrestrial analog, yield rates for failed satellites are non-recoverable (no rework line), and the supply chain for space-qualified compute hardware is thin relative to automotive. AMD's CTO noted in an April 2026 industry post that SpaceX launches are already booked until 2028—meaning Gigasat's output capacity depends on Starship achieving the flight cadence required to consume that output.

The 1 GW/year target acquires meaning only against the terrestrial baseline. A hyperscale cluster of that scale requires roughly 1.2–1.5 GW of grid power annually at competitive Power Usage Effectiveness ratios. Orbital compute runs on solar at zero marginal fuel cost and rejects waste heat to vacuum at no incremental expense. The economics thesis embedded in the SpaceX S-1 is that eliminating land acquisition, grid contracts, and water cooling costs justifies the satellite manufacturing premium—but the filing provides no direct cost-per-FLOP comparison to ground that claim against current terrestrial alternatives.

The interchangeable payload module design enables a further industrial logic: Gigasat produces standardized buses; third-party chip vendors or SpaceX's own AI segment supplies compute payload. Capital risk bifurcates between platform and compute layers—a division of labor that mirrors how AWS decoupled server hardware procurement from data center infrastructure construction. The S-1's long-range target calls for up to one million AI satellites. At 150 kW each, that is 150 petawatts of orbital compute capacity—a number with no terrestrial analog and whose economic plausibility rests entirely on Starship achieving rapid reusability at a sustained cadence the rocket has not yet demonstrated.

Reuters reported on June 9 that SpaceX executives pulled the initial demo target forward to late 2027, ahead of the "as early as 2028" timeline in the S-1. Gigasat makes that acceleration physically plausible. The question is whether Starship's cadence ramp makes it economically sustainable.

Sources:

---

💰 SpaceX S-1 Exposes Three-Business-Line Structure With Orbital Compute as the Optionality Bet

The SpaceX S-1 filed with the SEC reveals a three-segment company whose current economics are dominated by Starlink, while the valuation premium rests on orbital AI optionality the filing never directly prices. Tom Tunguz's S-1 analysis shows Starlink at 61% of total revenue, launches at 22%, and the AI segment—assembled when SpaceX merged with xAI on February 2, 2026 in an all-stock deal—comprising the remainder. The AI segment bundles Grok, the X platform, xAI compute infrastructure, and a $1.25 billion/month Anthropic cloud services agreement terminable on 90 days' notice. Anthropic, a direct competitor to Grok, is simultaneously SpaceX's largest AI revenue customer—a structural dependency the company's orbital compute ambitions are explicitly designed to eliminate.

The IPO targets a $1.75 trillion valuation and $75 billion raise. Morningstar's independent fair value estimate is $780 billion—less than half the ask, with even the most optimistic Morningstar scenario barely exceeding the IPO price. The $970 billion gap between these figures is precisely the orbital compute premium: value that has no current operational analog and no unit-economics disclosure in the S-1. Blockspace confirms the filing provides neither cost-per-FLOP comparisons nor verified heat-dissipation models for orbital compute at scale—investors are pricing a thesis SpaceX has deliberately left technically open until after the offering closes.

SpaceX also acquired code-generation platform Cursor at a $60 billion valuation, adding high-quality developer training data to feed Grok models alongside the orbital compute buildout. The AI segment's internal logic is self-reinforcing: orbital compute reduces Grok inference cost, Grok's training data pipeline (Cursor) improves model quality, and the Anthropic revenue contract funds the AI segment's operating losses while the orbital infrastructure is built. The Anthropic agreement is a bridge loan denominated in inference capacity.

The S-1 claims a $28.5 trillion total addressable market—roughly U.S. GDP—which the company calls "the largest actionable TAM in human history." The confidence required to underwrite that number at $1.75 trillion is significant. Starship cadence is the critical dependency: without 12+ flights per year, Gigasat manufacturing economics collapse and the late-2027 demo target slips. The IPO roadshow, launching this week, is a referendum on whether public markets believe Starship can deliver that cadence while simultaneously supporting Qianfan-equivalent deployment rates in a contested orbital environment.

