Elon Musk responds to Google launching TPU processors into space for AI


Space is increasingly emerging as a potential new frontier for artificial intelligence. Google has announced plans to send a satellite equipped with its advanced Tensor Processing Units (TPUs) into space next week as it explores whether AI infrastructure could eventually be hosted in orbit. The company is working with Elon Musk’s SpaceX for the mission.

In a blog post, Google said it would launch a prototype satellite carrying its TPUs into low Earth orbit next week under Project Suncatcher. The initiative is a long-term research programme aimed at determining whether space could eventually support large-scale AI computing infrastructure.

Google CEO Sundar Pichai announced the upcoming launch on X with a reference to Neil Armstrong’s famous words from the Moon landing, writing, “One small step for TPUs....”

The first orbital test will be part of SpaceX’s upcoming Transporter-18 rideshare mission, in partnership with Planet Labs. The mission will examine whether Google’s AI hardware can withstand the forces of launch and continue functioning amid the radiation and extreme temperatures found in space.

Elon Musk reacts to Google’s AI plans for space

Elon Musk, one of the strongest advocates for building data centres in space, responded to Google’s announcement on X. He suggested that computing resources could eventually be overwhelmingly located in space.

“The amount of compute in space will obviously round up to 100 per cent of all compute,” Musk wrote.

Google has stressed that this launch is an early-stage experiment intended to determine whether the concept is technically viable over the long term. It is not being presented as the beginning of an operational orbital data centre.

Terrestrial data centres require enormous amounts of electricity and water, with companies consuming billions of litres of water every year for cooling their hardware. Satellites in low Earth orbit, meanwhile, can receive sunlight for much longer periods and potentially generate up to eight times more solar power than equivalent systems on Earth.

This has made orbital computing an attractive possibility for companies such as Musk, who argue that moving AI infrastructure into space could enable significantly greater computing capacity without placing additional pressure on Earth’s resources.

“Global electricity demand for AI simply cannot be met with terrestrial solutions, even in the near term, without imposing hardship on communities and the environment,” Musk said earlier this year. “In the long term, space-based AI is obviously the only way to scale.”

Google’s plans for first TPU flight to space

Google’s first objective is to determine whether its AI hardware can function reliably in the space environment. A typical rocket journey to low Earth orbit takes around 10 minutes, during which spacecraft experience intense vibrations and acceleration forces that can reach up to 10 times the force of gravity, or 10g.

Individual components such as TPU chips can experience forces of between 50g and 100g. To prepare for these conditions, Google subjected its satellite to vibration testing along all three axes to replicate the conditions experienced during launch.

Radiation presents another major challenge in space. Solar activity and cosmic rays can interfere with electronic systems and trigger errors. Google tested its TPUs while running AI workloads at UC Davis’ Crocker Nuclear Laboratory to examine how they respond to radiation.

According to Google, early testing indicated that its Trillium TPUs could tolerate a total ionising radiation dose greater than the amount they are expected to encounter during a five-year mission in space.

However, the company noted that certain conditions cannot be accurately replicated on Earth, making an actual orbital test essential for collecting real-world flight data.

The mission will also help Google investigate the problem of cooling AI hardware in space. TPUs produce significant amounts of heat within a relatively small area, but satellites cannot rely on airflow to dissipate that heat as data centres on Earth do.

Instead, Google is testing a cooling system based on heat pipes and radiators. The company has already evaluated the technology inside a thermal vacuum chamber designed to recreate the vacuum and temperature conditions found in space.

Project Suncatcher is ultimately intended to go beyond a single satellite. Future versions could consist of groups of satellites, with each spacecraft carrying dozens of TPU chips. These satellites could potentially be connected to work together on larger AI workloads while in orbit.

For such a system to operate effectively, satellites would need extremely precise knowledge of their positions relative to one another and the ability to exchange large amounts of data through high-bandwidth laser links over short distances.

Google plans to test this part of the technology in 2027, when it aims to send two satellites into orbit.

Google is not the only company exploring the possibility of putting AI computing infrastructure in space. Alongside Project Suncatcher and Musk’s plans for orbital data centres, Jeff Bezos’ Blue Origin has also explored similar concepts, including plans for data centres in lunar orbit.

India-based AI company Sarvam has announced plans to develop orbital data centres in partnership with aerospace company Pixxel. Meanwhile, Starcloud, which is working with Nvidia, is also developing plans to launch satellites equipped with advanced GPUs into Earth orbit.


 

buttons=(Accept !) days=(20)

Our website uses cookies to enhance your experience. Learn More
Accept !