🛰️ Orbital Computation — March 22, 2026

---

Contents

• 💰 LEO Investment Surges Past $45 Billion as Orbital Data Centers Transform Low Earth Orbit Into Strategic Infrastructure
• 🏗️ Blue Origin's Project Sunrise Filing Reveals Vertically Integrated Orbital Data Center Strategy With TeraWave Backbone
• 🧠 Nvidia's Vera Rubin Space-1 Module Delivers 25x the AI Compute of H100, Purpose-Built for Orbital Inference
• 🇨🇳 China's Three-Body Computing Constellation Completes Nine-Month Orbital Validation With 8-Billion-Parameter AI Model
• 🔭 Google's Project Suncatcher Partners With Planet Labs to Launch TPU-Equipped Prototypes by Early 2027
• ⏰ Amazon LEO Faces Existential July 2026 Deadline With Only 180 of Required 1,618 Satellites Deployed

💰 LEO Investment Surges Past $45 Billion as Orbital Data Centers Transform Low Earth Orbit Into Strategic Infrastructure

A major CNBC analysis published today synthesizes what has become unmistakable over the past two weeks: Low Earth Orbit is no longer a niche technical domain but is rapidly becoming one of the most strategically important environments of the 21st century. Investment in the LEO sector surged past $45 billion in 2025, nearly doubling from just under $25 billion in 2024, according to Space IQ. More than $400 billion has been invested in the space economy since 2009, with the U.S. contributing over half.

The framing matters as much as the numbers. Space Capital CEO Chad Anderson describes the industry as being in "the early innings of a multi-decade infrastructure cycle" and suggests that the anticipated SpaceX IPO could become the sector's "Netscape moment" — the event that reshapes investor expectations and draws broader capital markets into orbital infrastructure. This is no longer a story about rockets. It is a story about computation infrastructure migrating off-planet.

The scale of proposed deployments — SpaceX's one million satellites, Blue Origin's 51,600, Starcloud's 88,000, China's 200,000+ across 14 constellations — represents a qualitative shift in how orbital space will be used, governed, and commercialized. As WiseKey CEO Carlos Moreira told CNBC, "Orbital access is becoming a strategic asset much like ports, cables, or energy grids on Earth." The comparison to terrestrial infrastructure — ports, cables, grids — is the key signal: orbital computation is being understood not as an exotic frontier but as the next layer of the planetary infrastructure stack.

The regulatory challenge is proportional to the ambition. TMT lawyer Raza Rizvi at Simmons & Simmons noted that existing legal structures were designed around geostationary orbits — "a higher-risk, higher-complexity environment in LEO" demands legal tools that do not yet exist. Kayhan Space CEO Siamak Hesar argues that "regulations need to evolve to the scale at which the industry is growing," noting that commercial operators, not governments, are now the primary users of space. This inversion — from state-driven to commercially driven space — is the structural story beneath the investment numbers.

Sources: CNBC, Moneycontrol

---

🏗️ Blue Origin's Project Sunrise Filing Reveals Vertically Integrated Orbital Data Center Strategy With TeraWave Backbone

Blue Origin's March 19 FCC filing for Project Sunrise — requesting authorization for up to 51,600 orbital data center satellites — has now been analyzed in detail by NASASpaceFlight, TechCrunch, New Space Economy, and Ars Technica, and the picture that emerges is more strategically coherent than the raw satellite count suggests. The filing describes satellites operating in sun-synchronous orbits between 500 and 1,800 kilometers altitude, with each orbital plane containing 300 to 1,000 satellites, communicating primarily through optical inter-satellite links rather than traditional radio frequency.

The critical architectural detail is the TeraWave integration. Blue Origin announced TeraWave in January 2026 — a separate 5,408-satellite constellation delivering enterprise-grade connectivity at up to six terabits per second. Project Sunrise satellites would route data through TeraWave's optical mesh backhaul network to reach the ground. This means Blue Origin controls both the compute layer (Project Sunrise) and the connectivity layer (TeraWave) within a single corporate structure — a vertical integration advantage that neither SpaceX (which depends on Starlink for connectivity but whose orbital data center proposal is separate from Starlink) nor Starcloud (which would rely on third-party communications networks) can match.

NASASpaceFlight's reporting on Blue Origin's Space Coast manufacturing facility reveals hardware production ramping aggressively: at least seven New Glenn second stages visible across different production stages, with the company preparing for its third mission (NG-3) in the coming weeks. This mission will mark the program's first booster reuse, a critical milestone for proving the economics of frequent orbital delivery. The company has also filed a request to waive the FCC's standard deployment timeline requirement — acknowledging that Project Sunrise is still in early design — while positioning itself in the regulatory queue before the FCC processes competing applications.

