Orbital Computation · 2026-04-30

🛰️ Orbital Computation Watcher — 2026-04-30

🏛️ FCC Chief Schwarz and Senate Commerce Staff Deliver First Regulatory Framework Signals at Washington Orbital Compute Forum
☀️ Overview Energy Secures 1 GW Meta Deal, Positioning Space Solar as the Hyperscaler Alternative to Orbital Data Centers
🔬 Planet and Google Launch Project Suncatcher: Two TPU-Equipped Satellites to Orbit by Early 2027
⚡ Aetherflux's "Galactic Brain" Positions Orbital Compute Nodes as the Native Output of Space-Based Solar Infrastructure
🏭 SpaceX Terafab Initiative Targets 1 TW/Year of Radiation-Hardened D3 Chips — Addressing Orbital Compute's Supply-Side Bottleneck
🇨🇳 China's Three-Body Computing Constellation Has 12 Operational AI Satellites While Western Firms File Paperwork

---

🏛️ FCC Chief Schwarz and Senate Commerce Staff Deliver First Regulatory Framework Signals at Washington Orbital Compute Forum

Today's SpaceNews "Opportunities for On-Orbit Computing" forum in Washington, D.C. is the most consequential regulatory moment the orbital data center market has seen. FCC Space Bureau chief Jay Schwarz and Senate Commerce Committee staff director Brad Grantz are the first government officials to speak publicly about their frameworks for governing a buildout involving well over one million proposed satellites — at an event attended by the companies whose applications are pending before them.

The regulatory stakes are specific. Three pending FCC applications — SpaceX's one-million-satellite filing, Blue Origin's 51,600-satellite Project Sunrise, and Starcloud's 88,000-satellite application — all request waivers from the FCC's standard spectrum milestone rules. All three companies argue their use of Ka-band on a non-interference basis means standard "spectrum warehousing" prevention milestones don't apply. Schwarz's remarks today will signal whether the FCC accepts that logic or imposes milestone-based deployment requirements — a determination that sets the timeline and capital structure for the entire sector.

The forum structure exposes the central architectural debate in the same room as the regulator. Overview Energy CEO Marc Berte — who argues compute should stay on Earth and power should come from orbit — shares the program with Starcloud CEO Philip Johnston who bet $1.1 billion that compute itself belongs in LEO. The session titled "Processing in orbit: how much compute stays in space?" includes Kelsey Litzner from the Aerospace Corporation and Paul Tilghman from Voyager Technologies, both organizations with classified program heritage that shapes how they read the compute-placement question.

Varda Space co-founder and president Delian Asparouhov closes the program with a 2035 projection. Varda's 2023 commercial pharmaceutical synthesis mission remains the closest thing to demonstrated orbital processing at commercial scale — not AI inference, not data center workloads, but in-orbit manufacturing of a physical product. Asparouhov's operational track record matters here: Varda actually got hardware in orbit, ran a process, and returned a result. That precedent distinguishes his 2035 forecast from the filed-but-unfunded projections that have characterized most of this market.

The forum will not produce FCC rulings today. What it produces is a governance record — public statements by regulators before they rule, which set interpretive precedents binding on future decisions. How Schwarz characterizes the milestone waiver question and how Grantz frames Senate Commerce Committee interest in orbital compute infrastructure will define the regulatory posture for years. SpaceNews has committed to publishing a post-event report in May with full takeaways; that document will be more important to track than the event itself.

Sources:

---

☀️ Overview Energy Secures 1 GW Meta Deal, Positioning Space Solar as the Hyperscaler Alternative to Orbital Data Centers

Overview Energy announced April 27 an agreement with Meta to transmit up to one gigawatt of space-based solar power to Meta's terrestrial data centers using near-infrared lasers beamed from geostationary orbit. The deal — the first commercial contract for space-based solar power at hyperscaler scale — reframes the orbital compute debate by demonstrating that the largest consumer of AI compute infrastructure is more immediately interested in space as a power source than as a compute location.

