The idea of constructing AI data centers in space is starting to gain traction, notably after SpaceX’s recent IPO that raised over $85 billion and valued the company in the trillions. This success, coupled with the backlash against such data centers in rural areas, has prompted competitors like Jeff Bezos to explore their own space-based plans.
According to CNBC, SpaceX’s successful public launch, which secured $85 billion and positioned Elon Musk as the world’s first billionaire to achieve this, has made the once-ambitious concept of orbital AI data centers seem more achievable. While doubts persist about the economic feasibility of this idea, some experts believe that significant capital could expedite the development of necessary infrastructure.
SpaceX already possesses many essential elements for a space-based data center, such as reusable Falcon rockets, plans for advanced launch vehicles, an xAI division with considerable computational needs, and an upgradable Starlink satellite network. With newfound capital flowing into its numerous ventures, the company is potentially well-equipped to integrate these components and serve both internal AI initiatives and commercial clients like Anthropic.
Previously reported by Breitbart News, Musk hinted at plans for an orbital AI data center shortly before SpaceX’s IPO.
The proposed satellite, which Musk described as an initial version of the eventual design, would stand 20 meters tall and have an impressive wingspan of 70 meters, marking it as SpaceX’s largest satellite endeavor. This satellite will include a rack of AI chips paired with large solar panels and liquid radiators for managing heat.
Musk clarified that the current design is a preliminary draft of what will become the production model, asserting that it is less complex compared to SpaceX’s Starlink satellites, which have been operational since 2019. Much of the technology utilized is derived from existing setups used for satellite internet services.
“I don’t think this is a very difficult problem compared to what we’re already working on,” Musk noted.
Duncan Davidson, a partner at Bullpen Capital, spoke on CNBC’s The Exchange, saying, “The company is going to get to data centers in space. That’s a big, long-term commitment,” while acknowledging that although engineering challenges are being met, the current economic landscape remains limiting. He expressed optimism about the long-term viability as launch costs are expected to drop while terrestrial data center expenses rise.
The proposed AI1 satellite is meant to build upon the existing Starlink communications satellite but will demand significantly more semiconductors. In response to this demand, SpaceX has joined forces with Tesla and Intel to establish Terafab, a 10 million square foot manufacturing facility planned to open in Austin by 2029, with projected costs reaching up to $119 billion.
The quest to create computing capabilities in space is not exclusive to SpaceX. Jeff Bezos has similar ambitions for his rocket company Blue Origin, as well as his AI project Prometheus. In a recent CNBC interview, Bezos remarked that building data centers in space is “very realistic,” yet he expressed skepticism over competitors’ aggressive timelines. “Some of the timelines we hear are very short. People talk about two or three years,” he stated. “That’s probably a little ambitious.”
Blue Origin has proposed Project Sunrise to the FCC, aiming to launch 51,600 data center satellites into low Earth orbit as part of the TeraWave constellation with plans to commence in late 2027.
Google has joined the competition, collaborating with satellite manufacturer Planet Labs on Project Suncatcher. This initiative will investigate an interconnected network of solar-powered satellites using Google’s Tensor Processing Unit AI chips. Research indicates that operational orbital data centers could match the economic viability of ground-based centers on a per-kilowatt basis if launch expenses dip below $200 per kilogram by the mid-2030s.
Several startups are also venturing into space-based computing. StarCloud is currently testing Nvidia H100 GPUs aboard a SpaceX Falcon 9 rocket. CEO Will Marshall mentioned to CNBC that putting resources in space is simply more economical, especially since they won’t compete with terrestrial communities for water and power. The backlash against AI data centers over their resource demands has been significant, making orbital facilities appealing as they use radiation cooling techniques due to the cold void of space.
Other companies are also working on addressing the energy needs of orbital data centers. Rendezvous Robotics is developing a modular spacecraft system that can autonomously assemble in orbit to supply power. The system consists of hundreds of interlinked hexagonal tiles, each around five feet across. Company president Joe Landon shared successful test results from Blue Origin’s New Shepard flight and the International Space Station, with another test planned for later this year. “We will be able to offer a full-scale system in 2028,” he stated.
Rocket Lab is also entering the scene, having launched nearly 90 satellites for NASA and others, and is creating a more powerful reusable rocket named Neutron to vie in the orbital data center market. “Rather than helping customers build their own infrastructure, we’d rather turn them into tenants of the infrastructure we own,” CFO Adam Spice commented.
Despite the massive investments and intense activity, fundamental economic questions remain. Mark Weinziel, a Harvard economist who focuses on space ventures, pointed out that current business models have yet to show cost competitiveness with their Earth-based counterparts. “One of the biggest questions is: Are we sure we can’t do it cheaper on Earth?” he remarked.
Nonetheless, Weinziel envisions a scenario where Earth-based costs rise while space-related expenses decrease, potentially making orbital data centers a viable option. This outlook relies on ongoing trends of decreasing launch costs and technological advancements in satellites, though he did note that improvements in chip efficiency might alter these expectations.
Increasing opposition to terrestrial data centers could sway opinions towards space alternatives. Over 100 proposals to halt data center projects have been introduced at various government levels due to growing public concern. A recent Heatmap News poll revealed that 70% of Americans disapproved of having data centers near their homes, an increase from 40% just a year prior. A climate analysis by First Street indicated that 79% of data center capacities face escalating risks from climate change.
Weinziel views the concept of space-based data centers as a calculated risk rather than a definitive solution, yet not an unreasonable idea. “If we’re optimistic that the cost declines we’ve observed with advances in launch and satellite tech and solar power will persist, we might reach that tipping point sooner. It’s always a gamble, but that seems reasonable to me,” he added.
AI data centers are becoming a significant focal point in political discussions across the nation. A recent bestseller from Wynton Hall, Code Red: Left, Right, China, and the Race to Control AI, explores how the MAGA movement can approach AI to benefit humanity without conceding control to tech giants or allowing China to dominate globally.
read more Click here for CNBC.