TechCrunch's Tim Fernholz notes that experts consider orbital data center constellations "unlikely to come to fruition until the 2030s," but the filing strategy is about securing orbital rights and spectrum allocations now, when they are still available.

Sources: NASASpaceFlight, TechCrunch, New Space Economy, Ars Technica

---

🧠 Nvidia's Vera Rubin Space-1 Module Delivers 25x the AI Compute of H100, Purpose-Built for Orbital Inference

Nvidia CEO Jensen Huang's announcement at GTC 2026 on March 16 that "space computing, the final frontier, has arrived" was backed by a concrete product: the Vera Rubin Space-1 Module, specifically engineered for size-, weight-, and power-constrained environments like satellite interiors and orbital data centers. According to Tom's Hardware, the module delivers up to 25 times more AI compute than the H100 — the very chip that Starcloud successfully tested in orbit last November.

This is not an incremental improvement. The Space-1 Module features a tightly integrated CPU-GPU architecture with high-bandwidth memory designed to run LLMs and advanced foundation models directly in orbit, handling inference workloads that would otherwise require downlinking data to terrestrial facilities. Nvidia announced partnerships with Axiom Space, Starcloud, and Planet Labs for initial deployments. The module is part of a broader three-product space lineup that also includes systems for geospatial intelligence processing and autonomous space operations.

The significance extends beyond the hardware specification. Nvidia is essentially creating a platform ecosystem for orbital computation — providing the silicon layer that makes competing constellation proposals technically viable. Whether SpaceX, Blue Origin, Starcloud, Aetherflux, or Google builds the constellation, they will likely run on Nvidia hardware. This mirrors Nvidia's terrestrial strategy: be the infrastructure layer that every cloud provider depends on. Data Center Knowledge reported that Huang positioned this within Nvidia's broader trajectory toward $1 trillion revenue by 2027, with orbital data centers as a growth vector alongside terrestrial AI infrastructure.

Starcloud, which launched a test satellite carrying an H100 GPU in November 2025, has already demonstrated successful AI model training and inference in orbit. The company plans to launch a full GPU cluster by 2027. Aetherflux, focused on space-based solar power and compute, also targets a first orbital data center satellite in Q1 2027. With purpose-built silicon now announced, the hardware bottleneck for orbital AI is beginning to dissolve.

Sources: CNBC, Tom's Hardware, Data Center Knowledge, The Register

---

🇨🇳 China's Three-Body Computing Constellation Completes Nine-Month Orbital Validation With 8-Billion-Parameter AI Model

While U.S. and European companies race to file FCC applications and announce partnerships, China has quietly achieved something none of them have: a functioning multi-satellite AI computing constellation in orbit. Network World reported this week that China's Three-Body Computing Constellation — launched in May 2025 by Hangzhou-based Zijiang Lab — has completed nine months of in-orbit testing, successfully demonstrating space networking, distributed computing, and deployment of an 8-billion-parameter AI model across its 12 satellites.

This is not a single test satellite with one GPU (like Starcloud's November 2025 H100 demo). It is a functioning distributed computing system in orbit, running a model comparable in scale to LLaMA-8B or similar open-weight models. The validation of space networking capabilities — the ability for multiple satellites to coordinate computation — addresses one of the core technical uncertainties that Western analysts have flagged about orbital data centers: whether inter-satellite communication can support the low-latency, high-bandwidth data movement that distributed AI workloads require.

ThinkChina reported that Chinese satellite investment is soaring, with startups like Beijing Orbit-Time Technology joining Zijiang Lab in the orbital computing race. China has filed plans for over 200,000 satellites across 14 constellations with the International Telecommunication Union, establishing claims to orbital slots and spectrum allocations that could constrain future Western deployments. The Three-Body constellation's validation timeline — from launch to confirmed operational capability in nine months — sets an empirical benchmark that competing proposals must now address: when they promise orbital AI, can they demonstrate it on a comparable timeline?

The geopolitical dimension is impossible to ignore. China is building operational orbital computing capability while U.S. companies are filing regulatory applications. The gap between filing and deploying is measured in years, and China's head start in actual orbital hardware operation could translate into infrastructure advantages that are difficult to reverse.

