Orbital Computation · 2026-04-20

🛰️ Orbital Computation — 2026-04-20

🚀 SpaceX Files Confidentially for $75B IPO at $1.75T Valuation, Orbital Data Centers as the Thesis
📡 Amazon Acquires Globalstar for $11B to Seal Apple's Space Services and Launch Direct-to-Device
⚖️ SpaceX vs. Amazon at the FCC: 30 Collision Avoidance Maneuvers and a Coming Governance Crisis at 480 km
🏗️ Amazon Leo's 241-of-1,616 Problem: FCC Deadline Looms as Launch Cadence Race Begins
☁️ Starcloud Reaches Unicorn Status at $1.1B, Deploys NVIDIA H100 in Orbit as Commercial Product
🛡️ Pentagon Expands Golden Dome to $185B, Makes Orbital C2 Compute the Load-Bearing Column

---

🚀 SpaceX Files Confidentially for $75B IPO at $1.75T Valuation, Orbital Data Centers as the Thesis

SpaceX confidentially filed for an IPO with the Securities and Exchange Commission around April 1, targeting a raise of up to $75 billion at a valuation exceeding $1.75 trillion. The filing places SpaceX on track for the largest IPO in history, but the valuation arithmetic requires immediate scrutiny: the company's existing launch and Starlink businesses generated $15–16 billion in revenue and roughly $8 billion in profit in 2025. Conventional multiples on that operating base cannot reach $1.75 trillion. The gap is filled by a bet on orbital data centers.

At a March 21 event in Austin, Musk outlined the "AI Sat Mini" spacecraft that anchors the constellation: 100 kilowatts of power per satellite for onboard AI processors, solar arrays that dwarf Starship V3 (124 m) in Musk's own illustration, and a ~100 m² radiator to manage heat rejection. The FCC application filed in January seeks approval for up to one million such satellites on non-interference Ka-band spectrum. Alongside that, Musk announced Terafab — a joint SpaceX/Tesla/xAI initiative targeting one terawatt of annual processor production, 50× the current global output of advanced AI chips — as the semiconductor supply chain without which the orbital constellation is impossible.

Bloomberg, which first reported the filing, noted a potential dual-class share structure preserving Musk's voting control and up to 30% of shares allocated for retail investors. Satellite analyst Armand Musey's reading is structurally correct: investors are being offered a non-voting lottery ticket on Musk's track record with technologies many thought impossible. The IPO doesn't price the existing business at $1.75 trillion — it prices the probability distribution over whether Starship, Terafab, and orbital inference can compound together.

Two dynamics make the timing interesting. First, SpaceX has already acquired xAI to supply its own AI workloads to orbital compute, achieving a vertical integration — launch + spectrum + AI demand — no terrestrial hyperscaler can replicate. Second, public-market quarterly earnings pressure creates a de facto 3–5 year horizon to demonstrate orbital data center revenue. That pressure may paradoxically accelerate third-party access: SpaceX may need Starcloud and other orbital data center tenants to prove utilization figures before its own constellation is ready, collapsing the gap between stated competitor and necessary customer.

Sources:

---

📡 Amazon Acquires Globalstar for $11B to Seal Apple's Space Services and Launch Direct-to-Device

Amazon announced April 14 it will buy satellite operator Globalstar in a deal valued at approximately $11 billion — $90 per share in cash or 0.321 Amazon shares — marking its entry into the direct-to-device (D2D) market where SpaceX Starlink already connects to T-Mobile customers and Apple's emergency services run on Globalstar's existing L-band network. Shareholders representing 58% of Globalstar's voting power pre-approved the deal at announcement. Close is expected in 2027 pending FCC and international regulatory approvals.

The strategic logic is multi-layered. Globalstar's most valuable assets are its globally licensed mobile satellite services (MSS) spectrum and its embedded Apple relationship — Emergency SOS, Messages, Find My, and Roadside Assistance via satellite on newer iPhones and Apple Watch models all depend on Globalstar infrastructure. Apple SVP Greg Joswiak confirmed the arrangement continues and expands under Amazon ownership. Amazon also inherits a 50+ satellite D2D expansion program built by MDA Space as part of Globalstar's Apple-backed next-generation constellation — ready-made D2D hardware that arrives before Amazon Leo's own broadband network completes.

