Space & Industry

Extreme US challenge: artificial intelligence in orbit

SpaceX has applied to the FCC for a data centre system capable of supporting up to one million satellites. Nvidia is adapting Vera Rubin for the Starmind programme. Google has sent AI into orbit with Suncatcher. Against this backdrop, the IEA estimates that energy consumption by terrestrial data centres will reach around 950 terawatt-hours by 2030.

by 24Ore NextMed

Aggiungi Il Sole 24 Ore
ai preferiti su Google
Missione Artemis della Nasa (Kennedy Space Center di Cape Canaveral in Florida) REUTERS

4' min read

Translated by AI
Versione italiana

4' min read

Translated by AI
Versione italiana

Whilst stock markets are in the red and an increasing number of analysts are linking the term ‘bubble’ to AI, the race towards new frontiers shows no sign of stopping. A race is currently underway between American private sector giants to take their servers to where the sun never sets. This issue is being raised directly with European governments in the form of a question. Not so much whether artificial intelligence can work in space, but whether space investments will translate into reliable services and verifiable benefits for Earth.

Recent developments in the US

The signs over recent months are clear. On 4 February 2026, the US Federal Communications Commission approved SpaceX’s application for a network of data centres comprising up to one million satellites, positioned at altitudes of between 500 and 2,000 kilometres. On 24 August, Nvidia announced that it was working on an optimised version of the Vera Rubin NVL72 platform for Starmind, the programme officially launched on 17 November 2026. On 30 October, Google announced that the Suncatcher prototype, in orbit since 1 October, had completed its first machine learning tests and returned a response generated by artificial intelligence.

Loading...

Taken together, these three announcements outline an industrial trajectory, not a scientific demonstration. And this is precisely why Europe cannot afford to stand idly by: funding capacity without having defined its uses means taking technological and commercial risks, whilst failing to experiment means leaving it to others to set the rules of the game.

Energy is the driving force behind everything

Energy is the driving force behind it all. In April 2026, the International Energy Agency estimated that data centres would consume 485 terawatt-hours in 2025 and predicted this figure would rise to around 950 by 2030. The strain on electricity grids explains the interest in orbital computing, but does not prove its viability. Power, heat dissipation and communications must be able to cope with a real-world application load. A system that analyses images on board a satellite and transmits only selected results to Earth has very different requirements from a cluster of accelerators that continuously exchange information during a distributed computation.

The environmental impact must also be assessed in its entirety, from launch to decommissioning. In June 2024, Thales Alenia Space reported that, according to the Ascend study’s assumptions, significant reductions in emissions would require a launcher with approximately ten times lower emissions over its life cycle. The absence of water consumption for in-orbit cooling remains one of the most compelling arguments, but it does not resolve the equation involving production, launch, replacement and disposal.

Robots, rovers and printers: the next frontier

The most challenging aspects are the technical ones. The most ambitious involves integrating computing with infrastructure that enables people to live and work beyond Earth. The Armadas experiment, described by NASA in January 2024, used three robots to assemble a structure several metres in size in a laboratory, comprising hundreds of prefabricated elements, in just over a hundred operating hours spread over several weeks.

On Mars, progress tells a different story. The NASA dated February 2026 on the autonomous positioning of Perseverance describes a rover that compares surface images and orbital images to determine its own position, reducing its reliance on Earth-based processing that could take a day or more.

Local production, moreover, should be measured in terms of usable stocks. In a design example published in 2020, NASA indicated approximately 25 tonnes of oxygen and seven tonnes of fuel for the launch of a four-person crew. Producing that quantity would require industrial capacity, as well as experimental validation. In July 2026, the ESA reported the production of the fifth sample using its metal 3D printer on board the International Space Station and the return of three samples for analysis on Earth. Full self-replication would also require the replication of critical control, measurement, maintenance and production capabilities: a capability that the experiments described do not demonstrate.

Iride, the Italian test bed

For Italia, this testbed has a name: Iride. It provides a practical platform on which to assess the value of satellite data processing before committing to more extensive infrastructure. On 1 October 2026, the Italian Space Agency announced the initial availability of 18 geospatial services and 110 products for the public administration, with applications in the fields of water, ground movements, emergencies and environmental monitoring. Developed by the ESA in collaboration with the ASI through the NRRP and the Complementary National Plan, the programme is gradually being transferred to Italian operational management, whilst Planetek Italia and e-GEOS coordinate their respective service segment consortia.

Italia also acts as a bridge between terrestrial applications and the development of space exploration. Thales Alenia Space in Italia has developed PhiFireAI, the fire classification application associated with Isat-2, whilst the Mph programme is addressing the issue of lunar habitation. In May 2026, ASI reported a favourable NASA assessment regarding progression to the Preliminary Design Review the following year, without presenting it as already completed. Industrial policy could bring these activities together through verifiable expertise in software, integration, inspection and maintenance, whilst preserving the varying degrees of maturity of the programmes. Public procurement contracts should specify the improvements they require and ensure a comparison with viable alternatives: companies would face more credible demand, whilst public authorities would limit purchases that lack the necessary procedures for their use.

Loading...

The choice facing local authorities

In the coming years, for public authorities using Iride, FireWatch and the Copernicus services, the choice will be a binary one: either make broader commitments contingent on agreed outcomes, or pass the still-unresolved uncertainty on to the public purse. Postponing testing and integration would leave users to bear the cost of fragmented information and slow procedures; funding capacity without demand or verified performance would shift the risk of under-utilisation and obsolescence onto taxpayers.

Starmind and Suncatcher will be able to justify further investment through continuity, clients and results; the lunar and Martian programmes through maintainable and repeatable physical functions. The common policy is to draw up contracts that allow us to move forward as the evidence improves and to change direction when another solution better meets the same need.

Copyright reserved ©
Loading...

Brand connect

Loading...

Newsletter

Notizie e approfondimenti sugli avvenimenti politici, economici e finanziari.

Iscriviti