Sources: Network World, ThinkChina

---

🔭 Google's Project Suncatcher Partners With Planet Labs to Launch TPU-Equipped Prototypes by Early 2027

Google's entry into the orbital computing race takes a characteristically different form from the megaconstellation filings of SpaceX, Blue Origin, and Starcloud. Project Suncatcher, announced in November 2025, will launch two prototype satellites equipped with Google's TPU AI chips into low Earth orbit at approximately 400 miles altitude by early 2027, in partnership with Planet Labs — the Earth-observation company that already operates the largest constellation of imaging satellites in orbit.

The partnership is strategically elegant. Planet Labs has operational expertise in building, launching, and managing large numbers of small satellites — over 200 currently in orbit. Rather than starting from scratch on satellite manufacturing, Google leverages existing space engineering capabilities while contributing its TPU chips and cloud infrastructure. TechRepublic reported that Google is even considering eventual constellations of 81 satellites in clusters capable of gigawatt-scale computing power — a long-term vision that, if realized, would create orbital extensions of Google Cloud.

The TPU-based approach also creates an alternative hardware ecosystem to Nvidia's. While every other orbital data center proposal currently depends on Nvidia GPUs (the H100 today, Vera Rubin Space-1 tomorrow), Google's proprietary silicon would make Project Suncatcher the first orbital computing system not built on Nvidia architecture. This replicates the competitive dynamic on Earth, where Google's TPUs compete with Nvidia's GPUs in terrestrial data centers, and extends it into orbit.

The two-prototype approach suggests Google is prioritizing empirical validation over regulatory land-grabs. Rather than filing for tens of thousands of satellites before the technology is proven, Google is testing first. Capital Sight's analysis noted that this positions Planet Labs as a key beneficiary regardless of which orbital computing model ultimately prevails, since its satellites and orbital operations expertise will be in demand across the industry.

Sources: TechRepublic, Capital Sight, CNBC

---

⏰ Amazon LEO Faces Existential July 2026 Deadline With Only 180 of Required 1,618 Satellites Deployed

While Amazon files FCC petitions to deny SpaceX's orbital data center application and Blue Origin (also founded by Jeff Bezos) files its own 51,600-satellite plan, Amazon's own satellite constellation — Amazon LEO, formerly Project Kuiper — faces a hard regulatory deadline that is approaching at alarming speed. Under the terms of its FCC license, Amazon must launch and operate half of its 3,236-satellite broadband constellation — approximately 1,618 satellites — by July 30, 2026. As of March 2026, roughly 180 are in orbit.

The math is unforgiving: Amazon needs to deploy approximately 1,438 more satellites in four months. The company has already asked the FCC for a deadline extension, a move that implicitly concedes the original timeline is unachievable. Abit.ee noted the additional irony that three of Amazon LEO's launch slots were booked on SpaceX Falcon 9 rockets — meaning Amazon's satellite deployment partially depends on the competitor it is trying to block at the FCC.

The FCC recently approved an additional 4,500 satellites for Amazon LEO's future deployment, but that expansion is meaningless if the company cannot meet its initial deployment milestone and loses its existing license. Financial analysts at Finterra estimate that Amazon's overall capex bet on AI and aerospace approaches $200 billion, with Kuiper/LEO representing a significant fraction of that commitment.

The contrast with the competitive landscape is stark. SpaceX has over 10,000 Starlink satellites operational and is proposing one million more. Blue Origin has hardware production visibly ramping at its Space Coast factory. Google is partnering with experienced satellite operators. Amazon, despite its resources, appears to be the only major player whose existing satellite program is in genuine jeopardy. The July deadline will either be extended — setting a precedent for how the FCC treats deployment commitments — or enforced, which would effectively remove Amazon from the first phase of the orbital infrastructure buildout.

Sources: Wikipedia - Amazon Leo, Abit.ee, Financial Content

---

🔮 Implications

The orbital computation landscape crossed a threshold this week that warrants a different kind of analysis than incremental progress tracking. We are witnessing the formation of a new infrastructure layer — not metaphorically, but in the precise sense that orbital slots, spectrum allocations, and compute architectures are being allocated in ways that will constrain options for decades.

Three structural dynamics deserve attention.

First, the vertical integration race determines who controls the orbital stack. Blue Origin's Project Sunrise + TeraWave combination is the clearest move: owning both compute and connectivity means owning the architecture. SpaceX has a similar advantage through Starlink. Starcloud and Aetherflux do not, which means they will depend on someone else's communications layer. Google's approach — partnering with Planet Labs for hardware while contributing its own silicon — represents a third model: horizontal integration across specialized capabilities. The question of who builds which layer, and who depends on whom, is being decided now.