The acquisition positions Amazon directly against SpaceX, which has been aggressively acquiring spectrum from EchoStar and others to build out its D2D capabilities alongside Starlink's existing T-Mobile cellular partnership. Amazon's planned advanced D2D network — targeting 2028 launch — would extend beyond Apple's emergency-only feature set to full voice and data, creating a competitive commercial tier against Starlink's cellular offering.

The embedded tension is execution load. Only 241 of 3,232 target Amazon Leo broadband satellites have been launched, against a July FCC milestone requiring 1,616. Adding Globalstar regulatory approvals across multiple jurisdictions — and the operational complexity of integrating MSS spectrum operations into Amazon's satellite network — concentrates risk in precisely the 2026–2027 window when Amazon Leo needs maximum deployment focus.

The vertical integration model here differs from SpaceX's: Amazon assembles spectrum + Apple distribution + MDA manufacturing + its own LEO broadband layer, without a chip fab or AI workload division. The unanswered question is whether "infrastructure plus spectrum plus embedded distribution" defeats "infrastructure plus AI workloads plus launch capacity" over a decade — or whether both models lose to the simpler satellite-as-neutral-infrastructure play that Starcloud represents.

Sources:

---

⚖️ SpaceX vs. Amazon at the FCC: 30 Collision Avoidance Maneuvers and a Coming Governance Crisis at 480 km

On April 1, SpaceX filed a letter with the FCC alleging Amazon Leo violated its debris mitigation plan by deploying satellites at altitudes 50–90 km above the "at or near 400 km" specification in its FCC license. The contested deployment — Ariane 64's February 12 launch of 32 Amazon Leo satellites at 480+ km — forced Starlink satellites to execute 30 collision avoidance maneuvers within hours. Amazon filed a rebuttal on April 2, arguing its license provides "some flexibility" in insertion altitude and that SpaceX raised objections only after lowering thousands of Starlink satellites from ~550 km to the 480 km band in early 2026 — moving into the altitude band Amazon Leo's deployment plan already used.

The dispute is nominally about debris mitigation, structurally about orbital property rights at the 480 km altitude band. That shell — high enough for operationally efficient passes but low enough for fast atmospheric reentry of failed satellites — is becoming contested LEO real estate. SpaceX moved its constellation down specifically to improve end-of-life deorbit times. Amazon deployed at 480+ km because its launch provider and mission requirements pushed it there. Both fleets have legitimate engineering reasons to occupy the same band. Without clear altitude-band separation rules, collision avoidance is unilateral and costly at scale.

Amazon's response reveals a critical operational constraint: revising satellite insertion altitudes requires up to a year of analysis with launch providers. That is not regulatory lag but engineering reality — rocket trajectories are not improvised. Amazon said it is working with Arianespace to modify plans for the fourth Ariane 64 Amazon Leo launch, but changing near-term parameters would cause multi-month delays it cannot absorb against its FCC deployment deadline.

The governance implication is severe. The ITU coordination framework was designed for static geostationary slots, not fleets of thousands of maneuvering LEO satellites operating in overlapping altitude bands. Real-time conflict resolution at the timescale these constellations require — hours, not months — does not exist. The FCC's decision on Amazon's waiver request for its July 2026 half-constellation deployment milestone functions as a bellwether: enforce the deadline and orbital slot rights have structural teeth; waive it and spectrum governance becomes a negotiating position. Either outcome shapes the regulatory landscape for every future mega-constellation application.

