Starcloud Achieves Milestone in Space-Based AI with NVIDIA H100

Starcloud, an NVIDIA-backed startup, has made history by becoming the first company to train large language models (LLMs) in space. Utilizing its Starcloud-1 satellite, which was launched last month, the company successfully employed an NVIDIA H100 GPU to train models, including Andrej Karpathy’s nano-GPT on the complete works of Shakespeare and conducted inference on Google DeepMind’s open Gemma model.

In a LinkedIn post, Philip Johnston, founder and CEO of Starcloud, expressed excitement over this achievement, stating, “We just trained the first LLM in space using an NVIDIA H100 on Starcloud-1! We are also the first to run a version of Google’s Gemini in space!” He emphasized the significance of this development, noting that it could help shift computing resources from Earth to space, thereby reducing the strain on terrestrial energy resources.

Pioneering Space-Based Computing

Starcloud CTO Adi Oltean commented on the technical challenges of operating the H100 GPU in space, highlighting the innovation and dedication of the engineering team. He noted that the team executed inference on a preloaded Gemma model, with plans to test additional models in the future.

Founded in 2024, Starcloud advocates for orbital computing as a solution to environmental issues associated with conventional data centers. According to the International Energy Agency, electricity consumption from data centers is projected to more than double by 2030. Traditional facilities are also grappling with water scarcity and rising emissions, while orbital platforms can harness uninterrupted solar energy, mitigating cooling challenges.

Starcloud envisions a future where a 5-gigawatt space-based data center, entirely powered by solar panels, spans four kilometers in width and height. The startup claims this system would exceed the output of the largest power plants in the United States while being more compact and cost-effective than equivalent terrestrial solar farms.

Competition in Space-Based Data Centers

Starcloud is not alone in its pursuit of space-based data centers. Other tech giants, including Google, SpaceX, and Blue Origin, are also exploring this frontier. Google recently unveiled Project Suncatcher, aimed at deploying AI data centers in orbit. This initiative involves satellites equipped with custom tensor processing units that will form a distributed compute cluster linked by high-throughput free-space optical connections.

During a recent interview, Google CEO Sundar Pichai described the project as a “moonshot,” indicating the company’s ambition to harness solar energy efficiently. Initial tests using small machine racks on satellites are scheduled for 2027, with hopes for broader adoption within a decade.

Meanwhile, Elon Musk announced plans for SpaceX to develop orbital data centers utilizing next-generation Starlink satellites. Musk aims to establish these facilities as the most cost-effective AI compute option within five years. He claimed that Starlink V3 satellites have the potential to become the backbone of future orbital computing infrastructure.

A recent report indicated that SpaceX is preparing for an initial public offering in 2026, aiming to raise over $25 billion at a valuation exceeding $1 trillion. Proceeds from the IPO are intended to fund the development of space-based data centers and procure the necessary chips for operation.

“Starship should be able to deliver around 300 GW per year of solar-powered AI satellites to orbit, maybe 500 GW. The ‘per year’ part is what makes this such a big deal,” Musk stated on X. He noted that at 300 GW/year, AI in space could surpass the entire energy consumption of the US economy in intelligence processing every two years.

The advancements being made by Starcloud and other tech companies signal a potential transformation in how we approach computing and energy consumption, promising a future where space may become the next frontier for technological innovation.