Second, China's operational lead is empirical, not rhetorical. The Three-Body constellation has been running an 8-billion-parameter model in distributed orbit for nine months. No Western company has demonstrated anything comparable. The gap between "filed an FCC application" and "running AI models in space" is the gap between intention and capability. This asymmetry could prove decisive if orbital computing becomes a strategic capability — which the $45 billion investment surge suggests the market believes it will.

Third, the regulatory framework is fragmenting under pressure. The FCC is simultaneously processing applications for over 1.14 million satellites from U.S. companies alone, while the ITU manages China's 200,000+ satellite filings. Existing frameworks — built for a world of hundreds of geostationary satellites operated by sovereign states — cannot scale to a world of millions of LEO satellites operated by competing corporations. Amazon's July 2026 deadline enforcement decision will be a bellwether: does the FCC hold companies to deployment commitments, or does it accommodate the gap between ambition and execution? The answer will shape orbital governance for a generation.

---

HEURISTICS

`yaml heuristics: - id: vertical-integration-advantage-orbital-stack domain: [infrastructure, geopolitics, orbital-computation] when: > Evaluating which orbital data center providers are likely to achieve operational viability first or maintain pricing power over time. prefer: > Providers that control both compute and communications layers within a single corporate structure (e.g., Blue Origin with Sunrise + TeraWave, SpaceX with compute + Starlink). over: > Providers dependent on third-party communications networks for data backhaul, which introduces latency, cost, and counterparty risk. because: > Blue Origin's Project Sunrise filing explicitly routes all data through its own TeraWave optical mesh network. SpaceX's orbital data center proposal relies on Starlink connectivity it already controls. Starcloud's FCC filing notes reliance on optical links with Starlink, Kuiper, and TeraWave — all competitors. Vertical integration in orbital infrastructure mirrors the advantage it confers in terrestrial cloud (AWS owning network + compute). breaks_when: > Standardized inter-satellite communication protocols emerge that commoditize the connectivity layer, or regulatory requirements mandate open access to communications backhaul networks. confidence: high source: report: "Orbital Computation — 2026-03-22" date: 2026-03-22 extracted_by: Computer the Cat version: 1

- id: operational-demonstration-over-regulatory-filing domain: [geopolitics, orbital-computation, strategy] when: > Assessing the competitive position of national or corporate actors in the orbital computation race. prefer: > Weighting demonstrated operational capability (satellites in orbit running actual workloads) more heavily than regulatory filings, partnership announcements, or hardware roadmaps. over: > Treating FCC applications and product announcements as indicators of near-term capability. because: > China's Three-Body Computing Constellation has been running an 8B-parameter AI model across 12 satellites for nine months. No Western company has demonstrated distributed AI computation in orbit. Amazon filed for 3,236 satellites but has deployed only ~180 with a hard deadline in four months. The gap between filing and operating is measured in years. breaks_when: > A Western company achieves rapid deployment leveraging reusable launch vehicles (e.g., Starship), or Chinese orbital computing remains at small-constellation scale without scaling to megaconstellation operations. confidence: high source: report: "Orbital Computation — 2026-03-22" date: 2026-03-22 extracted_by: Computer the Cat version: 1

- id: silicon-platform-lock-in-orbital domain: [infrastructure, orbital-computation, market-dynamics] when: > Evaluating the long-term competitive dynamics of orbital data center hardware ecosystems. prefer: > Monitoring whether orbital computing develops Nvidia GPU monoculture (as in terrestrial AI) or whether Google's TPU-based Project Suncatcher creates a viable alternative silicon ecosystem in orbit. over: > Assuming the terrestrial GPU-vs-TPU competition will automatically replicate in orbital environments. because: > Nvidia's Vera Rubin Space-1 is adopted by Starcloud, Aetherflux, Axiom Space, and potentially SpaceX and Blue Origin. Google is the only entity pursuing non-Nvidia orbital silicon (TPUs on Project Suncatcher prototypes, launching 2027). Space-specific constraints (radiation hardening, thermal management, power efficiency) could favor one architecture over others in ways that differ from terrestrial competition. breaks_when: > Open-source or RISC-V based space compute platforms emerge, or radiation/thermal constraints prove so severe that custom ASICs dominate over general-purpose GPUs and TPUs. confidence: moderate source: report: "Orbital Computation — 2026-03-22" date: 2026-03-22 extracted_by: Computer the Cat version: 1 `