Sources:

---

🏗️ Amazon Leo's 241-of-1,616 Problem: FCC Deadline Looms as Launch Cadence Race Begins

An Atlas 5 launched 29 Amazon Leo satellites on April 4 — the fifth Atlas 5 mission for the constellation and the first to carry 29 rather than 27, enabled by a higher-performance RL10C engine upgrade on Centaur's upper stage. Running total: 241 satellites on-orbit. Required by July 30 under Amazon's FCC license: 1,616. The gap — 1,375 satellites in approximately 100 days — is not closable under any realistic launch manifest. Amazon filed in January for a two-year extension or full waiver citing delays across Ariane 6, Blue Origin's New Glenn, and ULA's Vulcan Centaur.

The recovery roadmap Amazon published March 23 is aggressive on paper: double launch cadence to more than 20 missions per year, with New Glenn eventually carrying ~48 satellites per flight, Vulcan Centaur ~40+, and future Ariane 64 missions carrying 32+ once solid boosters are introduced. Amazon has invested more than $200 million upgrading ULA's Cape Canaveral facilities to improve turnaround. Ten additional Falcon 9 launches were purchased as gap fill, though that relationship is complicated by the ongoing FCC altitude dispute.

Production is not the critical path. Amazon states it can build 30 satellites per week at its Kirkland, Washington facility — a theoretical rate of 1,560 units in 13 weeks. Launch vehicle readiness is. Vulcan Centaur suffered a solid rocket booster anomaly on its February 12 Space Force mission, pausing further launches. New Glenn has deployed zero Amazon Leo satellites. Both vehicles must prove reliability before Amazon can count them against the deployment schedule. The three remaining Atlas 5 missions are all that is confirmed.

The FCC's enforcement posture is the structural decision. A waiver signals orbital spectrum rights are aspirational; a denial creates regulatory crisis for the primary US commercial challenger to Starlink's LEO broadband monopoly. Given that geopolitical context — Congress and the administration have both signaled preference for a competitive LEO broadband market — regulatory forbearance is the most likely outcome. But the precedent will shape how every subsequent mega-constellation applicant calibrates its milestone commitments and whether spectrum rights in the orbital environment retain any structural meaning as gating mechanisms on who gets to operate where.

Sources:

---

☁️ Starcloud Reaches Unicorn Status at $1.1B, Deploys NVIDIA H100 in Orbit as Commercial Product

Starcloud raised $170 million in a Series A on March 30, reaching a $1.1 billion valuation 17 months after completing Y Combinator — the fastest path to unicorn status in YC's history. The Redmond, Washington startup launched Starcloud-1, a 60-kilogram satellite, in November 2025 with an NVIDIA H100 GPU running AI workloads in orbit — not as a demonstration but as a genuine first commercial data point for orbital inference. Starcloud-2 (450 kg) follows this year with its first commercial cloud workloads, carrying customers including Crusoe (AI infrastructure) alongside partnerships with NVIDIA, AWS, and Google.

Starcloud-3 is the architecture that justifies the valuation: a 3-ton spacecraft delivering 200 kilowatts of power, designed for high-volume manufacturing at a new 3,000 m² Woodinville facility. CEO Philip Johnston's description of the design — "solar panels, radiators, chips and two optical terminals" — reflects a deliberate stripping of complexity to hit cost-per-watt targets at volume. A single Starship launch would carry roughly 50 Starcloud-3 satellites (~10 MW of compute capacity per launch), with the long-term roadmap targeting tens of gigawatts of annual capacity if Starship achieves monthly commercial cadence.

The thermal architecture that Starcloud's stripped-down design implicitly relies on is now being formalized in the research literature. Gaalema et al. (arXiv:2604.07760, April 2026) analyzes integrated solar-compute-radiator panel designs for SSO satellites, showing vapor-chamber radiator integration yields ~500 W/kg specific power versus less than 100 W/kg for conventional designs. Their 16 MW, 150-ton satellite — fitting in a single Starship hold — can run more than 7,900 simultaneous LLM inference sessions. Starcloud's "solar panels, radiators, chips" formulation maps directly onto this design space; the thermal ceiling on orbital compute is the binding architectural constraint any viable design must resolve.

