Google’s orbital AI data centres hinge on 1,800 Starship launches
Google has sent its first orbital computing prototype into space to test whether its Tensor Processing Unit can operate reliably there. The experiment is an early step towards Project Suncatcher, a proposed network of satellites that could run AI workloads in orbit, but building such data centres at scale depends on much cheaper launches.
The prototype, built by Planet Labs, will test power, cooling and chip performance, initially running the TPU in 15-minute bursts. Google envisages 81 satellites working together and plans a two-satellite communications demonstration next year. A new peer-reviewed paper estimates that launch costs could approach $200 per kilogram by 2035; achieving that trajectory would require SpaceX’s Starship to launch about 370,000 tonnes of payload, or roughly 1,800 flights. The headline’s figure of 1,600 differs from the article text.
- Google has launched its first orbital computing prototype.
- Project Suncatcher envisages 81 satellites processing workloads together.
- Google estimates Starship may need about 1,800 launches to cut costs.
New here? Start with this
Google is exploring whether it could run artificial intelligence computing from satellites orbiting Earth instead of from traditional data centres on the ground. This concept could potentially offer advantages in power efficiency and data transmission speed, though it remains highly experimental and faces major technical and economic challenges.
Launching and maintaining such a satellite network would be extraordinarily expensive with current technology and costs. Researchers estimate that making such a system economically viable would require launch costs to fall substantially, potentially needing thousands of individual rocket flights to deploy a full-scale network.
To establish whether the concept is actually workable, the company is conducting early-stage tests to see if its specialised artificial intelligence computing chips can operate reliably in space. These experiments are the first step in a longer-term research initiative exploring the feasibility of processing data from orbit.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Orbital data centres represent a genuine technological frontier offering unprecedented computing capabilities through space-based solar power and ultra-low latency services for global applications. SpaceX has demonstrated credible cost reduction trajectories through engineering innovation, making the required launch cadence ambitious but plausible rather than implausible. The potential gains in computing efficiency and entirely new capabilities justify the scale of investment and development required.
The case against
The scale of 1,800 launches creates significant space debris and collision risks that could jeopardise all orbital activities for decades, raising serious concerns about long-term orbital sustainability. Terrestrial data centres powered by renewable energy already provide mature, proven, and cost-effective computing solutions that demonstrably meet existing needs without these risks. The astronomical costs and technical complexity of orbital infrastructure remain difficult to justify when established alternatives offer superior reliability, security, and environmental accountability.
Celebrity Entertainment Space Technology
Read the full article at the source →
Originally published by TechCrunch as “Google thinks SpaceX’s Starship has to launch 1,600 times before space data centers get off the ground”.