Starcloud Launches Orbital AI Data Center With NVIDIA H100 GPU

On November 2, 2025, space-based data center startup Starcloud launched its first AI data center into orbit aboard a SpaceX rocket. The single-GPU system, running an NVIDIA H100 in a custom package roughly the size and weight of a dorm-room refrigerator, reached a key milestone this week by successfully running inference workloads and returning responses using Google’s Gemma large language model.

According to CNBC, one of the first messages transmitted by the orbital data center read:

“Greetings, Earthlings! Or, as I prefer to think of you—a fascinating collection of blue and green. Let’s see what wonders this view of your world holds. I’m Gemma, and I’m here to observe, analyze, and perhaps occasionally offer a slightly unsettlingly insightful commentary. Let’s begin!”

Data Center Frontier previously reported on Starcloud’s plans, including its broader Project Suncatcher roadmap, shortly after the company’s initial announcement.

From Terrestrial Limits to Orbital Compute

Starcloud is a deep-tech startup based in Redmond, Washington, focused on building orbital data centers: AI compute infrastructure deployed in space rather than on Earth. The company’s core premise is that terrestrial, large-scale data centers are increasingly constrained by land availability, grid capacity, water requirements for cooling, permitting timelines, and environmental impact. As AI compute demand accelerates, Starcloud argues that scaling exclusively on Earth will become progressively harder, pushing the industry to look beyond traditional geographies, and eventually beyond the planet itself.

According to the company’s white paper, orbital data centers could theoretically scale to gigawatt-class capacity, well beyond today’s typical terrestrial deployments. While that ambition may sound extreme, the scale of current hyperscale and AI factory investment suggests the industry is already moving toward facilities measured in hundreds of megawatts, with gigawatt-scale campuses increasingly part of long-range planning discussions.

The appeal of orbital data centers centers on power and cooling. Space offers two structural advantages: near-continuous solar energy exposure and radiative cooling through the vacuum of space, eliminating the need for water-intensive cooling systems or mechanical chillers. Starcloud often summarizes this concept as “cloud computing…above the clouds”—a shorthand for shifting compute growth away from Earth-based infrastructure, with its land, power, and environmental constraints, to an environment where energy is abundant, cooling is inherent, and physical scale is less tightly bound.

Starcloud-1: A Proof Point for Orbital AI Compute

The first concrete step in Starcloud’s roadmap is the Starcloud-1 satellite, a 60-kilogram platform roughly the size of a small refrigerator that sits somewhere between a technology demonstrator and a true proof-of-concept. Uniquely, the satellite carries what Starcloud says is the first terrestrial, data-center-class GPU ever deployed in orbit: an NVIDIA H100.

General-purpose computers have flown aboard the International Space Station for years, but according to Starcloud and independent reporting, the H100 represents a step change in capability, delivering on the order of 100 times more compute performance than any prior space-based system.

The mission’s primary objective is to determine whether a high-power GPU can operate reliably in the harsh conditions of space, including vacuum exposure, radiation, and repeated thermal cycling. At the same time, the deployment is intended to demonstrate that the core elements of an “AI data center in space” (power delivery, thermal management, and sustained operation) can be maintained throughout the satellite’s operational life.

Starcloud has said the satellite is designed to enable high-powered inference and fine-tuning workloads that could eventually support other spacecraft. In that sense, Starcloud-1 is not just a self-contained experiment, but a foundational building block for a broader orbital AI infrastructure. The company’s publicly shared roadmap outlines an ambition to scale from small demonstrators like Starcloud-1 toward gigawatt-class orbital data centers, potentially far larger than today’s Earth-based facilities.

More precisely, the milestone is not the training of an AI model in space, but the successful operation of a data-center-class GPU in orbit. By packaging that capability into a compact satellite platform, Starcloud is signaling how it envisions the technical foundation for much larger, and more ambitious, orbital compute systems.

Reality—or “Pie in the Sky”?

The deployment of an NVIDIA H100 into orbit represents both a symbolic and practical milestone: the first time a data-center-class GPU has operated in space. That matters not only for Starcloud, but for the broader AI infrastructure ecosystem. It advances the idea that compute and energy could eventually be decoupled from Earth’s physical and environmental constraints. On Earth, energy scarcity, water availability, land use, and permitting timelines increasingly shape where and how data centers can be built. In space, many of those constraints largely disappear.

