Google Sends Advanced Chips Into Orbit, and Estimates 1,800 Starship Launches for Space Data Centres
Google’s Project Suncatcher has put a prototype satellite carrying Tensor processors into orbit to test whether advanced chips can run in space. A new paper estimates that scaling up orbital data centres would require about 1,800 Starship launches over a decade.
October 2 (IT Home) — According to the foreign outlet TechCrunch, on the 1st local time, Google’s in-orbit computing prototype satellite lifted off from California aboard a SpaceX rocket, launching Project Suncatcher, which aims to build large-scale computing clusters in Earth orbit. The satellite was built by Planet Labs and will verify whether Google’s Tensor processors can operate normally in space. The test requires a continuous supply of 1 kW of power, cooling for the chips, and running a series of models to check for problems.
As IT Home understands it, this is the first time Google has sent advanced chips into space. Travis Beals, the Google executive in charge of the project, said: “We have done ground testing, but no amount of testing can fully match the actual environment.”
Through Project Suncatcher, Google hopes to build large-scale computing clusters in Earth orbit.
Once commissioning is complete, the processors on the satellite will run for 15 minutes at a time, to avoid overloading the power and thermal management systems. The prototype satellite uses Planet Labs’ standard platform, and the two companies are also preparing a demonstration mission expected to launch next year, using two specially designed satellites better suited to advanced computing to take on heavier computing workloads. Later, the satellites will also attempt to operate together using laser communication links.
Beals calls the mission “a long-term frontier exploration”: the focus is on preparing for future space infrastructure and AI computing tasks. Google envisions an in-orbit data centre made up of 81 satellites flying in a tight formation and processing computing tasks in parallel. “When you run computing tasks across multiple racks, the bandwidth and latency between Tensor processors become very, very important... We want to look ahead, not only at what computing tasks exist today, but at what will exist in five years.”
The report notes that Google is looking so far ahead largely because the rockets needed to scale up in-orbit data centres at a reasonable cost do not yet exist.
On the 1st local time, Google also published a peer-reviewed version of its in-orbit data centre white paper. Among existing studies examining how to deploy computing power into orbit, the paper is one of the most rigorous analyses, and will appear in the journal Joule.
One highlight of the paper is Google’s assessment of the cost of getting into space. The researchers stress that the paper does not examine economic feasibility, but Google’s prediction of how rocket costs will gradually fall is still worth a look. Like every data centre company, Google is counting on SpaceX to carry spacecraft into space.
The paper argues that, starting with the launch of the Falcon 1 rocket, Musk’s rocket-building team has followed a “learning curve” that delivered price reductions of about 20% a year. It is therefore reasonable to expect SpaceX to bring launch prices down to close to US$200 per kilogram (about 1,343 yuan at current exchange rates) by 2035.
Based on the cumulative payload launched by Falcon 9, Starship would need to deliver 370,000 tonnes of payload to orbit to reduce costs along a similar path. Over the next 10 years that would require roughly 1,800 launches — about 180 a year — each carrying 200 tonnes of payload.
Starship has never launched more than five times in a year, so reaching that frequency would be a tall order. SpaceX expects future launch rates to far exceed that level. Musk has even claimed that Starship will be able to launch once an hour by 2029.

评论
0 条讨论
登录后发表评论
立即登录暂无评论
成为第一个分享想法的人吧!