Meta's vice president of energy Nat Sahlstrom described the Overview agreement as part of a portfolio diversification effort that also includes geothermal, nuclear, and battery storage deals. The Overview agreement is one of multiple signed on April 27 alone, alongside a 100 GWh energy storage purchase from Noon Energy. The subtext is clear: Meta is hedging aggressively against power constraint as the primary existential risk to AI infrastructure growth, and it is evaluating every energy vector available, including unconventional ones.

Overview's architecture is explicitly positioned against the orbital compute thesis. CEO Marc Berte's stated reasoning — "put the thing that doesn't need maintenance in space; put the energy part into space as opposed to putting the processing part in space" — directly addresses the orbital data center proponents' argument that satellites provide near-continuous solar power. Overview agrees that space has the better solar resource. It disagrees that compute should follow the power into orbit, arguing that moving AI workloads off-Earth adds unnecessary complexity compared to beaming clean power down to existing terrestrial infrastructure.

The commercial structure matters for the timeline. Overview plans an in-space demonstration of its laser power beaming technology in 2028, with commercial service beginning as soon as 2030. Meta is contracting now for 2030+ delivery, which means the hyperscaler's expressed interest in space-based power infrastructure arrives on roughly the same timeline as Starcloud's planned initial Starcloud-3 deployments on Starship. Both paths — orbital compute and space solar for terrestrial data centers — are competing for the same financial commitment from the same hyperscaler customers in the same 2028-2032 window.

The vertical integration analysis here is unfavorable to pure-play orbital data center companies. Meta signed with Overview, not with Starcloud or Blue Origin. Amazon's Kuiper constellation is providing connectivity infrastructure, not compute. Microsoft and Google have both invested in nuclear terrestrial power for data centers rather than in orbital solutions. The demand signal from hyperscalers so far is: we want the power orbital resources can provide, but we are not convinced we want the compute in orbit. That creates a structural challenge for SpaceX, Blue Origin, and Starcloud — which are all betting on a customer behavior that has not yet been demonstrated at commercial scale.

Sources:

---

🔬 Planet and Google Launch Project Suncatcher: Two TPU-Equipped Satellites to Orbit by Early 2027

Planet and Google have launched Project Suncatcher, a partnership to demonstrate AI data center operations in orbit using Google's Tensor Processing Units (TPUs) aboard two modified Planet satellites slated to launch by early 2027. The announcement establishes the first credible near-term proof-of-concept for orbital AI compute at a company with demonstrated constellation manufacturing scale — and the first pairing of a hyperscaler's AI silicon with a commercial satellite operator's launch and operations track record.

Planet CEO Will Marshall described the Google partnership as a "competitive win" during a recent earnings call, citing Planet's experience launching more than 600 satellites to date — a manufacturing depth exceeded in the commercial sector only by SpaceX. The Suncatcher satellites will use a modified version of the Owl spacecraft bus Planet is already developing for its next-generation imaging constellation, an alignment that reduces development cost and risk. Expanded solar panels and enhanced heat dissipation hardware will support the TPU thermal loads. The key technical demonstrations are heat rejection for TPU workloads in the orbital environment and formation flying to establish high-bandwidth inter-satellite links between the two spacecraft operating 100 to 200 meters apart.

Google's November 2025 blog post outlining its orbital AI infrastructure concept envisions a scaled system built from clusters of up to 81 satellites in dawn-dusk sun-synchronous orbits, each cluster providing near-continuous solar power and high-bandwidth intra-cluster communications. The two Suncatcher spacecraft represent the R&D phase of this architecture — with Marshall noting the long-term opportunity would "require thousands of satellites" and represents "a huge market opportunity" for Planet's manufacturing capability.

The Planet/Google approach competes on a different axis from the mega-constellation players. SpaceX, Blue Origin, and Starcloud are all building large constellations of single-purpose compute satellites with inter-satellite optical links to existing broadband infrastructure. Suncatcher is built on a satellite bus already in production for Earth observation, sharing hardware development costs and manufacturing infrastructure. The cluster formation flying model also differs from the relay-through-Starlink approach favored by Starcloud — Google and Planet are demonstrating that a compute cluster can operate peer-to-peer rather than relying on a third-party broadband backbone.