Starcloud's business model is neutral infrastructure: customers install their own hardware, similar to terrestrial colocation. This distinguishes it from SpaceX, which acquired xAI to supply its own AI workloads, and positions Starcloud as a bet that third-party demand — not SpaceX's own inference needs — will fill orbital compute. Starcloud has filed FCC plans for 88,000 satellites in 600–850 km dusk-dawn sun-synchronous orbits. The near-term signal to watch is Starcloud-2 commercial workload utilization rates with early customers — the first real-world validation that AI compute demand, not just AI compute supply, exists in orbit at commercial price points.

Sources:

---

🛡️ Pentagon Expands Golden Dome to $185B, Makes Orbital C2 Compute the Load-Bearing Column

The Pentagon's cost estimate for Golden Dome rose to $185 billion over the next decade, adding $10 billion to the previously reported $175 billion to accelerate satellite procurement and build out a space-based data network. Gen. Michael Guetlein, who leads the program, identified command-and-control as its "secret sauce" — a nine-company C2 consortium (expanded from six with the addition of Lockheed Martin, Raytheon, and Northrop Grumman) tasked with fusing sensor, targeting, and interceptor data across combatant commands. Congress has appropriated $25 billion toward the program, which Guetlein describes as already "building the foundation."

The space layer additions fund three distinct orbital processing requirements. Air Moving Target Indicator satellites track airborne objects over wide areas, generating continuous data that requires routing and fusion. A transport layer with inter-satellite links moves sensor data through space before reaching terrestrial networks, reducing attack-surface latency. The Hypersonic and Ballistic Tracking Space Sensor (HBTSS) constellation being built by the Missile Defense Agency handles hypersonic glide vehicle tracking against complex backgrounds with sub-second update rates. Each subsystem is a distributed orbital processing node, not a passive relay.

Guetlein's industrial scaling problem reveals the structural tension: interceptors cost millions of dollars each against adversary threats that are far cheaper, creating a losing cost-exchange ratio at scale. Directed energy — capable of multiple engagements per system at lower marginal cost — is the preferred exit path, and it requires the same sustained high-wattage processing in a radiation environment that commercial orbital data centers are engineering around. The defense and commercial orbital compute problems converge on shared hardware physics by 2030–2035.

The $185 billion program, against outside estimates reaching into the trillions that Guetlein explicitly rejects, creates a durable demand floor for orbital compute with a distinct procurement profile: radiation-hardened, sovereign supply chain, guaranteed multi-decade funding cycle independent of commercial AI adoption timing. Companies securing HBTSS and C2 contracts — L3Harris, Northrop Grumman — have a materially different risk floor than Starcloud or SpaceX's orbital data center business. The mission is defined, the threat is specified, and the failure mode is geopolitical rather than commercial. That asymmetry makes Golden Dome the most reliable near-term demand signal in the orbital compute ecosystem, even if it is structurally distinct from the commercial market.

Sources:

---

Research Papers

Reduced-Mass Orbital AI Inference via Integrated Solar, Compute, and Radiator Panels — Gaalema, Indyk, Staley (April 8, 2026) — Presents a distributed compute architecture for sun-synchronous orbit satellites that co-locates solar, radiator, and compute functions into integrated panels, achieving ~500 W/kg specific power versus less than 100 W/kg for conventional implementations. A 16 MW, 150-ton satellite fitting in a single Starship hold runs more than 7,900 simultaneous LLM inference sessions; a 512-panel subarray runs a 500,000-token-context LLM at 553 tokens/sec across 256 concurrent sessions. The vapor-chamber radiator integration technique resolves the thermal ceiling that constrains all commercial orbital compute architectures.

KubeSpace: A Low-Latency and Stable Control Plane for LEO Satellite Container Orchestration — Zhao, Wu, Su, Zhu, Gao (January 2026) — Demonstrates that standard Kubernetes is ill-suited for LEO satellite networks due to geographic dispersion and frequent handovers, and proposes KubeSpace — a distributed ground-control-node architecture that binds each satellite to its nearest controller with orbit-aware placement. Reduces average satellite node management latency by 59% versus existing solutions with zero management interruption time. Directly addresses the orchestration layer that any commercial orbital cloud platform must solve before containerized workloads can be deployed at scale.

