Introduction: The Dawn of Orbital Intelligence
On June 12, 2026 — the very day this paper is published — SpaceX (NASDAQ: SPCX) made its historic debut on the Nasdaq stock exchange, raising $75 billion at a valuation that opened above $1.77 trillion and soared nearly thirty percent in intraday trading, making it the single largest initial public offering in the history of financial markets, surpassing Saudi Aramco’s record $29 billion raise in 2019.
Elon Musk, whose combined stakes in SpaceX and Tesla now value his portfolio at approximately $1.147 trillion, became the world’s first verified trillionaire on the day of the listing.[1] On a JPMorgan Chase livestream before the opening bell, Musk declared that SpaceX had been cash-flow positive since roughly 2015 and that he wanted to take the company public to raise capital for “a significant growth phase,” which he described as including more than 100,000 satellites in orbit and the construction of artificial intelligence data centers in space.[2]
It is fitting, then, that a paper on Orbital Intelligence Networks should be written on precisely this day. The SpaceX IPO is not only a capital markets event; it is a civilizational marker. It signals that the satellite industry has crossed a threshold from infrastructure-as-coverage to infrastructure-as-intelligence. Starlink’s connectivity unit generated $11.39 billion in 2025, representing 61% of total SpaceX revenue — and climbing to 69% of total sales in the first quarter of 2026 alone — demonstrating that recurring revenue from orbital assets is now a verifiable economic reality at scale.
This paper arrives at a propitious moment for a second reason. The great challenge of the terrestrial AI era is energy. The International Energy Agency reported in its April 2026 update that the largest technology companies exceeded $400 billion in data center capital expenditure in 2025, and that this figure is expected to jump by another 75% in 2026 alone. Electricity consumption from AI-focused data centers grew 50% in 2025.[3] In a prior IEA analysis, projections showed that energy demands from data centers and AI could more than double from 460 terawatt-hours in 2022 to more than 1,000 TWh by 2026 — “roughly equivalent to the energy consumption of Japan,” as the agency stated.[4]
Against that backdrop, moving compute to orbit is no longer a speculative fantasy. Space-based platforms powered by near-continuous solar energy, radiating waste heat passively into the vacuum, and freed from land constraints, permitting delays, and freshwater cooling needs, represent a logically coherent response to a genuine planetary resource constraint. Whether the economics will ultimately work at scale remains contested — but the engineering demonstrations are now real, the filings are now before regulators, and the capital is now committed.
This paper introduces the conceptual framework of Orbital Intelligence Networks and situates it in the current competitive, regulatory, and technical landscape. It is organized in seven sections: the evolution from communication to intelligence (Section 1); the three-layer architecture of space, air, and ground (Section 2); why decentralized in-orbit AI compute matters (Section 3); the challenges and risks (Section 4); the strategic opportunities (Section 5); the competitive landscape of established players and newcomers as of mid-2026 (Section 6); and the seven pillars that synthesize the lessons of this emerging field (Section 7).
Section 1: The Evolution from Space Communications to Space-Based Intelligence
1.1 Three Eras of Satellite Infrastructure
The satellite industry has unfolded in three discernible eras, each defined by the dominant value proposition satellites delivered to human civilization. Understanding the logic of each transition is essential for grasping why the third era — intelligence — is not merely an incremental upgrade but a structural reorganization of what satellites are for.
The first era, which spans roughly from Sputnik in 1957 through the early years of commercial Earth observation in the 1990s, was defined by observation. Satellites looked downward. Their value lay in seeing — in capturing imagery, collecting geophysical data, detecting atmospheric changes, and photographing terrain that ground-based instruments could not reach. The intelligence resided entirely on Earth. Satellites were sensors, and their data traveled on a one-way street to ground stations for analysis by human experts or terrestrial computers.
The second era, which accelerated dramatically in the 2010s with the rise of large low-Earth-orbit broadband constellations, was defined by connectivity. Starlink, OneWeb, and Amazon’s Project Kuiper redefined what satellites could do for everyday users. They became relay nodes — not instruments of observation but conduits of communication, bridging the digital divide for maritime operators, remote communities, airline passengers, and military units far from terrestrial fiber infrastructure. This era’s satellite was a high-speed pipe made extraordinarily dense and inexpensive by the economics of reusable rockets and mass manufacturing.
The third era, which this paper argues has definitively begun in 2025–2026, is defined by intelligence. Satellites will not merely sense or relay; they will interpret, infer, decide, and act. This transition is enabled by four simultaneous developments: the miniaturization of high-performance AI accelerators culminating in NVIDIA’s Vera Rubin Space-1 Module announced in March 2026; the dramatic reduction in launch costs brought by SpaceX’s Falcon 9 and the coming Starship; the maturation of satellite-to-satellite optical communication links; and the explosive demand for inference at the edge created by the AI revolution on Earth.
1.2 The Old Model: Collect, Downlink, Analyze
Traditional satellite systems have relied on a sequential and inherently slow workflow. An Earth observation satellite captures an image. It stores that image onboard until it passes over a ground station. The ground station receives a raw downlink, which may amount to dozens or hundreds of gigabytes per pass. The data travels through terrestrial networks to a processing datacenter. Analysts or automated systems classify, orthorectify, and interpret the imagery. Insights eventually reach the end user — a military commander, a disaster response team, a precision agriculture platform — hours or even days after the moment of collection.
This architecture was adequate when satellite coverage was sparse and the number of satellites was small enough that downlink windows were a manageable bottleneck. But it is structurally incompatible with the future now being built. As Earth observation constellations grow from dozens to thousands of satellites, as synthetic aperture radar and hyperspectral sensors generate orders of magnitude more data per orbit, and as the use cases shift toward real-time applications — wildfire detection, maritime domain awareness, missile warning, autonomous aviation routing — the downlink-and-process model breaks.
1.3 The Bottleneck Problem and Its Consequences
The bottleneck is not merely technical; it is strategic. A disaster response team that receives imagery of a flood zone six hours after collection cannot use that data to route emergency responders in real time. A naval commander who receives vessel detection alerts with a two-hour lag is operating on stale intelligence. A precision agriculture system that gets soil moisture indices at the end of the day cannot optimize irrigation decisions in the morning. In each case, the latency introduced by the collect-downlink-process pipeline erases much of the value that space-based sensing could theoretically deliver.