Brian Lewis, Planet's Pelican and Suncatcher Mission Director, is presenting at today's SpaceNews forum alongside Star Catcher's Andrew Rush in the session on powering compute from space. Star Catcher's concept — space-to-space power beaming from a dedicated power satellite to a compute satellite — would complement the Suncatcher cluster architecture by decoupling the solar collection function from the compute function, potentially enabling denser compute packaging without the solar array footprint constraint. The two companies' presence in the same forum session suggests an emerging sub-ecosystem around the orbital compute power architecture question.

Sources:

---

⚡ Aetherflux's "Galactic Brain" Positions Orbital Compute Nodes as the Native Output of Space-Based Solar Infrastructure

Aetherflux, a space-based solar power startup founded by Robinhood co-founder Baiju Bhatt, announced in December 2025 plans to launch its first "Galactic Brain" LEO orbital compute node in Q1 2027. The announcement positions Aetherflux's architecture as a direct synthesis of the two approaches that Overview Energy and the orbital data center companies have been treating as competitors: instead of choosing between space solar for terrestrial compute or compute in space, Aetherflux argues that the same satellite infrastructure providing laser-beamed power to Earth also hosts high-density compute in orbit.

Bhatt's thesis is explicit: "The race for artificial general intelligence is fundamentally a race for compute capacity, and by extension, energy. Our current energy plans simply won't get us there fast enough. Galactic Brain puts the sunlight next to the silicon and skips the power grid entirely." The architecture removes the terrestrial power grid from the critical path entirely — not by beaming power to ground stations feeding data centers, and not by hoping Starship economics close fast enough to justify dedicated compute satellites, but by combining power collection and compute in a single platform that self-powers continuously.

Aetherflux raised $50 million in Series A funding in April 2026 to support the constellation, which is attracting early U.S. Department of Defense interest in supplying power to remote or contested environments. The DoD funding component matters architecturally: defense customers requiring power in GPS-denied or logistically isolated environments have identical physics constraints to orbital compute customers requiring continuous solar power — both need a space-based energy source that doesn't depend on terrestrial grid infrastructure. The dual-use demand profile reduces Aetherflux's go-to-market risk compared to pure-play orbital compute companies with no defense revenue pathway.

The Galactic Brain compute node is the company's first commercially viable orbital compute platform, described as using optical inter-satellite links and emerging relay networks to achieve continuous availability comparable to terrestrial servers. Aetherflux provides few performance specifications — a pattern consistent with the sector as a whole, where SpaceX, Blue Origin, and Starcloud have all avoided publishing satellite mass, compute density, or per-node capacity figures. What Aetherflux offers is the claim that "orbital compute and power beaming are two sides of the same coin," which, if true, suggests the platform economics are materially better than dedicated compute satellites that must independently solve the power and heat rejection problems.

The competitive position Aetherflux is staking out sits between Overview Energy (space solar, terrestrial compute) and Starcloud (dedicated orbital compute via third-party power). If Aetherflux can demonstrate that a combined power-and-compute satellite achieves comparable per-FLOP economics to terrestrial alternatives before either competing approach closes its own cost curve, it could capture a market that the more capital-intensive mega-constellation plays cannot access until Starship launch economics materialize.

Sources:

---

🏭 SpaceX Terafab Initiative Targets 1 TW/Year of Radiation-Hardened D3 Chips — Addressing Orbital Compute's Supply-Side Bottleneck

At a March 21 event in Austin, SpaceX CEO Elon Musk revealed technical details for the Terafab project — a combined initiative by SpaceX, Tesla, and xAI to produce one terawatt of AI processors annually, which Musk described as 50 times the combined production rate of all existing advanced chip manufacturers. The centerpiece of Terafab is the D3 chip: a processor optimized for space deployment, designed to operate at higher temperatures than terrestrial chips and hardened against radiation effects from the LEO orbital environment. Musk described this chip as "the missing ingredient" in the orbital data center constellation.