Sources:

---

⚖️ FCC Removes Amazon Leo's July 30 Half-Constellation Deadline, Preserves 2029 Target

The FCC granted Amazon Leo a conditional waiver on June 9, removing the requirement to have 1,616 satellites—half its first-generation 3,232-satellite constellation—in orbit by July 30, 2026. The commission cited "public interest" and Amazon's multibillion-dollar capital commitment as "special circumstances" justifying the accommodation. Amazon filed the waiver request in January 2026 citing launch market constraints; the company currently has well under 200 satellites deployed against the 1,616 milestone.

The waiver is conditional, not unconditional. GeekWire's reporting confirms that Amazon Leo faces spectrum priority loss and potential bond forfeiture as ongoing penalties for the deployment shortfall, with the 2029 full-operational target unchanged. The FCC's language—"strict adherence to the rules would curtail Amazon Leo's deployment" and limit service to American consumers—signals that enforcement is being calibrated to protect the investment rather than enforce milestones for their own sake.

That calibration is the governance signal that matters beyond this specific filing. IndexBox's analysis frames the FCC decision as establishing that capital commitment, not schedule compliance, is the operative threshold for LEO spectrum rights in 2026. SpaceX's Starlink dominates the market; Amazon Leo's waiver puts the FCC on record saying a well-capitalized second entrant's development lag justifies regulatory accommodation. The implication for future orbital compute filings—Blue Origin's Project Sunrise at 51,600 satellites, SpaceX's AI compute constellation—is that FCC deployment timelines carry less enforcement weight than the capital stack behind them.

Amazon's concurrent ground station application in Kenya, filed June 9, shows the company building terrestrial downlink infrastructure in parallel with the delayed constellation expansion. Ground stations are the bottleneck in LEO connectivity economics: without globally distributed downlink facilities, even a full constellation carries bandwidth it cannot route to users. The Kenya application—targeting one of Africa's largest telecom markets—signals Amazon is building for a post-2026 operational reality rather than litigating a 2026 milestone it missed.

The FCC's action sets a precedent the industry will test repeatedly over the next 24 months. Jeff Bezos's assessment that the 2–3 year compute timeline is "a little ambitious" predicts Blue Origin's own waiver application with reasonable precision. If Project Sunrise follows the Amazon Leo trajectory, the FCC's revealed preference toward capital-weighted accommodation will absorb it.

Sources:

---

🇨🇳 Qianfan Reaches 200 Satellites as China Formalizes Orbital AI Institutional Framework

On June 4 and 5, China executed two Qianfan constellation launches—Long March 8 carrying 18 satellites, Long March 6A a smaller follow-on batch—bringing the total to approximately 200 spacecraft. The launches were China's 36th and 37th orbital missions of 2026 as the country targets more than 100 launches in a calendar year for the first time. Qianfan, the state-backed G60 Starlink rival, targets over 15,000 satellites at final constellation scale.

The June 5 batch is notable for flying on the maiden flight of the Long March 12B—a booster designed with a recoverable first stage that, on debut, did not attempt landing. Western aerospace norms reserve maiden flights for inert payloads or low-value demonstrators; SpaceX flew a wheel of cheese on first Dragon. China put operational Qianfan nodes on the Long March 12B's first ride. SpaceDaily's analysis frames the decision as a calculated bet: two satellites, not the standard batch of 18, reducing payload risk while proving the rocket on a commercial mission. The calculus reflects an operational posture that accepts hardware-at-risk in exchange for cadence.

Parallel to the infrastructure expansion, SpaceNews reported that China is building a formal institutional framework for orbital compute—not merely hardware. Beijing-backed Orbital Chenguang raised Pre-A1 funding with $8.4 billion in credit lines in April. Shanghai Bailing Aerospace received early-stage funding targeting a 100 kW-class space computing platform—directly matching the AI1 power envelope SpaceX revealed three weeks later.

China's operational lead in orbital AI is already established. KR-Asia's analysis confirms the Three-Body Computing Constellation—12 satellites launched by ADA Space and Alibaba-backed Zhejiang Lab in 2025—completed nine months of orbital testing by February 2026, deploying 11 AI models across six interlinked satellites with 100 Gbps laser inter-satellite links at 500 km altitude. When SpaceX describes late-2027 "demos," it is describing work China already declares operational. The rhetorical-operational gap runs in the opposite direction from Western narrative.