NASA’s Space Communications and Navigation (SCaN) program describes daily command and data exchange between spacecraft and Earth as one of the most critical bottlenecks in current space operations.[5] The agency’s High-Performance Spaceflight Computing initiative aims to deliver a next-generation system-on-chip with more than 100 times the computing capability of current space processors — a direct acknowledgment that onboard intelligence is the path forward.[6]
1.4 The New Model: Process First, Transmit Later
Orbital Intelligence Networks reverse the logic of the old model. Instead of transmitting everything and processing nothing, the new architecture processes onboard and transmits only what matters. A satellite equipped with a capable AI accelerator can run a cloud detection algorithm before storing imagery, discarding frames that are obscured. It can run a vessel detection model and transmit only the bounding boxes of identified ships, not the raw image pixels. It can run an anomaly detection model over a hyperspectral data stream and alert ground operators only when a signature of interest — a wildfire, a chemical plume, an unusual spectral pattern — has been identified.
The European Space Agency’s Phi-sat-2 mission, which launched in August 2024 and entered its science phase shortly after, provides the clearest existing demonstration of this approach. ESA reports that Phi-sat-2 demonstrates onboard AI for Earth observation, including eliminating cloudy images before downlink, detecting maritime vessels, and transforming images into maps for disaster response.[7] These are not experimental capabilities being validated for some future mission; they are operational functions running on a satellite in LEO today.
1.5 From Satellite Networks to Intelligence Networks
The most important conceptual step in understanding the Orbital Intelligence paradigm is the move from thinking about satellites as individual nodes to thinking about constellations as distributed intelligence networks. A single satellite with onboard AI is an improvement. A constellation of satellites that can share context, coordinate inference tasks, route data to whichever node has the best compute availability, and collectively build a real-time picture of a geographic area or event is something qualitatively different — it is a distributed computing fabric that happens to orbit the Earth at 28,000 kilometers per hour.
NASA’s Jet Propulsion Laboratory describes sensorweb capabilities where data from one satellite can drive the tasking of another, with edge computing and intersatellite links making notifications more rapid.[8] This is the architectural vision toward which the field is converging: not satellites as separate instruments, but satellite constellations as coherent intelligence systems.
Section 2: The Architecture of Orbital Intelligence Networks
The architecture of Orbital Intelligence Networks cannot be understood by looking at any single layer in isolation. The power of this framework lies precisely in the integration of three distinct layers — space, air, and ground — into a unified, dynamic compute fabric. Scholars in the field of space-air-ground integrated networks (SAGIN) have argued that this layered architecture represents the foundation for next-generation ubiquitous connectivity and intelligence.[9] The academic literature on edge-computing-enhanced SAGINs further establishes that this architecture can support compute tasks where terrestrial infrastructure is unavailable — in remote regions, islands, disaster zones, and open ocean.[10]
2.1 The Space Layer: LEO, MEO, GEO, and Cislunar
The space layer spans four orbital regimes, each with distinct characteristics that make it suitable for different functions within the overall network. Low-Earth orbit — typically defined as altitudes between 200 and 2,000 kilometers — offers the lowest latency and the highest data density for Earth observation and broadband. LEO satellites orbit the Earth roughly every 90 minutes, providing frequent revisit rates and round-trip signal delays of 20–40 milliseconds. Large LEO constellations — Starlink’s planned 100,000+ satellites, Blue Origin’s TeraWave, and the proposed orbital AI constellations — are required to provide continuous coverage.
Medium-Earth orbit, at altitudes of 2,000 to 35,786 kilometers, is the home of GPS and navigation constellations. MEO satellites remain in view of any location for hours rather than minutes, making them useful for coordination, aggregation, and persistent coverage functions. Geostationary orbit at 35,786 kilometers — where satellites appear stationary relative to the Earth’s surface — serves as a persistent command and coordination layer. Cislunar space, the region between Earth and the Moon, represents the frontier: Lonestar Data Holdings is actively developing cislunar and lunar data storage and compute infrastructure, with its first commercial LEO service rolling out in Q4 2026 and six lunar data storage spacecraft planned for 2027–2030.
“Throw in all the problems with climate change, natural disasters, human error, wars, nation states going after immutable data held in data centers. Data center customers want to put their data somewhere that is secure, accessible, and in compliance with data sovereignty laws. And space beckons.”
— Chris Stott, CEO, Lonestar Data Holdings [11]
2.2 The Air Layer: Drones, HAPS, and Autonomous Aviation
The air layer is the most underappreciated component of the Orbital Intelligence architecture, but it is functionally essential. High-altitude platform stations — solar-powered unmanned aircraft, stratospheric balloons, and other long-endurance aerial vehicles operating at 20 to 50 kilometers altitude — occupy a middle tier between ground infrastructure and LEO satellites. They combine some of the persistence of GEO with the low altitude of LEO, enabling low-latency sensing and communication. Drones and autonomous aircraft at lower altitudes serve as mobile relay nodes, edge compute platforms, and flexible sensors that can be rapidly deployed to areas where satellite revisit rates are insufficient. The air layer provides the geographic flexibility and temporal continuity that neither ground nor space can offer alone.
2.3 The Ground Layer: Datacenters, Edge Nodes, and Command Infrastructure
The ground layer remains foundational even in a world of orbital AI. Ground-based datacenters provide the computational depth required for large-scale model training, historical data archiving, and complex analytics that cannot be performed in the power-constrained environment of orbit. In the Orbital Intelligence architecture, the ground layer is reconceptualized from a terminus — the final destination of data downloaded from space — into a peer in a distributed compute fabric. A satellite over the Pacific might detect an anomalous vessel, run a preliminary classification onboard, and route the alert to a ground-based model that retrieves historical track data and returns a confidence-weighted assessment to the operator within minutes.
2.4 In-Orbit AI Compute: The Central Innovation
In-orbit AI compute is the enabling technology that transforms the three-layer architecture from a connectivity network into an intelligence network. Without capable onboard compute, satellites remain passive sensors. With it, satellites become active inference nodes capable of running object detection, anomaly identification, natural language processing of signal intercepts, and autonomous task coordination — all before any data touches a ground station.
The NVIDIA Vera Rubin Space-1 Module, announced at GTC 2026 in March, delivers up to 25 times more AI compute per GPU than the H100 for orbital inference workloads.[12] Jensen Huang, NVIDIA’s CEO, declared at the announcement:
“Space computing, the final frontier, has arrived. As we deploy satellite constellations and explore deeper into space, intelligence must live wherever data is generated.”