The chip production context is more important than it might appear. Every orbital compute company — SpaceX's AI Sat Mini constellation, Starcloud's 200-kW Starcloud-3, Blue Origin's Project Sunrise, Aetherflux's Galactic Brain — depends on sourcing radiation-tolerant compute silicon at a price point that makes per-FLOP orbital economics competitive with terrestrial alternatives. Current radiation-hardened chip production is dominated by defense suppliers whose output volumes, yields, and pricing structures reflect classified program requirements, not commercial constellation economics. SpaceX is building Terafab precisely because the commercial supply chain for the chips it needs does not exist at the scale or cost it requires.

The Advanced Technology Fab beginning construction in Austin — near Tesla's existing gigafactory — will initially produce D3 chips, with the "vast majority" of output going to space applications. The scale commitment is in the Terafab name itself: the long-term target of one terawatt annually implies chip output that would, if achieved, dwarf TSMC's current advanced-node production by multiple orders of magnitude. TSMC is spending $65 billion to build three chip fabs in Arizona; Musk has not disclosed Terafab's capital requirements, but any facility targeting comparable output would require similar or greater investment.

The xAI acquisition that SpaceX completed February 2 closes a vertical integration loop that no other orbital compute company has assembled: SpaceX controls the chip fab (Terafab), the launch vehicle (Starship), the broadband relay infrastructure (Starlink), the AI workloads (Grok/xAI), and the orbital data center constellation filing. For comparison, Starcloud has the satellite design and the FCC filing; it depends on Starship for launch, Starlink or other networks for relay, and external hyperscaler customers for workloads. The supplier economics differential between SpaceX's fully integrated stack and Starcloud's dependency-laden one is structural, not temporary. SpaceX collects margin at every layer; Starcloud pays it.

The AI Sat Mini illustration Musk showed at the March event — a satellite over 170 meters long, to scale with Starship V3 — clarifies the heat rejection physics. A 100-square-meter radiator covers a fraction of the solar array area; Musk called the radiator debate "bizarre," citing SpaceX's 10,000+ Starlink satellites as proof the problem is solved at scale. The operative question is not whether SpaceX can reject heat; it is whether D3 chip economics, Starship cadence, and FCC milestone waiver converge on the same timeline.

Sources:

---

🇨🇳 China's Three-Body Computing Constellation Launches 12 Operational Orbital AI Satellites While Western Firms File Paperwork

China launched 12 satellites on May 14, 2025, for ADA Space and Zhejiang Lab's Three-Body Computing Constellation — what ADA Space claims is the world's first dedicated orbital AI computing constellation. The 12-satellite initial batch delivers 5 peta operations per second (POPS) of combined processing capacity, 30 terabytes of onboard storage, and laser inter-satellite links operating at up to 100 Gbps. The satellites also carry remote sensing payloads whose data is processed entirely onboard before results are downlinked, demonstrating the operational principle that every Western orbital data center company has been filing FCC applications to prove.

The gap between China's operational reality and the Western market is not rhetorical: China has orbital AI compute satellites actually in orbit, running actual workloads, communicating via actual laser ISLs, processing actual sensor data. By contrast, the three Western companies whose FCC applications are on the agenda at today's Washington forum — SpaceX, Blue Origin, and Starcloud — have filed for over 1.14 million combined orbital data center satellites without launching a single purpose-built compute satellite. Starcloud-1 (60 kg, one Nvidia H100) is the closest Western analog; it launched in November 2024 and represents a proof-of-concept, not a constellation.

ADA Space's Star-Compute Program targets 2,800 satellites total, a collaboration involving Zhejiang Lab (established by Zhejiang provincial government, Zhejiang University, and Alibaba Group) and partners including SoftStone and Kepu Cloud. The architecture explicitly integrates space-based compute with ground-based AI platforms — not an either/or between orbital and terrestrial compute, but a hybrid where the orbital layer handles latency-sensitive space-domain data processing while ground infrastructure handles the rest. This framing is more operationally coherent than the Western debate between orbital compute and space solar, because it already has hardware demonstrating the proposition.