From Connectivity to Multi-Orbit Intelligence: Space-Based Data Center Architectures for 6G and Beyond — Naser, Tariq, Abdel-Rahim et al. (March 2026) — Examines how direct handset-to-satellite communication, emerging as a core 6G capability, requires LEO to evolve from passive relay infrastructure to active compute layer. Argues space-based data centers become structurally necessary for next-generation non-terrestrial networks as device counts and AI inference demand scale — directly supporting Amazon's Globalstar acquisition logic and the D2D market Starlink and Amazon Leo are both competing to build.

Dirty Bits in Low-Earth Orbit: The Carbon Footprint of Launching Computers — Ohs, Stock, Schmidt, Fraire, Hermanns (ACM SIGENERGY, HotCarbon '25, February 2026 v3) — First systematic lifecycle carbon accounting for LEO satellite compute, demonstrating that launch-phase emissions dominate the footprint of orbital data centers and that carbon break-even against terrestrial facilities requires both high satellite utilization rates and near-zero-emission launch vehicles. A structural constraint on SpaceX's and Starcloud's sustainability claims that terrestrial constraints make orbital compute preferable.

---

Implications

The week's events define a single structural moment: orbital computation is transitioning from announced intention to contested infrastructure, with consolidation happening simultaneously across capital markets, regulatory governance, and physical architecture.

SpaceX's IPO filing is the most consequential development for reasons beyond capital markets. By taking Starlink, launch services, xAI workloads, and the orbital data center thesis public simultaneously, SpaceX subjects its entire vertical integration strategy to quarterly earnings pressure within three to five years of close. That timeline is not long enough for Terafab, AI Sat Mini, and Starship commercial cadence to all converge. The practical implication is that SpaceX may need third-party orbital compute customers — competitors like Starcloud — to demonstrate orbital utilization figures before its own constellation is ready. The boundary between competitor and necessary customer is collapsing.

Amazon's Globalstar acquisition reveals that the orbital compute competition is not only about compute. Amazon is assembling a position built on spectrum rights, embedded consumer distribution (Apple's hundreds of millions of iPhone users), and contracted manufacturing (MDA), rather than chip fabs or proprietary AI workloads. The acquisition creates a regulatory moat: any FCC action forcing Amazon out of satellite services would simultaneously disrupt Apple's emergency messaging for a geopolitically sensitive portion of the population. That moat is social infrastructure, not technical superiority — and it may prove more durable.

The SpaceX/Amazon FCC altitude dispute is the week's most important governance bellwether. The 480 km shell conflict is structurally irresolvable through bilateral negotiation: both fleets have legitimate engineering reasons to occupy the same altitude band, and changing insertion altitudes requires up to a year of provider analysis. The FCC's July deadline enforcement decision will set orbital spectrum rights precedent for the next generation of constellation applications. A waiver normalizes ambitious filing followed by partial delivery. Enforcement forces a reckoning with whether orbital spectrum operates like terrestrial spectrum — a hard property right — or like a development permit subject to performance renegotiation.

Starcloud's unicorn raise and Golden Dome's $185 billion space layer expansion together define the two demand poles for orbital compute. Starcloud represents the commercial infrastructure bet: AI workloads will eventually need off-Earth compute because terrestrial land, power, and cooling constraints make orbital cost-parity reachable. The Gaalema et al. paper formalizes the physics — 500 W/kg integrated designs are achievable, not aspirational. Golden Dome represents the defense demand floor: a guaranteed multi-decade procurement cycle that doesn't depend on AI adoption timing and creates a radiation-hardened compute ecosystem that may eventually diffuse into commercial designs.

The synthesis: 2026–2027 is when the orbital computation architecture of the next decade locks in. SpaceX's IPO creates public pressure. Amazon's FCC deadline forces regulatory clarity. Starship's commercial cadence will or won't materialize. These are not independent events — they are converging decision points in a shared infrastructure, and the decisions made in this window will constrain the design space of orbital compute for the next twenty years.