— Jensen Huang, CEO, NVIDIA Corporation, GTC 2026 [12]
The Space-1 Module is engineered for size-, weight-, and power-constrained environments, and is being adopted by Axiom Space, Starcloud, Planet Labs, Kepler Communications, Sophia Space, and Aetherflux.[13]
2.5 Decentralized Orchestration: Beyond Single Points of Control
The long-term architectural vision is one in which no single ground station, satellite, or company serves as the sole orchestration point. Academic research on space-ground fluid AI proposes extending edge AI to space and using satellite mobility for horizontal and vertical task and model migration across space-ground networks.[14] A 2026 arXiv paper on multi-orbit space-based data center architectures proposes that LEO satellites support radio access and real-time inference while higher orbital layers provide aggregation and orchestration — a hierarchical but decentralized design that mirrors the architecture of content delivery networks on Earth.[15]
Section 3: Why Decentralized In-Orbit AI Compute Matters
3.1 The Latency Advantage in Time-Critical Applications
The fundamental argument for in-orbit AI compute is latency. In the collect-downlink-process model, the time between a satellite sensor detecting an event and a human or automated system acting on that detection is measured in minutes to hours. In a world of orbital AI, that latency can be reduced to seconds — or, for autonomous systems, to the computational time required to run the inference onboard. Consider the applications where seconds matter: a wildfire expanding at the edge of a populated area, a missile launch detected by an infrared sensor, a ship operating in contested waters that changes course and transponder status. In each case, the value of the intelligence scales inversely with its latency. Orbital AI eliminates the latency imposed by the downlink-and-process bottleneck at the point where it matters most: the moment between detection and decision.
3.2 Bandwidth Efficiency and the Compression of Intelligence
The second argument for in-orbit AI compute is bandwidth efficiency. A cloud-detection model running onboard can immediately discard the 60–70% of Earth observation imagery that is obscured by cloud cover, reducing downlink requirements by the same fraction. A vessel detection model can compress a 100-megabyte image to a 10-kilobyte set of bounding boxes and confidence scores. A wildfire detection algorithm can replace a 500-megabyte thermal image with a 1-kilobyte alert message containing location, estimated perimeter, and confidence level. Intelligence, not data, is what the ground needs. Orbital AI produces intelligence. Ground links transmit it.
3.3 Resilience and Redundancy in Contested Environments
A network that depends on a centralized ground processing architecture is vulnerable to the disruption of that architecture. Ground stations can be destroyed, jammed, hacked, or made inaccessible by natural disaster or military action. A decentralized orbital intelligence network is qualitatively more resilient: each satellite node can operate autonomously, maintaining its own situational picture, running its own inference pipeline, and communicating directly with other satellites via intersatellite links if ground connections are unavailable. The network degrades gracefully under stress rather than failing catastrophically.
DARPA’s Blackjack program described this goal precisely: to demonstrate elements of a global high-speed LEO network providing resilient, persistent, and highly connected coverage — explicitly designed to function without dependence on vulnerable ground infrastructure.[16]
3.4 Strategic Autonomy and the Control of Orbital Compute
The countries and companies that control orbital compute infrastructure will hold a significant advantage in the domains of intelligence, communications, defense, climate monitoring, and AI-enabled logistics for the remainder of the twenty-first century. A nation that can field an orbital intelligence network capable of persistent surveillance, autonomous vessel tracking, real-time wildfire mapping, and resilient communications is less dependent on shared ground infrastructure, allied relay systems, and foreign-hosted cloud platforms.
As Stanford Law School’s 2025–2026 policy practicum on governing autonomous and AI systems in outer space notes, the rise of AI-driven autonomy in space is rapidly outpacing the legal and governance frameworks designed to regulate it.[17] The strategic stakes of that gap are enormous.
3.5 The Emergence of a New Economic Layer
Orbital intelligence is not merely a feature added to existing satellite services; it is the foundation of an entirely new industry layer. Just as cloud computing created a new economic category — infrastructure-as-a-service — that did not exist before Amazon Web Services launched in 2006, orbital compute is creating a new economic category: intelligence-as-orbit, inference-as-orbit, and resilience-as-orbit.
“Running advanced AI from space solves the critical bottlenecks facing data centers on Earth. Orbital compute offers a way forward that respects both technological ambition and environmental responsibility.”
— Philip Johnston, CEO, Starcloud [18]
The economic model of orbital AI is still being written. But the foundational demonstrations — Starcloud’s training of the first large language model in orbit in December 2025, Axiom Space’s AxDCU-1 running cloud computing and AI/ML workloads on the ISS in fall 2025, and Lonestar’s April 2026 announcement of StarVault, the world’s first commercial space-based data storage service — establish that this economic layer is real and beginning to generate revenue.[18][19][20]
Section 4: Challenges and Risks of Orbital Intelligence Networks
Intellectual honesty requires that we examine the challenges and risks of Orbital Intelligence Networks with the same rigor we bring to the opportunities. What follows is not a catalogue of reasons to doubt the concept but a rigorous mapping of the engineering, economic, governance, and strategic obstacles that must be overcome for orbital AI to fulfill its potential.
4.1 Power Constraints: The Solar Ceiling
AI inference is energy-hungry. A terrestrial datacenter running large language model inference at scale may draw tens of megawatts from the grid continuously. A satellite, by contrast, is powered by solar panels whose surface area is constrained by the satellite’s mass, launch fairing dimensions, and orbital mechanics. A typical small satellite in LEO generates one to five kilowatts of power. Even the most ambitious proposals — Starcloud’s gigawatt-scale compute roadmap or SpaceX’s 100 GW of AI computing capacity per year — require dramatic advances in solar power generation and power management before they can run serious AI workloads continuously.
Researchers at the University of Pennsylvania’s School of Engineering and Applied Science presented a tether-based architecture for solar-powered orbital AI data centers at the 2026 AIAA SciTech Forum, which they described as “the first design that prioritizes passive orientation at this scale.” Associate Professor Igor Bargatin argued that the tether approach allows orbital datacenters to be scaled to sizes that would meaningfully reduce the energy and water demands of terrestrial datacenters.[21]
4.2 Thermal Management: Cooling Without an Atmosphere
On Earth, datacenters use air, liquid, and evaporative cooling to reject waste heat from compute hardware. In the vacuum of space, none of these mechanisms are available. Heat can only be rejected by radiation, and the rate of radiative heat rejection depends on the radiator’s area, emissivity, and the temperature difference between the radiator and the surrounding environment. For high-power AI workloads, the radiator area required is enormous relative to the satellite’s compute payload.