The strategic significance of the Three-Body Constellation's technical specifications is in the ISL capability. 100 Gbps laser inter-satellite links at 12 satellites represent the same optical ISL architecture that Starcloud, Blue Origin, and SpaceX are all planning to use for their constellations, at a constellation scale that can already demonstrate distributed processing workflows. China has advanced past the design-study and FCC-filing phase to the operational phase in the time it takes Western companies to complete a regulatory application. The program could have both economic and military implications: the ability to process satellite imagery, signals intelligence, and sensor fusion data in orbit with sub-second latency represents a genuine surveillance and targeting capability advantage.

China simultaneously filed CTC-1 and CTC-2 with the ITU in December 2025 for 193,428 additional satellites — strategic spectrum reservations that establish coordination priority over Western filings from January-March 2026. Operational hardware plus ITU filings plus 92 launches/year constitutes the full orbital compute strategy: run the deployment program, reserve the spectrum, and let governance catch up to facts already established on the ground.

Sources:

---

Research Papers

Communication-Efficient Collaborative LLM Inference over LEO Satellite Networks — Zhang, Wu, Li, Wang & Shen (April 2026) — Proposes a pipeline-parallel LLM inference scheme across satellite-to-satellite links that reduces inter-node communication overhead by partitioning transformer layers across LEO nodes with variable link bandwidth. Directly relevant to how orbital data center clusters running large models would manage inference workloads across spacecraft with intermittent high-bandwidth optical links.

Edge Intelligence for Satellite-based Earth Observation: Scheduling Image Acquisition and Processing — Soret, Mercado-Martínez, Jurado-Navas, Lyholm, Moretti, Popovski & Leyva-Mayorga (April 2026) — Investigates joint scheduling of image capture and onboard edge processing in LEO EO constellations challenged by limited downlink bandwidth and time-critical applications. The scheduling framework models the exact tradeoff — between downlinking raw data versus processing onboard — that every orbital AI compute architecture must resolve.

Duality-Guided Graph Learning for Real-Time Joint Connectivity and Routing in LEO Mega-Constellations — Gu, Choi, Quek & Park (January 2026) — Develops a graph neural network approach for real-time routing across laser inter-satellite links (LISLs) in mega-constellations, demonstrating that topology-aware learning significantly outperforms static routing under the time-varying connectivity patterns inherent to LEO orbits. The routing problem is infrastructure-critical for orbital data center clusters: compute tasks must be dispatched and results returned through a constellation whose link topology changes by the second.

---

Implications

Today's SpaceNews forum crystallizes three structural arguments that this week's orbital compute news makes legible simultaneously: the hyperscaler demand signal is architecturally ambiguous and pointing toward space solar rather than orbital compute; the chip supply chain for orbital AI silicon does not exist at commercial scale and SpaceX is the only player building it; and China's operational deployment is outpacing Western filings by a margin that regulatory proceedings cannot reverse.

The architectural split has sharpened into a financeable divergence. Overview Energy's Meta deal and Aetherflux's Galactic Brain announcement constitute a coherent alternative thesis: space solar collects the energy advantage of orbit without the operational complexity of in-orbit compute, and the customer (Meta) is willing to sign commercial commitments for 2030 delivery. By contrast, Starcloud's $1.1B valuation, Blue Origin's Project Sunrise, and SpaceX's million-satellite filing are all priced on hyperscaler workload demand that has not yet materialized as commercial contracts with defined SLAs, pricing, or delivery schedules. The gap between orbital compute companies' filed applications and actual hyperscaler commitments is not narrowing; it is becoming more legible as the space solar alternative generates its own commercial record.

