A prototype of the Project Suncatcher satellite at the Planet Labs facility in San Francisco, California.
Planet Labs PBC
hide caption
Planet Labs PBC
Google has just placed a refrigerator-sized satellite in space, part of a research project that the company hopes will pave the way for orbiting AI data centers powered by the sun.
The satellite prototype contains four specialized chips called Tensor Processing Units, or TPUs, that Google designed for machine learning and which have already been deployed in data centers on Earth. The chips will run a version of the company’s Gemma AI model for 15 minutes at a time due to heat management constraints, using the open weight model to respond to simple queries. In a blog postGoogle said the mission would evaluate how TPUs handle “the physical stress of spaceflight and the radiation and extreme temperatures of space.”
The use of AI has increased sharply in recent years, fueling demand for data centers. At Google’s annual developer summit in May, CEO Sundar Pichai said demand for AI services is outpacing supply and he expects capital spending this year to be between $180 billion and $190 billion — more than six times that of 2022 — as the company builds computing infrastructure and develops chips.
At a time when opposition to the proliferation of energy-hungry data centers on Earth is growthGoogle is one of a handful of companies proposing to put them into orbit where there is virtually unlimited free energy and no protesters.
“The sun produces almost all of the energy in our solar system. All other energy sources harnessed by humanity represent only a tiny fraction of a percent,” said Travis Beals, principal director and manager of the Suncatcher project. “So in some sense this project is about harnessing the best way to use solar energy to run AI computing.”
The satellite will be in a sun-synchronous orbit where its solar panels will almost never be in the shade. This eliminates the need to carry heavy batteries or backup power on the spacecraft.
Last November, a venture-backed startup called Starcloud launched a spaceship which featured an Nvidia H100 chip and demonstrated a version of Google’s Gemini AI from space. Elon Musk’s SpaceX is also working on space data centers and has said it plans to begin deploying “orbital AI computing satellites” as early as 2028.
Google, in partnership with aerospace and satellite imaging company Planet, is considering groups of satellites that could eventually orbit together to form space data centers. Each satellite will carry dozens of TPU chips and communicate with the others – and with Earth – via lasers. Next year, the company plans to put two more satellites into orbit to test their connection.
The satellite launched today is a prototype that Beals called a “very minimal test” to ensure the chips can work in space. A SpaceX rocket put it into orbit and Planet will operate it. Beal said Google will launch and test TPUs next. The goal is for the mission to be operational for a year.
The satellite prototype was tested in a thermal vacuum chamber that simulates the space environment.
hide caption
Brandon Lucia, a professor of electrical and computer engineering at Carnegie Mellon University, said the idea sounds “very science fiction” but that much of the technology has been proven at the research level.
However, enormous challenges remain. Heat is a major problem. Computer chips heat up, and on Earth, heat can be dissipated by wind or water. This is not possible in the vacuum of space.
“In a particular satellite, using more power to perform the calculations means rejecting more heat into the confined environment inside the satellite,” Lucia said.
In its blog post, Google called Cooling chips is “a crucial research challenge” and described some of the approaches he is working on, including using a combination of pipes and radiators to carry heat away from chips.
Beal said radiators are one of the heaviest components on the current mission — a disadvantage because heavier objects are more expensive to launch — and his team is working on ways to cool the chips more efficiently.
Lucia points out that there are also important questions regarding the interview. When a chip fails in a terrestrial data center or something goes wrong, a technician can come in and fix it. It’s more difficult when the data center is running hundreds of miles above the Earth.
“All of this is seeing its cost and complexity increase by a factor of 10, maybe a factor of 100. So there must be a big win,” Lucia said.
And there are also environmental issues. While space data centers would run on solar power, the rockets that launch them run on carbon-based fuels. And eventually, they will re-enter Earth’s atmosphere, potentially adding pollution as they burn.
Elon Musk said this space is the “only way to scale” data centers to meet future AI demand. But the economic aspects remain a question mark. Launches are expensive, repairs are difficult, and laser communication over hundreds of miles and through Earth’s atmosphere is slower. And it will take constellations of many satellites to match the computing capacity of a terrestrial data center.
Beals said he and his small team think a lot about these questions. “It doesn’t help much if it’s technically possible, if it’s always going to be too expensive to be practical,” he said.
He added that it is difficult to know when space data centers will be economically viable. “I don’t think it will cost less over the next five years. I think it will take longer than that,” he said.
And he doesn’t think space data centers will replace terrestrial data centers anytime soon. “I think there will be use cases that make more sense to do on Earth and use cases that make more sense to do in space,” Beals said. “In the very long term, maybe the set of things you do in space will get bigger and bigger until that’s where most of the computing takes place. But I think for a long time the two will co-exist.”
The whole project, he said, is a “moonshot.”
“It’s going to take a long time,” he said. “It won’t be an entirely simple journey.”
Google financially supports NPR.
Gn bussni