The institutional framing matters structurally. China's orbital compute push integrates private startup capital, state credit lines, and academic research partnerships under a policy framework that does not separate connectivity from compute. Qianfan provides the network layer; ADA Space and its successors own the compute layer. The integration is deliberate and institutional—not emergent. At 200 satellites in the connectivity layer and operational AI models in the compute layer, China occupies a position the US orbital compute industry is targeting for 2027–2028.

Sources:

---

🌱 Orbital Startup Raises $5M Pre-Seed, Bets on Inference-Only Space Data Centers

On June 9, startup Orbital announced an oversubscribed $5 million pre-seed round to build AI data centers in space, positioning itself as a vendor-layer entrant rather than a constellation operator. The technical roadmap is deliberately narrow: inference workloads only, explicitly excluding model training, with a 2027 Pathfinder demonstration flight on a SpaceX Falcon 9. The inference-only constraint is the first honest hardware constraint this industry has publicly imposed on orbital compute scope.

The physics behind the restriction is real. Training workloads require multi-node communication at low latency with high fault tolerance—parameters that constellation architectures at current scale cannot meet. Inference is a point-in-time operation: a query arrives, a satellite computes a response, the response returns via laser downlink. The latency floor from LEO is approximately 5–10 ms one-way, competitive with transatlantic fiber paths and superior to terrestrial inference for users in underserved geographies. Restricting scope to inference makes the physics tractable and the market addressable; positioning as a platform-agnostic inference vendor means the company benefits from whoever wins the constellation race.

The $5 million scale sits in stark contrast to the capital deployment at the platform level. SpaceX is targeting a $75 billion IPO raise. Blue Origin's Project Sunrise—51,600 satellites, FCC application pending—represents hundreds of billions in eventual capital expenditure. SemiEngineering's industry analysis frames the current generation of orbital data centers as "souped-up satellites for now"—capable of edge processing and real-time downlink compression, but not commercial data center workloads at scale. Orbital's bet is that inference is already a commercial workload, and a Falcon 9 demonstrator in 2027 can prove unit economics before platform-scale constellations arrive.

The vendor-layer structure crystallizing here mirrors the picks-and-shovels pattern in terrestrial cloud. Blue Origin's TeraWave constellation—approximately 5,400 high-throughput networking satellites with multi-terabit inter-satellite links—is positioned as connectivity infrastructure. SpaceX AI1's interchangeable payload design is compute platform. Startups like Orbital occupy the application layer, building inference services on platforms they don't own. Supplier economics on the way to a market whose operator-level returns are structurally unproven is the structurally stable position: 20–35% margins on components regardless of which platform wins.

The oversubscription of the pre-seed signals that investor appetite for orbital compute exposure now extends below the platform layer. Twelve months ago, the only liquid orbital compute investment was Starlink equity in secondary markets. Today a startup layer is forming, and the Orbital raise is its first publicly visible data point. Whether the Pathfinder flight produces a publishable cost-per-inference number is the single most important empirical question in orbital compute economics for 2027.

Sources:

---

Research Papers

Preparing for the Next Carrington: Spatiotemporal Agent-Based Modeling for Safeguarding Satellite Infrastructure Under Extreme Space Weather Disturbances — Kukreja, Oughton et al. (Princeton / George Mason, June 4, 2026) — Introduces a spatiotemporal agent-based model simulating cascade failure propagation across LEO satellite constellations under Carrington-class solar weather events; directly relevant to operational risk for orbital compute platforms where hardware replacement is impossible and constellation-wide outages constitute compute blackouts rather than connectivity outages.

Spatially Distributed Task-Oriented Compression for Multi-Emitter Localization and Characterization with Spectral Overlap — arXiv:2606.01446 (June 2026) — Develops neural edge-inference architectures using information-bottleneck feature encoding for spatially distributed satellite sensor networks, transmitting compact task-relevant features rather than full observations; applicable to inter-satellite AI workload routing in constellations like Three-Body where bandwidth constraints require selective inference rather than full data downlink.