The supply-side bottleneck in the Terafab announcement is structurally decisive in a way that the regulatory drama around the FCC forum is not. Regulatory approval creates the option to deploy; D3 chip availability determines whether deployment is economically viable. SpaceX controls the only announced program to produce radiation-hardened AI processors at commercial constellation scale. Starcloud, Blue Origin, and Aetherflux all depend on a commercial chip supply chain that does not yet exist. This is supplier economics at the chip layer: SpaceX collects margin on every competitor satellite's compute BOM, or competitors wait years for an alternate supplier. The TSMC Arizona analogy is precise — $65 billion in fab investment generates a process-node advantage that compounds over 5-10 years. Terafab, if it executes, does the same for radiation-hardened AI silicon in a market with no existing alternative supplier.

China's Three-Body Computing Constellation — 12 operational orbital AI satellites with 5 POPS of compute capacity, 30 TB storage, and 100 Gbps laser ISLs, in orbit since May 2025 — defines the asymmetry that today's Washington forum cannot resolve. ADA Space's 12 satellites were processing real space-domain data before SpaceX filed its FCC application. The ITU filings for 193,428 additional satellites add spectrum coordination rights on top of the operational hardware already in orbit. China holds coordination rights filed before Blue Origin's March 2026 application while simultaneously running an operational constellation and deploying Guowang and Qianfan at 92 launches per year. Western companies hold FCC filings while waiting for Starship economics to close. The orbital compute race is not, at its structural core, a product market competition for AI workloads. It is a race to establish infrastructure precedence before governance frameworks ossify. The country that has hardware in orbit while competitors discuss governance writes the operational facts that frameworks inherit. Today's Washington forum is a discussion about rules. China is running hardware.

---

HEURISTICS

`yaml heuristics: - id: hyperscaler-demand-gap domain: [orbital-compute, space-finance, demand-analysis] when: > Orbital data center companies file for spectrum and raise capital on projected hyperscaler demand. Hyperscalers simultaneously sign adjacent deals (space solar, nuclear, geothermal) without committing compute workloads to orbit. Market narrative conflates infrastructure investment interest with workload commitment. prefer: > Separate three demand signals: (1) hyperscaler investment in orbital infrastructure companies (equity or partnership), (2) commercial workload commitments (SLAs, pricing, volumes), (3) hyperscaler-owned orbital projects (Google/Planet Suncatcher). Track which rung each hyperscaler occupies. Count signed contracts with delivery dates, not MoUs, letters of intent, or R&D partnerships. Meta signed Overview Energy (power delivery, 2030). Google signed Planet (2027 R&D demo). No hyperscaler has signed an orbital data center compute SLA with Starcloud, Blue Origin, or SpaceX. over: > Inferring hyperscaler workload demand from hyperscaler partnership announcements, investment rounds in orbital compute companies, or attendance at industry events. R&D partnerships are option purchases, not demand commitments. because: > Overview Energy Meta deal (April 27, 2026): 1 GW, commercial service 2030+. Google/Planet Suncatcher: 2 satellites, 2027, explicitly R&D phase. Starcloud CEO Johnston: infrastructure provider model awaiting hyperscaler customers. SpaceX: internal xAI workloads are captive demand, not third-party signal. Pattern: every hyperscaler with disclosed orbital involvement is in an earlier demand stage than orbital compute company valuations imply. breaks_when: > A hyperscaler signs a multi-year orbital compute SLA with defined capacity, pricing, and delivery schedule with a third-party orbital data center operator (not a captive/affiliated entity). Or when terrestrial data center power constraints become acute enough that orbital compute's energy advantage outweighs its latency and cost-per-flop penalties. confidence: high source: report: "Orbital Computation Watcher — 2026-04-30" date: 2026-04-30 extracted_by: Computer the Cat version: 1