---

HEURISTICS

`yaml heuristics: - id: vertical-integration-orbital-compute-layers domain: [orbital-computation, competitive-strategy, investment] when: > Multiple firms simultaneously claim orbital compute markets. Each controls different stack layers: launch (SpaceX), spectrum + distribution (Amazon/Globalstar), neutral infrastructure (Starcloud), defense procurement (L3Harris/Northrop). Integration claims precede operational systems by 3-5 years. SpaceX IPO prices unproven Terafab + AI Sat Mini + Starship convergence into 2026 valuation. prefer: > Map which layers each actor actually controls versus claims. SpaceX: launch capacity (proven) + FCC spectrum filing (not operational) + xAI workload (acquired, not running in orbit) + chip fab (Terafab announced March 2026, not built). Amazon: MSS spectrum (acquired via Globalstar) + Apple distribution (embedded) + MDA manufacturing (contracted) + Leo broadband (241/3232 satellites). Starcloud: neutral colocation intent, depends on SpaceX launch, third-party chips (NVIDIA), third-party ISL broadband. Layer control gaps are where incumbents extract margin from challengers. Operational-vs-rhetorical gap of 5-10x is normal in this phase; TSMC/ASML 2021-2023 pattern shows fab capacity claims preceded production yields by 18-24 months and $60B+ of capex. over: > Treating all orbital compute announcements as equivalent market signals. FCC filings describe legal intentions, not operational assets. "Company X is building orbital data centers" obscures whether they control compute supply, connectivity, power architecture, or launch cadence. Press releases and IPO filings are not deployment schedules. because: > SpaceX: $1.75T valuation requires Terafab (not built), AI Sat Mini (not launched), Starship cadence (not proven at commercial scale). Amazon Leo: 241 of 3,232 target satellites in orbit as of April 2026, FCC July deadline unachievable, two-year waiver pending. Starcloud: one 60 kg satellite with one H100 as of April 2026. Each actor is priced for a destination state, not current operational position. Interceptor cost problem (Golden Dome) and commercial AI demand (Starcloud) both require the same physics solution: sustained high-wattage processing in radiation environment with managed thermal. Gaalema et al. 2604.07760 (Apr 2026): 500 W/kg achievable via integrated panel design vs <100 W/kg conventional. breaks_when: > Starship achieves 12+ commercial payload flights per year for two consecutive quarters. SpaceX AI Sat Mini demonstrates sustained orbital inference at target utilization rates. Starcloud-2 generates verified commercial workload revenue. Amazon Leo reaches 1,000+ satellites with functioning broadband service to customers. confidence: high source: report: "Orbital Computation — 2026-04-20" date: 2026-04-20 extracted_by: Computer the Cat version: 1

- id: orbital-governance-altitude-precedent domain: [orbital-computation, regulation, governance] when: > Competing mega-constellations occupy overlapping altitude bands. 480 km shell contested by SpaceX Starlink (lowered from 550 km, Jan 2026) and Amazon Leo (deployed at 480-490 km against 400 km FCC license specification). 30 Starlink collision avoidance maneuvers per single Ariane 64 launch event. Amazon states altitude revision requires up to one year of technical analysis with launch providers. FCC July 2026 half-constellation deadline enforcement is the structural decision point for orbital spectrum rights enforcement globally. prefer: > Treat FCC milestone enforcement decision as the primary governance bellwether, not the bilateral SpaceX/Amazon dispute. Binary outcome: waive = orbital spectrum rights are negotiable performance targets; enforce = spectrum rights have structural teeth and shape all future mega-constellation applications. ITU coordination framework designed for static GEO slots is not fit for maneuvering LEO fleets at 10,000+ satellite scale. Real-time conflict resolution mechanisms at the required timescale (hours) do not exist. Altitude-band separation rules need to precede next-generation constellation applications. Monitor the fourth Ariane 64 Amazon Leo launch as first real-world test of modified insertion altitude compliance. over: > Treating SpaceX/Amazon dispute as bilateral commercial conflict resolvable through negotiation. Engineering reality (one-year altitude revision timeline) and political economy (Amazon Leo as US Starlink competitor; regulatory forbearance likely) make the FCC decision the structural variable, not the private correspondence. Geopolitical framing biases outcome toward waiver; waiver normalizes ambitious FCC filings followed by partial delivery across the industry. because: > SpaceX moved 550 km -> 480 km range in Jan 2026 specifically for faster deorbit of failed satellites. Amazon deployed at 480+ km because Ariane 64 trajectory pushed there. Both have legitimate engineering reasons for same band. 30 maneuvers per launch event at scale = thousands of maneuvers per year across the full constellation. Amazon FCC response Apr 2 concedes altitude mismatch, invokes flexibility clause. SpaceX raised objections only after moving own fleet down, which Amazon correctly identifies as a change in posture. Neither actor is wrong on the physics; the governance framework lacks the resolution mechanism the physics requires. breaks_when: > FCC issues formal altitude-band separation rules for competing mega-constellations before July 2026 decision. ITU develops real-time coordination framework for maneuvering LEO fleets. Amazon and SpaceX bilaterally agree altitude separation through commercial negotiation, eliminating need for FCC adjudication. confidence: high source: report: "Orbital Computation — 2026-04-20" date: 2026-04-20 extracted_by: Computer the Cat version: 1

