NVIDIA’s H100 GPU Launches AI Processing Into Orbit

Nvidia’s groundbreaking H100 GPU has embarked on its first space mission, marking a pivotal moment for artificial intelligence and data processing beyond Earth’s atmosphere. Launched on November 2, this powerful hardware is testing the feasibility of orbital data centers. With 80 gigabytes of RAM, the H100 delivers a hundredfold performance increase over any previous spaceborne computer. It will handle demanding AI tasks such as analyzing Earth observation imagery and running Google’s large language model. This initial test flight, hosted on the Starcloud-1 satellite developed by Virginia startup Starcloud, represents the opening phase of a bold strategy to relocate energy-intensive data infrastructure into space.

Supporters of the concept highlight clear advantages. Orbiting high above the planet, data centers would no longer occupy valuable terrestrial real estate. They would also slash energy and water consumption used for cooling, while completely avoiding greenhouse gas emissions. Although Nvidia’s Jetson machine-learning boards have previously operated on small satellites, the H100’s deployment initiates an entirely new era for AI data processing in space, potentially enabling commercial services to begin as early as next year.

Philip Johnston, CEO and cofounder of Starcloud, emphasized the leap in capability, stating, “The H100 is roughly one hundred times more powerful than any GPU computer previously operated in orbit. This will be the first instance of a terrestrial-grade data-center GPU functioning in space.”

The three-year mission launched aboard a SpaceX Falcon 9 rocket. The 60-kilogram Starcloud-1 satellite will circle Earth in a very low orbit, approximately 350 kilometers high. While in orbit, it will receive streams of synthetic aperture radar (SAR) data from Capella’s Earth-observation satellite fleet. The H100 will process this information in real time, transmitting only the distilled insights back to ground stations.

Johnston explained the transformational benefit of this approach: “Historically, downlinking raw SAR data has been extremely challenging due to its enormous volume. By processing it on orbit, we can downlink just the essential insight, for example, identifying a vessel’s location, speed, and direction. That insight might be a mere one-kilobyte data packet, compared to the hundreds of gigabytes required for the raw dataset.”

Orbital data processing for satellites is just one aspect of Starcloud’s broader vision. The company believes that with ongoing advances in rocket technology, particularly the cost reductions anticipated from SpaceX’s Starship, future large-scale computing infrastructure could be situated in space rather than consuming precious Earth-based resources.

Josh Parker, Nvidia’s head of sustainability, outlined the environmental advantages in an email: “With AI’s escalating energy requirements, orbital data centers offer a transformative environmental solution. They can reduce greenhouse gas emissions dramatically and eliminate the need for complex cooling systems. By leveraging continuous solar power and avoiding land use and fossil fuels, Starcloud’s technology enables rapid, sustainable expansion of data infrastructure, helping safeguard Earth’s climate and vital natural resources.”

Current projections from the International Energy Agency indicate that global data infrastructure could consume as much electricity by 2030 as the entire nation of Japan. Data centers also place immense strain on water supplies; according to World Economic Forum estimates, a single one-megawatt data center may use as much water daily as one thousand people in a developed country. As AI advances drive computing demand ever higher, these resource pressures intensify, often burdening local communities with rising costs and supply disruptions. Moving data centers into space could effectively address these challenges.

Johnston shared his outlook: “I anticipate that within a decade, nearly all new data centers will be constructed in space, primarily due to the energy constraints we face on Earth.” He noted that achieving fully green operations for terrestrial data centers demands massive investment in solar generation and battery storage. In space, however, no batteries are necessary since sunlight is available around the clock. Additionally, each solar panel in orbit generates about eight times more electricity than its Earth-based counterpart, further driving down operational expenses.

“The only extra cost we incur in space is launch,” Johnston added. “We project breaking even with launch costs around $500 per kilogram. With Starship, we anticipate launch expenses dropping significantly lower.” Industry estimates suggest that once SpaceX’s Starship becomes fully operational, launch prices could fall to between $150 and as little as $10 per kilogram. The rocket has completed six successful test flights to date and is slated to begin deploying satellites next year.

Starcloud is already preparing its follow-up mission, aiming to launch a data center ten times more powerful than Starcloud-1 next year. The Starcloud-2 mission will be equipped with Nvidia’s Blackwell GPU alongside multiple H100 units. Delivering seven kilowatts of computing power, this platform is expected to offer commercial services to clients including Earth-observation satellite operators and the U.S. Department of Defense.

An even larger satellite, capable of 100 kilowatts, is planned for 2027. By the early 2030s, Starcloud envisions operating a 40-megawatt data center in space, processing data at costs competitive with terrestrial facilities. Starcloud is not alone in pursuing off-world computing; Axiom Space announced similar plans earlier this year, while Florida’s Lonestar Holdings already sent a small data center to the Moon and aims to establish significant computing infrastructure there in the coming years.

(Source: Spectrum IEEE)