Google’s Project Suncatcher research paper acknowledges this challenge explicitly, noting that “advanced thermal interface materials and heat transport mechanisms would be required, preferably passive to maximize reliability, to efficiently move large heat loads from the chips to dedicated radiator surfaces.”[22]
4.3 Radiation and Hardware Durability
Space is a hostile radiation environment. Low-Earth orbit satellites are exposed to protons and heavy ions from the Van Allen belts, cosmic rays, and solar energetic particle events. These radiation sources cause total ionizing dose degradation of semiconductor materials and single-event upsets — bit flips in memory or logic that can cause computation errors or permanent hardware damage.
Google’s radiation testing for Project Suncatcher found that the Trillium TPU v6e’s high bandwidth memory subsystems began showing irregularities only after a cumulative dose of 2 krad(Si) — nearly three times the expected shielded five-year mission dose of 750 rad(Si).[22] The Starcloud-1 mission, which launched an NVIDIA H100 GPU into orbit in November 2025, was the first demonstration that commercial-grade AI accelerators can operate in LEO without radiation hardening, achieving 100 times more AI compute than any previous space hardware and validating the commercial silicon approach — though long-term durability over multi-year mission lifetimes remains an open engineering question.[23]
4.4 Cybersecurity and Command Integrity
Orbital AI networks introduce cybersecurity risks that are qualitatively different from those facing terrestrial cloud systems. A compromised satellite compute node is not merely a data breach; it is potentially a physical asset that can be misdirected, deorbited toward inhabited areas, or used to jam communications. The combination of autonomous decision-making capability, physical orbital maneuvering authority, and network connectivity creates an attack surface with physical-world consequences. Command integrity — ensuring that a satellite’s onboard AI system acts only on authenticated instructions — becomes dramatically more complex when the system is running distributed inference across a constellation and sharing model weights via intersatellite links.
As one governance expert has noted: “The hardware is launching faster than the law, and the gap is sharpest where it matters most for sovereign AI: jurisdiction over the data, not just the satellite.”[24]
4.5 Space Debris and Orbital Congestion
The proliferation of orbital compute infrastructure will dramatically increase the number of objects in LEO, compounding the already serious problem of space debris. SpaceX has filed with the FCC for approval to operate one million orbital AI data center satellites. Starcloud has applied to deploy up to 88,000. Blue Origin’s Project Sunrise proposes 51,600. Each of these constellations must be designed, operated, and ultimately deorbited in ways that do not contribute to cascading collision chains.
Stanford Law School’s 2025–2026 policy practicum notes that as of 2024, approximately 10,000 satellites orbit Earth — a number projected by the European Space Agency to grow nearly tenfold by 2030 — and that these satellites must navigate more than 140 million pieces of space debris moving at velocities several times faster than a bullet.[17]
4.6 Legal and Geopolitical Uncertainty
The legal framework governing orbital AI compute was not designed for this application. The Outer Space Treaty of 1967, the Liability Convention, the Registration Convention, and ITU Radio Regulations were all written for state-operated communications and Earth-observation satellites. They offer little clarity on questions of data sovereignty over information processed in orbit, liability for harm caused by autonomous satellite decisions, or export controls on AI models running on orbital hardware.
China has proposed two satellite constellations to the ITU totaling 96,714 satellites for orbital AI infrastructure, reflecting the geopolitical dimensions of spectrum allocation in this new domain.[25] The race to file constellation proposals with the ITU — which operates on a first-come, first-served principle — is itself a form of geopolitical competition in which filing speed translates directly into legal priority.
Section 5: Strategic Opportunities for Space-Based Intelligence
5.1 Disaster Response and Climate Monitoring
Among the most immediate and compelling applications of Orbital Intelligence Networks is the acceleration of disaster response and climate monitoring. Earth is experiencing an increasing frequency of high-impact environmental events — wildfires, floods, hurricanes, volcanic eruptions, and compound weather extremes — each of which demands rapid, accurate, and persistent monitoring from space. The collect-downlink-process model is categorically inadequate for these applications. An orbital AI network capable of running change detection, damage assessment, and emergency routing algorithms continuously and autonomously — transmitting only actionable intelligence to first responders — would fundamentally transform emergency management capability.
5.2 Defense and National Security
The defense and national security implications of Orbital Intelligence Networks are substantial and unavoidable. What changes with orbital AI is the speed, autonomy, and resilience of space-based intelligence. A satellite constellation capable of autonomous vessel tracking, automated change detection of military facilities, real-time signals intelligence processing, and distributed command-and-control relay — without dependence on any single vulnerable ground station — represents a qualitative upgrade in persistent ISR capability. The dual-use nature of orbital AI also creates significant arms control and escalation risks that will require new governance frameworks to manage.
5.3 Global Connectivity and the 6G Architecture
The integration of space, air, and ground infrastructure into a unified network has become a central theme of sixth-generation (6G) telecommunications research. A 2025 arXiv paper frames SAGIN as a key architecture for ubiquitous next-generation coverage and argues that AI is essential for managing the complex control and resource allocation requirements of such a network.[9] The orbital layer of a 6G architecture provides low-latency, high-reliability backbone for time-sensitive applications including real-time autonomous vehicle coordination, surgical robotics, and industrial automation systems.
5.4 Autonomous Systems and Robotics
The proliferation of autonomous systems — unmanned aerial vehicles, autonomous ships, self-driving vehicles, agricultural robots, and eventually autonomous spacecraft — creates an enormous and growing demand for the real-time positional intelligence, routing optimization, and collision avoidance that only a pervasive orbital intelligence network can provide at global scale. A swarm of satellites conducting a distributed aperture radar observation requires precise relative positioning and real-time task coordination that only an orbital AI network with low-latency intersatellite links can support.
5.5 Commercial Orbital Datacenters: The Frontier Opportunity
The most ambitious and contested opportunity in the Orbital Intelligence space is the emergence of commercial orbital AI datacenters — satellite-based compute platforms powered by solar energy, linked by optical intersatellite networks, and providing cloud-equivalent AI compute services to customers on Earth and in space. This concept has moved from whitepaper speculation to active engineering and regulatory engagement over the eighteen months from late 2024 to mid-2026.
Gartner distinguished VP analyst Bill Ray published a report in February 2026 arguing that “companies are wasting money by pouring funds into the orbital data center bubble because the economics do not work,” citing the prohibitive costs of launching hardware and the immense technical challenges of cooling orbital datacenters.[26]
NVIDIA’s Jensen Huang himself acknowledged on the company’s most recent earnings call that “launching datacenters into orbit is a poor economic decision, at least for now,” while simultaneously betting that it is better to be ready for a boom that never comes than to miss it if it does.[26]
“At the beginning of 2026, space-based data centers are no longer a single speculative headline, but a growing set of experiments, some already underway, many announced, testing whether orbital computing can evolve from edge computing into a genuine infrastructural layer.”