If viable at scale, orbital compute could enable a new phase of what might be called “compute industrialization,” unconstrained by terrestrial geography. The critical caveat, however, is economics. For this model to move beyond experimentation, launch and deployment costs would need to fall below, or at least meaningfully compete with, the total cost of building and operating equivalent AI factories on Earth.

From Novelty to Infrastructure

For AI researchers and organizations operating at the frontier—very large models, high-throughput analytics, global satellite telemetry, and Earth or climate monitoring—space-based compute could theoretically enable massive scale without the environmental trade-offs associated with terrestrial infrastructure. Strategically, this raises the possibility of a distributed “space cloud,” offering a new availability zone with its own trade-offs around latency, cost, resilience, and sustainability.

Yet the challenges familiar from Earth-bound cloud infrastructure do not disappear simply because compute moves into orbit. Automation, predictive operations, and AI-driven data center infrastructure management (DCIM) would remain essential. Without those capabilities, orbital data centers risk inheriting many of the same operational fragilities that have plagued terrestrial hyperscale environments.

To move beyond being a technological novelty, the economics must work not only at launch, but across ongoing operations and long-term asset lifecycles. If orbital compute can deliver on its promised advantages at scale, it could mark a meaningful shift in where and how AI workloads are deployed—particularly as global compute demand continues to outpace what Earth-based infrastructure can sustainably support.

That vision may already be aligning with institutional interest. On December 15, 2025, NASA issued a request for information (RFI) seeking AI technologies capable of supporting Earth-independent space operations. According to the agency, the goal is to enable crews to operate autonomously during missions that experience significant communication delays or outages, allowing onboard systems to assist with anomaly detection and decision-making when ground support is unavailable. A distributed orbital AI architecture of the kind Starcloud envisions could eventually play a role in such scenarios.

Broader efforts to embed AI into space operations are already underway. At the 2025 International Conference on Space Robotics, researchers presented a system designed to help Astrobee—a fan-powered robotic assistant—navigate the International Space Station autonomously. While modest in scale, the work illustrates how AI-driven decision-making is beginning to migrate into operational space environments.

Still, a fundamental question remains. Is orbital compute simply a more dramatic variation on past attempts to site data centers in unconventional locations—such as beneath oceans, in polar regions, or on glaciers—many of which proved technically viable but commercially marginal? If so, space-based data centers may face similar challenges in translating novelty into durable infrastructure. The unanswered question is where, among these unconventional approaches, orbital data centers ultimately find their footing.

Building a Space-Based Power Grid for Orbital Compute

Aetherflux is an American aerospace and renewable energy startup focused on rethinking how energy and eventually high-performance computing can be delivered by building infrastructure in space rather than on Earth.

The company’s stated mission is to create what it calls “an American power grid in space.” That vision builds on the long-studied concept of space solar power (SSP): harvesting solar energy outside Earth’s atmosphere, where sunlight is stronger and uninterrupted by night or weather, and transmitting it to where it is needed.

Unlike earlier SSP proposals centered on massive geostationary platforms beaming microwaves to large ground receivers, Aetherflux is pursuing a more distributed architecture. Its approach relies on constellations of smaller satellites in low Earth orbit (LEO), equipped with high-efficiency solar arrays and infrared laser-based power transmission. The design emphasizes modularity, scalability, and reduced ground-station footprints, with potential applications ranging from military energy resilience to disaster response and remote infrastructure support.

More recently, Aetherflux has outlined an even more ambitious initiative it calls “Galactic Brain.” The concept envisions an orbital constellation of data center satellites designed to host high-performance computing workloads (particularly AI-driven applications) powered directly by solar energy harvested in space. By locating compute infrastructure above the constraints that increasingly define terrestrial data center development, including land availability, grid capacity, cooling, and regulatory friction, the company aims to deliver continuously available, scalable compute capacity from orbit.

According to Aetherflux, the first orbital data center node supporting this vision is targeted for deployment in early 2027, a milestone that would mark a transition from conceptual architecture toward early commercial execution.

At Data Center Frontier, we talk the industry talk and walk the industry walk. In that spirit, DCF Staff members may occasionally use AI tools to assist with content. Elements of this article were created with help from OpenAI's GPT5.

Keep pace with the fast-moving world of data centers and cloud computing by connecting with Data Center Frontier on LinkedIn, following us on X/Twitter and Facebook, as well as on BlueSky, and signing up for our weekly newsletters using the form below.