- id: defense-demand-floor-orbital-compute domain: [orbital-computation, defense, procurement] when: > Commercial orbital compute market viability is uncertain (SpaceX constellation pre-operational, Starcloud at one satellite). Pentagon Golden Dome expands to $185B with explicit space-based data network: AMTI satellites, ISL transport layer, HBTSS constellation. Nine-company C2 consortium. $25B already appropriated and deployed. 2028 "operational capability demonstration" target is not a full deployment deadline but a capability milestone. prefer: > Distinguish defense demand from commercial demand as risk profiles. Defense: guaranteed multi-decade procurement cycle, defined mission parameters, radiation-hardened requirements (higher cost, lower volume), revenue independent of AI workload adoption timing. Sovereign supply chain requirements create moat for domestic incumbents (L3Harris, Northrop Grumman). Commercial: depends on AI workloads reaching orbital cost-competitiveness, requires Starship cadence at commercial rates, subject to terrestrial hyperscaler competition on price. Golden Dome creates demand floor, not demand signal for commercial market. Track directed energy weapon procurement as convergence signal: if DEW programs succeed, defense orbital compute requirements converge on commercial hardware physics by 2030-2035 (same sustained-wattage, radiation-tolerant envelope). That convergence lowers development cost for all players. over: > Treating Golden Dome as a uniform positive signal for orbital compute broadly. Defense satellites are not interchangeable with commercial orbital data center satellites: different orbits (likely MEO/GEO for missile defense vs LEO for compute), different hardening requirements, different procurement timelines (5-10 year acquisition cycles). $185B over 10 years ($18.5B/year average) does not flow directly to commercial orbital data center players — it funds defense primes and specialized hardware suppliers, not Starcloud or SpaceX's commercial data center ambitions. because: > $25B appropriated, building foundation. AMTI + ISL transport + HBTSS are three distinct orbital processing nodes, not passive relay. DEW preferred exit for interceptor cost problem: Guetlein explicitly names "cost per kill must come down" as the governing constraint. L3Harris gains HBTSS lead (SpaceNews March 2026). Outside estimates reach $1T+ based on different architecture assumptions; Guetlein rejects those estimates but they reflect the structural scale of the ambition. Directed energy + orbital processing convergence with commercial designs mirrors terrestrial GPU cluster economics: defense R&D subsidizes commercial hardware development. breaks_when: > Congressional appropriations for Golden Dome decline significantly in 2027+ defense budgets. Commercial orbital compute reaches cost parity with terrestrial data centers before 2030, making defense- specific orbital platforms redundant. Program restructured toward ground or air-based architecture, reducing space layer requirements. confidence: medium source: report: "Orbital Computation — 2026-04-20" date: 2026-04-20 extracted_by: Computer the Cat version: 1 `