— Giorgia Rau, Astrophysicist and Professor, Catholic University of America, writing in Aspenia, 2026 [27]
Section 6: The Competitive Landscape — Established Players and Newcomers
The competitive landscape for space-based intelligence and orbital AI compute has transformed with extraordinary speed over the eighteen months from January 2025 to June 2026. What was a theoretical concept in academic papers and startup whitepapers in early 2024 is now a domain of active competition among the largest technology companies in the world, a growing cohort of well-funded startups, and sovereign programs in China, Europe, and elsewhere.
6.1 SpaceX: The Pioneer Turns Public
SpaceX is the foundational actor in the Orbital Intelligence landscape. Starlink generated $11.39 billion in revenue in 2025, representing 61% of SpaceX’s total revenue, and climbed to 69% of total sales in Q1 2026.[1] SpaceX’s SEC filings ahead of its IPO revealed the company’s filing with the FCC for regulatory approval to operate up to one million orbital AI data center satellites, specifying 100 gigawatts of AI computing capacity per year.[28] Elon Musk has stated that next-generation Starlink satellites could become the lowest-cost AI compute platform within five years.[29]
The SpaceX IPO, which raised $75 billion at a $1.75 trillion valuation on June 12, 2026, provides the capital infrastructure to pursue this vision. The first public earnings report is scheduled for November 2026 and will be the first rigorous data point on which scenario — bull or bear — is unfolding.[30]
6.2 NVIDIA: Bringing the GPU to Orbit
NVIDIA’s March 2026 announcement of the Space-1 Vera Rubin Module at GTC 2026 was a watershed moment in the industrialization of orbital AI compute. The module delivers up to 25 times more AI compute per GPU than the H100 for space-based inferencing, and is engineered for size-, weight-, and power-constrained environments. The announcement was accompanied by partnerships with Axiom Space, Starcloud, Planet Labs, Kepler Communications, Sophia Space, and Aetherflux.[12] NVIDIA’s positioning mirrors its terrestrial strategy: the company provides the accelerated compute substrate on which the industry runs, ensuring that whatever orbital AI infrastructure emerges will likely run on NVIDIA silicon.
6.3 Google: Project Suncatcher and the TPU in Orbit
Google announced Project Suncatcher in November 2025, describing it as a research moonshot to scale machine learning in space. The project involves launching two Planet Labs satellites equipped with Google’s Trillium-generation TPU v6e chips by early 2027, testing TPU performance in the space radiation environment, and demonstrating high-bandwidth optical intersatellite links.[31]
“We will send tiny, tiny racks of machines and have them in satellites, test them out, and then start scaling from there. There is no doubt to me that, a decade or so away, we will be viewing it as a more normal way to build data centers.”
— Sundar Pichai, CEO, Alphabet / Google [31]
Google is also reportedly in talks with SpaceX to provide launch services for the Project Suncatcher prototype satellites.[32] Google’s radiation testing confirmed that the TPU v6e can withstand radiation levels across a five-year LEO mission — a critical engineering milestone validating that commercial AI accelerators can operate in orbit without dedicated radiation hardening.[22]
6.4 Amazon / Blue Origin: Project Sunrise and the Globalstar Acquisition
Amazon announced an agreement to acquire Globalstar, the mobile satellite services operator, combining Globalstar’s spectrum licenses, LEO satellite operations, and direct-to-device technology with Amazon’s Leo connectivity constellation and AWS cloud infrastructure.[33] On the orbital datacenter side, Blue Origin filed with the FCC in March 2026 to deploy up to 51,600 satellites under Project Sunrise, describing it as a means to address mounting demand for AI compute while reducing the high water and energy needs typically seen at terrestrial sites, with connectivity back to Earth handled through the planned TeraWave broadband system.[34]
6.5 Starcloud: The First AI Training in Orbit
Starcloud (formerly Lumen Orbit) is the most operationally advanced pure-play entrant in the orbital AI datacenter race. On November 2, 2025, the Y Combinator-backed company launched Starcloud-1, a 60-kilogram satellite carrying the first NVIDIA H100 GPU ever placed in orbit — hardware 100 times more powerful than any previous space computing hardware.[23]
In December 2025, Starcloud achieved an even more significant milestone: it trained the first large language model in orbit, running NanoGPT — OpenAI founding member Andrej Karpathy’s compact LLM — on the complete works of Shakespeare aboard Starcloud-1. The satellite also ran Google DeepMind’s Gemma model in orbit.[35] Starcloud-2, planned for October 2026, will carry multiple H100 chips and an NVIDIA Blackwell B200 GPU alongside an AWS server blade, with a 7-kilowatt solar array — a 100-fold increase in compute capacity over Starcloud-1. The company subsequently raised a $170 million Series A on the strength of the Starcloud-1 results.[36]
6.6 Axiom Space: The ISS as Orbital AI Testbed
Axiom Space deployed its AxDCU-1 prototype to the International Space Station in fall 2025, running Red Hat Device Edge software with cloud computing, AI/ML, data fusion, and cybersecurity applications.[37] On January 11, 2026, Axiom launched the first two operational orbital data center nodes on a Kepler Communications satellite. A fully optically-interconnected orbital data center node aboard the ISS is planned for 2027, serving as a testbed for future deployments on Axiom Station, the company’s planned commercial space station.[38]
6.7 Lonestar Data Holdings: Lunar and Cislunar Resilience
Lonestar Data Holdings occupies a distinctive position: rather than focusing on LEO compute for AI model inference, Lonestar builds secure, off-Earth data storage and resilience infrastructure for governments and enterprises that need to protect mission-critical information beyond the reach of terrestrial disasters or cyberattacks.[39] In April 2026, the company announced StarVault — the world’s first commercial space-based data storage service — and expanded its agreement with Sidus Space for the manufacture of six LizzieSat lunar data storage spacecraft planned for 2027–2030. Lonestar also holds a NASA Space Act Agreement with NASA Ames Research Center to advance space-based super compute and data storage.[40]
6.8 The Broader Field: China, Europe, and the International Race
China’s Adaspace program has deployed three test satellites equipped with domestic AI accelerators, with state media indicating plans for a 50-satellite constellation by 2028. China has proposed two satellite constellations to the ITU totaling 96,714 satellites for orbital AI infrastructure, with $8.4 billion in projected investment — setting up a space-based AI confrontation with American programs of significant geopolitical dimensions.[25] In Europe, the ASCEND program is planning a demonstration orbital data center module mission in 2026, reflecting the strategic concern that if orbital AI infrastructure is built primarily by American and Chinese companies, European sovereignty over AI-generated intelligence will be structurally compromised.
