Google's data center energy chief said this week that space-based data centers could cost up to 100 times more than ground-based ones. Google itself is testing AI server chips in orbit, and this statement poured a bucket of cold reality on the hot concept of space computing.
"Before we rush to space, we should first exhaust every viable option on the ground," said Amanda Peterson Corio at the WTF Summit, which brings together senior women from the technology and finance industries. Her remarks came as SpaceX's orbital data center plans serve as one of the key pillars supporting its roughly $1.8 trillion market valuation.
Peterson Corio also noted that the growing opposition to ground-based data centers in the United States is largely the responsibility of a group of inexperienced new companies, which should be subject to higher regulatory standards. She said that before this wave of AI computing fervor, companies that had spent years building data centers "operated in a fairly disciplined manner," and their projects rarely sparked controversy.
The launch cost of sending computing hardware into orbit is "extremely high, up to 100 times the cost of ground-based construction," Peterson Corio said. Beyond cost, operators also need to solve the challenges of maintaining and upgrading equipment in orbit. On the ground, data center staff can replace servers and rewire systems at any time.
Despite her cautious stance on space computing, Google's research team announced this month that it will partner with satellite imagery company Planet Labs to explore the feasibility of large-scale space data center clusters using satellites. Under a project called "Suncatcher," Google has already launched a small prototype space server into orbit, equipped with an AI tensor processing unit (TPU) custom-built with Broadcom.
Last autumn, Google researchers involved in the Suncatcher project, along with the company's chief researcher James Manyika, published a paper arguing that if launch costs fall to $200 per kilogram or below, the computing cost per unit of power in space could become roughly comparable to ground levels. Google researchers believe this threshold could be reached around 2035, though the timing depends on the development progress of SpaceX's Starship and the pace of increasing launch frequency.
But other companies are more optimistic. SpaceX CEO Elon Musk said the company could build a space data center as early as next year; Amazon founder and Blue Origin chief Jeff Bezos, whose company competes with SpaceX, predicted that a space data center with 1 gigawatt of power consumption will emerge within a decade. Other reports say Google and SpaceX have discussed plans to cooperate on building orbital data centers.
Peterson Corio did not completely dismiss the space computing concept, calling the direction "very interesting"; she said Google continues to pursue such high-risk frontier projects because the company is accustomed to "thinking 10, 20, 30 years ahead." But the technology's deployment is "by no means imminent."
OpenAI CEO Sam Altman was more blunt, saying that under current conditions, the idea of moving data centers to space is "quite impractical," and questioning the aggressive timeline given by Musk.
The real bottleneck: permits, not power
Regarding the argument put forward by Musk and others — that U.S. electricity supply is insufficient to support AI computing expansion — Peterson Corio offered a different view. "Is there enough power? Of course there is," she said.
The real bottleneck is the approval process and lengthy assessment studies required to connect projects to the grid. Alphabet's X lab has launched the Tapestry project, working with PJM, the largest grid operator on the U.S. East Coast, to use AI to accelerate the relevant assessment process. In Texas, state power regulators have slowed approvals for some data center projects, strengthening oversight to prevent new computing facilities from burdening the grid and driving up electricity prices.
Other industry executives also point out that the grid itself is far from being utilized at ideal levels. Brian Janous, co-founder of data center land and power services firm Cloverleaf Infrastructure, said in a recent interview that about 45% of grid capacity sits idle on average over the long term, reserved for peak demand. For the vast majority of the day, the grid has surplus capacity that could accommodate data center power needs.
It is worth noting that Google has an advantage over cloud peers on the power usage effectiveness (PUE) metric, which measures how efficiently a data center converts electricity into computing output. As for the rising anti-data-center wave in the United States, Peterson Corio attributed it to "a large number of new players entering the market who lack industry experience and maturity when acquiring land and building computing projects," though she did not name specific companies. In recent years, a flood of novice developers has poured into the AI data center sector, but it is not just new players — projects from well-known companies like Oracle have also faced resistance from local residents.
Higher industry standards
In any case, she supports introducing industry regulations and setting higher standards. She did not propose specific provisions, but cited examples of how Google is increasing investment in local communities where projects are located while putting contingency plans in place to avoid risks from abandoned projects. These include commitments to bear grid upgrade costs for 15 to 20 years, preventing local communities from being "forced to foot the bill" if a project stalls. This model of upfront commitments to cover grid infrastructure costs has already been implemented in more than nine U.S. states and is "expanding extremely fast."
Supporting standards also include: Google paying electricity bills and property taxes in full; posting bonds for land restoration if a project is abandoned; and canceling confidentiality agreements signed with local governments. Her remarks come against a backdrop of investors re-examining the valuations and viability of a group of upstart AI cloud vendors: most of these companies are heavily backed by Nvidia, rely on massive borrowing, and rush to build data centers despite lacking a long operating track record. The companies Peterson Corio says need tighter standards likely include such vendors.
Meanwhile, Google has also disclosed that its computing needs will increasingly rely on third-party computing resources, though it did not reveal specific partners. SpaceX is one of them. Google needs external computing capacity because it cannot build its own data centers fast enough to keep up with demand.
Comments