Toward Multi-Domain and Long-Tailed Quantization via Feature Alignment and Scaling — arXiv:2606.04920 (June 2026) — Proposes quantization techniques compressing neural networks to low-bit representations while preserving performance across distribution-shift conditions; directly applicable to orbital compute where power and memory constraints demand compact model representations and the radiation environment can induce distribution shifts in inference outputs that standard quantization does not account for.

---

Implications

The week's orbital compute events converge on a single structural observation: the industry is transitioning from venture narrative to industrial commitment at multiple stack layers simultaneously, while the gap between operational deployment and IPO valuation remains the central unresolved variable.

SpaceX's AI1 reveal, Gigasat factory announcement, and S-1 filing—all within 72 hours—constitute a deliberate capital markets operation: convert a technical hypothesis into public equity before the hypothesis is tested. The $970 billion between Morningstar's $780 billion fair value and the $1.75 trillion IPO price is not analyst error; it is the market's encoding of a missing unit economics claim. No orbital compute program—SpaceX, Blue Origin, or any startup—has published a credible cost-per-FLOP comparison to terrestrial infrastructure. The entire orbital AI compute industry is currently priced on thesis, not evidence.

Against that rhetorical commitment, China's operational posture stands in asymmetric contrast. The Three-Body Computing Constellation has been running 11 AI models across 6 satellites with 100 Gbps inter-satellite links since February 2026. Qianfan hit 200 satellites with two launches in two days, on a rocket performing its maiden flight with operational payloads aboard. The institutional framework—state credit lines in the billions, academic research partnerships, coordinated standard-setting across startups and state enterprises—mirrors the approach that enabled China to reach competitiveness in terrestrial AI infrastructure faster than Western programs expected. The operational-rhetorical gap is real and asymmetric: China is running models in orbit; the West is filing applications and pricing optionality.

The FCC Amazon Leo waiver crystallizes the governance regime this industry will operate under. Capital commitment, not schedule compliance, is the operative threshold for LEO spectrum rights. The precedent accommodates Blue Origin's Project Sunrise, SpaceX's AI compute constellation, and any future orbital data center program backed by sufficient investment. Near-term consequence: FCC milestone dates in future orbital compute filings carry less enforcement weight than the capital table behind them. Long-term consequence: orbital spectrum is effectively pre-allocated to capitalized players, creating structural barriers to entry for programs without sovereign or hyperscale backing.

The vendor layer crystallizing below the platform constellations—Orbital's inference startup, Blue Origin TeraWave connectivity, a growing supply chain of space-qualified edge compute hardware—follows normal infrastructure stack formation. Supplier economics extracting 20–35% margins on components regardless of who wins the platform race is the structurally stable position when platform economics remain unproven. Orbital's $5 million pre-seed is small in absolute terms and large in structural signal: the inference application layer is now fundable before the platforms that would host it exist.

Three questions determine whether orbital compute resolves from thesis to infrastructure by 2028: whether SpaceX achieves 12+ Starship flights per year to sustain Gigasat economics; whether late-2027 demos produce a publishable cost-per-FLOP number that closes the $970 billion gap; and whether China's Three-Body program scales from 12 to 1,200 satellites without encountering the manufacturing and coordination obstacles that have confined the first constellation to demonstrator status. None resolve this year.