Section 7: Seven Pillars of Orbital Intelligence Networks
The preceding six sections have traced the historical evolution, architectural logic, strategic rationale, technical challenges, and competitive landscape of Orbital Intelligence Networks. This final section synthesizes those threads into seven foundational pillars — the intellectual and strategic principles that should guide policymakers, investors, engineers, and strategists as they navigate this emerging domain. These pillars are not predictions; they are propositions about what must be true for orbital AI to fulfill its potential and what must be managed if that potential is to be realized without catastrophic unintended consequences.
Pillar 1: Intelligence Must Move Closer to the Sensor
The most durable principle of distributed computing — a principle that has governed the evolution from mainframes to personal computers to mobile devices to IoT edge nodes — is that intelligence migrates toward the data source. In the orbital context, this principle implies that future satellite infrastructure will not be defined by raw data transmission but by onboard interpretation. The closer intelligence moves to the sensor — whether optical imager, synthetic aperture radar, signals intelligence receiver, or environmental monitor — the faster, more efficient, and more resilient the network becomes. This is not a preference; it is an architectural inevitability driven by the physics of data transmission, the economics of bandwidth, and the strategic requirements of time-critical applications.
Pillar 2: Space-Air-Ground Integration Is the New Network Architecture
The future of global intelligence infrastructure will not be purely terrestrial and not purely orbital. It will be a continuously optimized, dynamically reconfigured fabric spanning satellites in multiple orbital regimes, high-altitude platforms, drones, autonomous aircraft, telecom towers, ground datacenters, and edge devices. These layers do not operate in isolation; they operate as a single coherent system, routing data, compute tasks, and intelligence to wherever in the fabric those operations can be performed most efficiently, reliably, and securely.
The academic literature on SAGIN architectures is converging on this conclusion from multiple directions — telecommunications research, edge computing theory, autonomous systems coordination, and military communications planning.[9][10] The practical demonstrations — SpaceX’s Starlink providing resilient communications in contested environments, Axiom’s orbital nodes relaying processed imagery to ground operators, Lonestar’s cislunar storage providing disaster recovery for enterprise customers — are confirming it empirically.
Pillar 3: Decentralization Creates Resilience
Centralized ground-based processing architectures create single points of failure that adversaries, disasters, and system failures can exploit. A decentralized orbital intelligence network — in which each satellite can process data independently, each intersatellite link can route around damaged nodes, and each orbital cluster can maintain situational awareness without ground connectivity — is orders of magnitude more resilient than any centralized alternative. This resilience is a strategic imperative for national security applications and a functional requirement for emergency response systems that must operate precisely when terrestrial infrastructure has been disrupted.
Pillar 4: Orbital Compute Is the New Strategic Space Asset
The first era of space power was defined by launch capability. The second era was defined by coverage. The third era will be defined by compute. Launch vehicles and satellites remain essential, but the strategic differentiator will be the AI compute capability that orbital infrastructure can deploy — the number of inference operations per second that a constellation can perform, the latency at which it can detect and respond to events, and the resilience with which it can maintain those capabilities under adversarial conditions.
As James Landay of Stanford has observed, the concentration of AI capabilities is a major geopolitical issue, with more than 70% of the computing power dedicated to advanced AI controlled by fewer than ten companies, primarily American and Chinese — and this imbalance is driving nations to invest in sovereign AI infrastructure.[41]
Pillar 5: The Energy Constraint Will Reshape Both Earth and Orbit
IEA data establishes that AI-focused data centers consumed 50% more electricity in 2025 than in 2024, and that the largest technology companies invested over $400 billion in data center capital expenditure in 2025, with a 75% increase projected for 2026.[3] Every gigawatt of AI compute deployed in orbit on solar-powered satellites is a gigawatt not competing for scarce terrestrial energy resources. As the marginal cost of terrestrial power for AI applications rises due to grid congestion, environmental permitting, and political resistance to large-scale datacenter development, the comparative economics of orbital compute improve.
“I don’t think we’ll have an operative data center in space in the next couple of years, but we’ll start seeing some of the building blocks tested in the next couple of years.”
— Josep Jornet, Professor of Computer and Electrical Engineering, Northeastern University [42]
That prediction, made in January 2026, is already being overtaken by events. The building blocks are being tested now, in real orbital hardware, by Starcloud, Axiom, Lonestar, and Google’s Project Suncatcher partnerships. The transition from building blocks to operational infrastructure will be determined by the rate at which launch costs fall, thermal management solutions scale, and radiation tolerance of commercial AI silicon is empirically validated over multi-year mission lifetimes.
Pillar 6: Orbital AI Requires Governance Before It Scales
The technical opportunity represented by Orbital Intelligence Networks is enormous. The governance gap is equally enormous, and the risks of allowing the former to outpace the latter are severe. Space debris, cybersecurity vulnerabilities, military escalation dynamics, data sovereignty conflicts, and spectrum allocation disputes are not hypothetical concerns for some distant future; they are active challenges already manifesting as orbital AI infrastructure scales.
The existing legal framework — the 1967 Outer Space Treaty, the Liability Convention, the Registration Convention, and ITU Radio Regulations — was not designed for constellations of one million compute-enabled satellites operated by private companies using autonomous AI systems to make real-time decisions with physical consequences. The gap between the existing framework and the emerging reality is a fundamental mismatch between the governance architecture and the technology it is meant to govern — one that must be addressed before the technology scales beyond easy governance.
Pillar 7: Orbital AI Will Redefine What It Means to Be a Data Center
The seventh and most profound pillar is also the most philosophical: the emergence of orbital AI compute represents a redefinition of the data center itself. The data center is no longer necessarily a building. It is a compute node — a collection of processing, memory, power, and networking capacity — that may be located in a basement, in a converted warehouse, on a high-altitude platform, or in a 60-kilogram satellite circling the Earth at 28,000 kilometers per hour.
This reconceptualization dissolves the geographic constraint that has defined where AI compute can be deployed and therefore where AI intelligence can be generated in real time. A wildfire monitor in a remote forest can receive AI-processed satellite imagery without waiting for a fiber connection to a distant datacenter. A maritime vessel in the middle of the Pacific can access inference services from an orbital compute node passing overhead. A military unit operating in a communications-denied environment can receive processed intelligence from a satellite constellation operating autonomously without any ground support. The implications extend beyond any single application or market: they suggest that the future of artificial intelligence as a physical computing substrate will be distributed across Earth orbit in ways that are only beginning to be imagined.