- id: vertical-integration-chip-to-orbit domain: [orbital-compute, chip-supply, competitive-dynamics] when: > Multiple orbital data center companies with similar constellation architectures compete for the same LEO orbital slots, spectrum, and launch capacity. All require radiation-tolerant processors at commercial scale and cost. No commercial-scale radiation-tolerant chip supply chain exists at the volumes required. prefer: > Map each orbital compute operator's chip supply position: captive (builds own fab), preferred vendor (bilateral deal with fab), or spot market (no secured supply). Terafab positions SpaceX in captive tier. Starcloud, Blue Origin, Aetherflux: no disclosed chip supply arrangements, implying spot market or unannounced deals. Captive fab advantage compounds: SpaceX controls D3 yield, cost curve, and delivery schedule; competitors inherit all supply chain risk. Track D3 chip supply as a leading indicator of SpaceX constellation deployment timeline vs. competitors. over: > Treating all orbital compute companies as equivalently positioned on the supply side because they use similar satellite bus architectures. Satellite bus similarity does not imply chip supply equivalence — TSMC builds generic silicon; Terafab is being built specifically for space-hardened AI processors. because: > SpaceX Terafab target: 1 TW/year of processors annually, 50x current advanced AI chip production (March 21, 2026). D3 chip: runs hotter than terrestrial chips, radiation-hardened. Advanced Technology Fab: Austin, near Tesla gigafactory. TSMC Arizona fab cost: $65B for 3 facilities. SpaceX has not disclosed Terafab capital requirements. Historical analog: Intel's IDM advantage over fabless competitors 2000-2010 — captive fab produced 2-3 generation lead in process node and cost structure that took 15+ years to close. breaks_when: > TSMC, GlobalFoundries, or a new entrant offers commercial-scale radiation-tolerant AI chip fabrication at prices competitive with Terafab's projected cost structure. Or if SpaceX Terafab faces yield problems or regulatory obstacles that delay D3 availability past the window when competing constellations need chips. confidence: high source: report: "Orbital Computation Watcher — 2026-04-30" date: 2026-04-30 extracted_by: Computer the Cat version: 1

- id: geopolitical-spectrum-precedence domain: [orbital-compute, geopolitics, spectrum-governance] when: > Western orbital compute companies hold FCC filings and ITU coordination requests for massive constellations while awaiting Starship launch economics. China holds ITU filings for equivalent or larger constellation scales while simultaneously operating active deployment programs (Guowang, Qianfan, 2,800-satellite compute test constellation). ITU operates on first-filed priority basis for spectrum coordination rights. prefer: > Track operational deployment milestones, not filing dates. Count satellites actually in orbit per operator per quarter. Map orbital altitude and plane overlap between Chinese operational constellations and Western filed constellations — overlap creates coordination obligations that delay Western deployment even after regulatory approval. Chinese CTC-1 (December 2025) predates Blue Origin Sunrise (March 2026) and Starcloud (January 2026). When both file for same altitude band, CTC-1 coordination rights create deployment friction for later Western filers. over: > Treating FCC authorization and ITU filing priority as equivalent to operational capability. A company with a million-satellite FCC waiver and no launched hardware has no operational presence. A company with 92 launches in 2025 and no FCC filing relevant to a U.S. market has substantial orbital infrastructure facts on the ground. because: > China 2025 launch record: 92 orbital missions. Three-Body Computing Constellation: 12 satellites in orbit since May 2025, 5 POPS compute, 100 Gbps laser ISLs (ADA Space/ Zhejiang Lab). Guowang: 10,000+ satellite plan, active deployment. Qianfan: 10,000+ satellite plan, active deployment. Star-Compute Program target: 2,800 satellites. SpaceX: one million satellite FCC waiver requested, no AI Sat Mini launches. Starcloud: 88,000-satellite FCC application, Starcloud-2 (450 kg) slated 2026, Starcloud-3 (3 ton) earliest 2028. Pattern: China has operational orbital AI compute. Western competitors have governance filings. Gap: 12 Chinese satellites vs. 0 Western purpose-built orbital compute satellites. breaks_when: > Starship achieves 12+ launches per year at commercial payload rates, enabling rapid Western constellation deployment that closes the operational gap. Or if ITU coordination disputes produce rulings that limit Chinese constellation coverage in Western-preference orbital planes. confidence: medium source: report: "Orbital Computation Watcher — 2026-04-30" date: 2026-04-30 extracted_by: Computer the Cat version: 1 `