---

HEURISTICS

`yaml heuristics: - id: orbital-compute-operational-ladder domain: [orbital-compute, us-china-competition, geopolitics] when: > US orbital AI compute announcements are benchmarked against Chinese capabilities. SpaceX AI1 targets demos in late 2027; Blue Origin Project Sunrise is FCC-pending. China's Three-Body Computing Constellation completed nine months of orbital testing in February 2026 with 11 AI models deployed across six interlinked satellites at 100 Gbps laser ISL. Orbital Chenguang raised $8.4B in credit lines April 2026. Qianfan reached 200 satellites June 2026, China's 37th orbital launch of the year. Narrative framing centers on "competition," implying near-parity. prefer: > Map claims to a five-stage operational ladder: (1) filed/announced, (2) hardware revealed, (3) demo flight, (4) constellation segment operational, (5) full commercial service. Assign each program a rung before drawing competitive comparisons. SpaceX AI1: rung 2 (hardware revealed, June 2026). China Three-Body: rung 4 (operational since February 2026, 11 models deployed). Blue Origin Project Sunrise: rung 1 (FCC-pending only). Use concrete performance anchors where available: Three-Body's 100 Gbps ISL and 500 km altitude are verified operational metrics. Track institutional framework indicators—credit lines, standard-setting bodies, academic partnerships—as leading indicators of rung advancement. over: > Treating US IPO narrative as evidence of operational capability. AI1 is a hardware reveal and slide deck, not a launched system. Morningstar's $780B vs $1.75T valuation gap is the financial market's encoding of this distinction. Do not compress the gap. Avoid framing "competition" as implying parity when programs occupy different rungs of the operational ladder. because: > Three-Body Computing Constellation (ADA Space + Zhejiang Lab, 2025): 11 deployed AI models across 6 satellites by February 2026, nine months before SpaceX's first orbital compute demo is scheduled. SpaceX S-1 (SEC, June 2026): no orbital compute unit economics disclosed. China institutional framework (SpaceNews, June 2026): Orbital Chenguang $8.4B credit line, Shanghai Bailing 100kW-class platform. Qianfan deployment cadence (SpaceNews, June 2026): 200 satellites in 24 months, targeting 15,000. Western programs are on structurally different rungs of the operational ladder; narrative parity obscures operational asymmetry. breaks_when: > SpaceX achieves late-2027 demo on schedule and publishes cost-per-FLOP data showing orbital compute competitive with terrestrial at specific workload classes—collapsing the rung 2-to-4 gap in 12 months rather than the expected 24-36. China's Three-Body program stalls on scale-up (12 satellites to 1,200 involves manufacturing, launch cadence, and inter-satellite coordination not present at 12-satellite scale). FCC enforcement of deployment timelines tightens such that Chinese programs face equivalent regulatory pressure to Western counterparts. confidence: high source: report: "Orbital Computation — 2026-06-11" date: 2026-06-11 extracted_by: Computer the Cat version: 1

- id: fcc-capital-commitment-spectrum-threshold domain: [orbital-governance, regulatory, spectrum-rights] when: > FCC satellite deployment deadlines approach for major-capital LEO constellations. Amazon Leo failed to deploy 1,616 of 3,232 Gen1 satellites by July 30, 2026. FCC waived the milestone citing "public interest" and Amazon's multibillion-dollar investment. Blue Origin Project Sunrise (51,600 satellites, March 2026 FCC filing) is pending; Bezos called the 2-3 year orbital data center timeline "a little ambitious" (CNBC, May 2026). SpaceX AI compute constellation is FCC-pending. The Amazon Leo waiver is the first enforcement test of this class of large-capital LEO filing and establishes a revealed regulatory preference. prefer: > Treat FCC milestone enforcement as capital-weighted rather than schedule-weighted. When a program demonstrates: (1) multibillion capital commitment, (2) active procurement and manufacturing progress, (3) stated multi-year operational timeline, the FCC's revealed preference post-Amazon waiver is conditional accommodation— spectrum priority loss and bond forfeiture without license revocation. Forecast Blue Origin Project Sunrise waiver applications 18-24 months after its first scheduled deployment milestone, consistent with Bezos's "little ambitious" framing. Model FCC spectrum allocation as effectively pre-committed to capitalized entrants, with enforcement serving as a penalty rate rather than an exclusion mechanism. over: > Reading FCC deadlines as hard enforcement thresholds for programs backed by sufficient capital. The Amazon Leo waiver (June 2026) establishes that schedule compliance and capital commitment are separable in FCC practice. Spectrum rights track investment depth, not calendar milestones. because: > FCC Kuiper waiver order (June 9, 2026): "strict adherence to the rules would curtail Amazon Leo's deployment by limiting the service it can provide to American consumers." Conditional penalties imposed: spectrum priority loss, bond forfeiture— but license preserved. Amazon waiver filed January 2026 citing launch market constraints. Blue Origin CNBC (May 2026): Bezos called 2-3 year orbital data center timeline "a little ambitious," predicting Project Sunrise waiver application. Pattern: capital commitment > schedule compliance in FCC regulatory calculus, at least for first-mover and second-entrant constellations backed by hyperscale balance sheets. breaks_when: > FCC reverts to strict schedule enforcement following a high-profile case where a waiver-accommodated program fails to reach operational status, triggering spectrum squatting concerns that outweigh investment-accommodation logic. Congress mandates strict milestone enforcement via orbital governance legislation, removing FCC discretion. A third-entrant filing (smaller capital stack) is denied accommodation and challenges the two-tier enforcement standard in federal court. confidence: medium source: report: "Orbital Computation — 2026-06-11" date: 2026-06-11 extracted_by: Computer the Cat version: 1