Conclusion: The Rise of Space-Based Intelligence
On the day that SpaceX made its debut on the Nasdaq — raising $75 billion, achieving a valuation above $1.77 trillion, and marking the largest initial public offering in financial history — it is tempting to view the orbital AI story as primarily a capital markets narrative. It is not. It is a structural transformation in the architecture of intelligence itself.
The first satellites were instruments of observation. The second generation were instruments of connectivity. The third generation — now being designed, built, launched, and operated — are instruments of intelligence. They will not merely look at the Earth or relay signals between distant points; they will interpret what they see, coordinate with one another, make decisions within their authority, and compress the time between event and response from hours to seconds for the applications that matter most.
The seven pillars articulated in this paper — intelligence at the sensor, space-air-ground integration, decentralization, orbital compute as a strategic asset, energy as the forcing function, governance as a prerequisite, and the redefinition of the data center itself — provide a framework for understanding both the opportunity and the responsibility that Orbital Intelligence Networks represent.
The competitive landscape as of June 2026 — SpaceX filing for one million orbital AI satellites on the day of its IPO, NVIDIA launching the Vera Rubin Space-1 Module at GTC 2026, Google partnering with Planet Labs on Project Suncatcher, Blue Origin filing for 51,600 Project Sunrise satellites, Starcloud training the first LLM in orbit, Axiom Space deploying nodes on the ISS and in the Kepler constellation, Lonestar announcing StarVault as the world’s first commercial space-based data storage service, and China proposing 96,714 ITU satellite slots for orbital AI — is not the conclusion of this story. It is the opening chapter.
What is certain is that the future of AI will not be located only inside terrestrial datacenters. Part of it is moving above Earth, distributed across constellations of intelligent satellites, connected to air and ground systems, and operating as a new layer of planetary intelligence infrastructure. Orbital Intelligence Networks are not merely a space technology concept or a capital markets story or a geopolitical competition. They are a framework for understanding how intelligence itself may become distributed across Earth orbit — and what that means for civilization, for sovereignty, for security, and for the long-term relationship between human intelligence and the machines we are building to extend it.
Footnotes and Endnotes:
[1] Yahoo Finance / CNBC. “SpaceX IPO Live Updates: Elon Musk’s SpaceX Opens at $150 Per Share in Record Debut.” Yahoo Finance / CNBC, June 12, 2026. https://finance.yahoo.com/markets/live/spacex-ipo-live-updates-elon-musks-spacex-set-to-make-record-debut-as-dow-sp-500-nasdaq-rise-230015961.html
[2] CNBC. “SpaceX IPO Live Updates: SPCX Pops More Than 25% on Historic Volume in Nasdaq Debut.” CNBC, June 12, 2026. https://www.cnbc.com/2026/06/12/spacex-ipo-spcx-live-updates.html
[3] International Energy Agency (IEA). “Key Questions on Energy and AI: Executive Summary.” IEA, April 14, 2026. https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary
[4] Northeastern University / IEA. “The Next Great Space Race: Building Data Centers in Orbit.” Northeastern Global News, January 6, 2026 (citing IEA Electricity 2024 Report). https://news.northeastern.edu/2026/01/06/ai-data-centers-in-space/
[5] NASA SCaN. NASA Space Communications and Navigation (SCaN) Program Overview. https://www.nasa.gov/directorates/heo/scan/
[6] NASA. “NASA High-Performance Spaceflight Computing.” NASA Technology Overview. https://www.nasa.gov/
[7] European Space Agency (ESA). “Phi-sat-2: Onboard AI for Earth Observation.” ESA Official News, 2024. https://www.esa.int/
[8] NASA / JPL. “FAME and Sensorweb Concepts.” AI.JPL.NASA.Gov, Jet Propulsion Laboratory. https://ai.jpl.nasa.gov/
[9] arXiv. “Space-Air-Ground Integrated Networks (SAGIN) for Next-Generation Coverage.” arXiv, 2025. https://arxiv.org/
[10] arXiv. “Edge Computing-Enhanced Space-Air-Ground Integrated Networks.” arXiv, 2021. https://arxiv.org/
[11] InformationWeek / Lonestar Data Holdings. “Lunar Data Centers Loom on the Near Horizon.” InformationWeek, April 21, 2025. Chris Stott, CEO, Lonestar Data Holdings. https://www.informationweek.com/it-infrastructure/lunar-data-centers-loom-on-the-near-horizon
[12] NVIDIA Newsroom / CNBC. “NVIDIA Launches Space Computing, Rocketing AI Into Orbit.” NVIDIA Press Release and CNBC Coverage, March 16, 2026. Jensen Huang, CEO, NVIDIA. https://nvidianews.nvidia.com/news/space-computing
[13] NVIDIA. “Space Computing: On-Orbit AI and Accelerated Computing.” NVIDIA Official Page. https://www.nvidia.com/en-us/edge-computing/space-computing/
[14] arXiv. “Space-Ground Fluid AI: Extending Edge AI to Space.” arXiv. https://arxiv.org/
[15] arXiv. “Space-Based Data Centers for 6G and Beyond: Multi-Orbit Architectures.” arXiv, 2026. https://arxiv.org/
[16] DARPA. “Blackjack Program Overview.” DARPA Official Site. https://www.darpa.mil/
[17] Stanford Law School. “Governing Autonomous and AI Systems in Outer Space (810G).” Stanford Law School Practicums 2025-2026. https://law.stanford.edu/education/only-at-sls/law-policy-lab/practicums-2025-2026/governing-autonomous-and-ai-systems-in-outer-space-810g/
[18] CNBC / Starcloud. “NVIDIA-Backed Starcloud Trains First AI Model in Space.” CNBC, January 9, 2026. Philip Johnston, CEO, Starcloud. https://www.cnbc.com/2025/12/10/nvidia-backed-starcloud-trains-first-ai-model-in-space-orbital-data-centers.html
[19] Cutter Consortium. “On-Orbit Data Centers: Mapping the Leaders in Space-Based AI Computing.” Cutter Consortium, September 17, 2025. https://www.cutter.com/article/orbit-data-centers-mapping-leaders-space-ai-computing
[20] Lonestar Data Holdings. “Lonestar Announces StarVault, The World’s First Commercial Space-Based Data Storage Service.” Lonestar Press Release, April 15, 2026. https://www.lonestar.space/
[21] University of Pennsylvania / AIAA. “Powering AI From Space, at Scale.” Penn Engineering, January 28, 2026. Prof. Igor Bargatin, MEAM, University of Pennsylvania. https://www.seas.upenn.edu/stories/powering-ai-from-space-at-scale/
[22] Google Research / Data Center Dynamics. “Project Suncatcher: Google to Launch TPUs into Orbit with Planet Labs.” Data Center Dynamics, May 7, 2026; Google Research Blog, November 2025. https://www.datacenterdynamics.com/en/news/project-suncatcher-google-to-launch-tpus-into-orbit-with-planet-labs-envisions-1km-arrays-of-81-satellite-compute-clusters/
[23] BlacKnight Space Labs. “First AI Training in Space: Starcloud’s NVIDIA H100 Orbital Mission.” BlacKnight Space Labs, April 3, 2026. https://blacknightspacelabs.com/blog/starcloud-nvidia-h100-orbit-first-ai-training-space
[24] Outlook Business. “Why India’s AI Start-ups Are Exploring Space-Based Data Centres.” Outlook Business, May 2026. https://www.outlookbusiness.com/deeptech/india-ai-startups-space-data-centres-orbital-computing
[25] Carbon Credits. “China’s $8.4B Orbital Data Center Push Sets Up Space-Based AI Showdown With SpaceX.” Carbon Credits, April 27, 2026. https://carboncredits.com/chinas-8-4b-orbital-data-center-push-sets-up-space-based-ai-showdown-with-spacex/
[26] The Register / Gartner / NVIDIA. “Nvidia Rolls Out Rubin Module for Space-Based Computing.” The Register, March 2026. Bill Ray, Distinguished VP Analyst, Gartner; Jensen Huang, CEO, NVIDIA. https://www.theregister.com/2026/03/17/nvidia_chips_in_spaaaaaace/
[27] We Build Value / Aspenia / Catholic University of America. Giorgia Rau, astrophysicist and professor, Catholic University of America, in Aspenia; cited in “Data Centers in Space: The Future of Digital Infrastructure.” We Build Value, 2026. https://www.webuildvalue.com/en/longstory/data-centers-space.html
[28] SpaceNews. “SpaceX FCC Filing: One Million Orbital AI Data Center Satellites.” SpaceNews, 2026. https://spacenews.com/spacex-files-plans-for-million-satellite-orbital-data-center-constellation/
[29] News9Live. “AI Model Trained in Earth Orbit for First Time: Starcloud Tests Succeed.” News9Live, December 11, 2025 (citing Elon Musk statement, November 2025). https://www.news9live.com/technology/artificial-intelligence/starcloud-ai-in-space-first-llm-training-nvidia-h100-satellite-tests-2911048
[30] TradingKey. “SpaceX IPO Is Live at $135: Bull, Base and Bear Cases for the First 90 Days.” TradingKey, June 12, 2026. https://www.tradingkey.com/analysis/stocks/us-stocks/261960721-spacex-ipo-is-live-at-135-bull-base-and-bear-cases-for-the-first-90-days-tradingkey
[31] Google Blog / Space.com. “Project Suncatcher Explores Powering AI in Space.” Google Official Blog, November 4, 2025. Sundar Pichai, CEO, Alphabet / Google. https://blog.google/innovation-and-ai/technology/research/google-project-suncatcher/
[32] Aerotime. “Google and SpaceX Discuss Launches for AI Data Centers in Orbit.” Aerotime, May 2026. https://www.aerotime.aero/articles/google-spacex-orbital-ai-data-centers
[33] SEC / Globalstar. “Amazon Acquires Globalstar for Amazon Leo Direct-to-Device Integration.” Globalstar Form 8-K, FY2026. https://www.sec.gov/Archives/edgar/data/1366868/000114036126014528/ef20070409_ex99-1.htm
[34] AI Business / Data Centre Magazine. “Bezos’ Blue Origin Joins Race to Put AI Data Centers in Space.” AI Business, March 23, 2026. Blue Origin FCC Filing, Project Sunrise, March 19, 2026. https://aibusiness.com/data-centers/bezos-blue-origin-joins-race-ai-data-centers-in-space
[35] Data Center Dynamics. “Starcloud Runs AI Model in Space.” Data Center Dynamics, December 2025. https://www.datacenterdynamics.com/en/news/starcloud-runs-ai-model-in-space/
[36] BlacKnight Space Labs. “First AI Training in Space: Starcloud’s NVIDIA H100 Orbital Mission.” BlacKnight Space Labs, April 3, 2026 (Starcloud-2 and Series A details). https://blacknightspacelabs.com/blog/starcloud-nvidia-h100-orbit-first-ai-training-space
[37] Cutter Consortium. “On-Orbit Data Centers: Mapping the Leaders in Space-Based AI Computing.” Cutter Consortium, September 2025 (Axiom AxDCU-1). https://www.cutter.com/article/orbit-data-centers-mapping-leaders-space-ai-computing
[38] Introl Blog. “Orbital Data Center Race 2026.” Introl, February 21, 2026 (citing SpaceNews: Axiom Space Kepler satellite nodes). https://introl.com/blog/orbital-data-centers-space-computing-race-2026
[39] Data Center Frontier / Lonestar. “Lonestar Data Makes it to the Moon on IM-1 Lunar Lander.” Data Center Frontier, 2024. Chris Stott, CEO, Lonestar Data Holdings. https://www.datacenterfrontier.com/edge-computing/article/33037610/lonestar-data-makes-it-to-the-moon-on-im-1-lunar-lander
[40] Lonestar Data Holdings / Sidus Space. “Lonestar Announces StarVault.” Lonestar Press Release, April 15, 2026; NASA Space Act Agreement, Lonestar and NASA Ames. https://www.lonestar.space/
[41] Aivancity / Stanford. “AI in 2026: Key Predictions from Stanford Experts.” Aivancity, January 14, 2026. Citing James Landay, Stanford University. https://aivancity.ai/en/blog/2026-se-dessine-ce-que-les-experts-de-stanford-prevoit-pour-lavenir-de-lintelligence-artificielle/
[42] Northeastern University. “The Future of Space Computing and the Hurdles for Orbital AI.” Northeastern University College of Engineering, January 8, 2026. Prof. Josep Jornet. https://coe.northeastern.edu/news/the-future-of-space-computing-and-the-hurdles-for-orbital-ai/