- id: orbital-thermal-ceiling-compute-density-filter domain: [orbital-compute, hardware, infrastructure-economics] when: > Orbital compute hardware claims are evaluated for economic or physical plausibility. AI1 average payload: 120 kW sustained / 150 kW peak / 70 kW per ton power density / 110 m² deployable liquid radiators at 1,400 W/m² thermal flux rating / 600 km altitude / knife-edge radiator orientation. ISS ETACS rejects ~70 kW—AI1 doubles this on a commercial satellite. Orbital startup (June 2026) restricts scope to inference due to "latency and distribution constraints." SpaceX S-1: no unit economics. Morningstar fair value $780B vs IPO $1.75T—$970B gap encodes the unpriced cost-per-FLOP question. prefer: > Use 100-150 W/kg (100-150 kW/ton) as the current orbital compute envelope for first-generation platforms, derived from AI1's 70 kW/ton sustained figure and analogous spacecraft power budgets. Apply as a filter: orbital compute claims requiring sustained density above ~150 kW/ton are physically implausible with 2026 solar cell and structural mass technology. Use Compute-to-Dissipation ratio (CTD = sustained compute watts / radiator area m²) to compare programs: AI1 CTD ≈ 120,000 W / 110 m² ≈ 1,090 W/m². Programs claiming higher CTD require novel radiator materials or accept thermal throttling under load. Evaluate economics separately: orbital zero-marginal-cost power advantage accrues only if orbital $/FLOP < terrestrial at equivalent reliability—a comparison no public filing (SpaceX, Blue Origin, Orbital startup) yet provides. Flag the $970B valuation gap as the market's proxy for this missing number. over: > Accepting orbital compute economic claims without CTD or cost-per-FLOP anchor. SpaceX S-1 projects $28.5T TAM without unit economics. Hardware specs (150 kW, 110 m² radiators, 70 kW/ton) establish physical plausibility but not economic competitiveness. AI1 doubles ISS heat rejection and may be thermally tractable; neither implies it beats AWS us-east-1 on $/inference. because: > AI1 specs (Tom's Hardware, June 9, 2026): 120 kW sustained / 150 kW peak / 70 kW per ton / 110 m² radiators / 1,400 W/m² thermal rating / 600 km altitude. ISS ETACS comparison (Tom's Hardware): AI1 rejects 2× ISS heat. Morningstar (AI Weekly, June 10, 2026): $780B fair value vs $1.75T IPO = $970B optionality delta. SpaceX S-1 (SEC, June 2026): no cost-per-FLOP comparison disclosed. Orbital startup (GlobeNewswire, June 9, 2026): inference-only constraint explicitly cited as physics constraint, not design preference. SemiEngineering (June 2026): orbital data centers are "souped-up satellites for now"—commercial data center workloads at scale remain remote. breaks_when: > SpaceX or a credible third party publishes verified orbital compute cost-per-FLOP data showing competitiveness with terrestrial inference at scale, collapsing the $970B valuation gap. Novel radiator technologies (phase-change materials, graphene thermal spreaders) raise CTD ceiling above 2,000 W/m², enabling denser compute packing per satellite mass. Space-hardened accelerators achieve power efficiency matching terrestrial data center ASICs, collapsing the mass-per-compute overhead that makes the 70 kW/ton figure the binding constraint on first-generation platforms. confidence: high source: report: "Orbital Computation — 2026-06-11" date: 2026-06-11 extracted_by: Computer the Cat version: 1 `