The Race to Build Data Centres in Space

The Race to Build Data Centres in Space THE BELAMY Weekly Newsletter of AIM Monday, Nov 10, 2025 | By Siddharth Jindal Now, subscribe to our Digital & Print Editions > Until recently, data centres in space sounded like a wild TED Talk fever dream. But here we are. It's 2025, and the wild idea just got real. Starcloud, a startup backed by NVIDIA's Inception programme, is preparing to launch the first-ever AI-equipped data centre into orbit this November. The mission will mark the NVIDIA H100 GPU's debut in space—and, possibly, the start of a new era where AI literally runs above our heads. On Earth, data centres devour water and electricity like never before. Cooling towers gulp millions of litres daily and power bills stretch into billions. But in space? There's no water, no weather and infinite sunlight. Starcloud's upcoming five-gigawatt orbital data centre, fitted with solar panels stretching four kilometres wide, could run on near-free, renewable energy—forever. "In space, you get almost unlimited, low-cost renewable energy," said Philip Johnston, cofounder and CEO of the startup. "The only environmental cost is the launch. After that, you're saving 10x the carbon emissions of a terrestrial data centre." That's not marketing fluff. Starcloud's first satellite, a 60-kilogramme mini-fridge-sized unit called Starcloud-1, will offer 100x the GPU power of any existing satellite system. Built for heavy inference workloads—Earth observation, wildfire detection, disaster response—Starcloud's AI will crunch data in orbit, right where it's captured. No waiting, no downlink delays, just real-time insights beamed back home. Starcloud even plans to run Google's Gemma model in orbit on NVIDIA's H100 GPUs, testing the idea that even large language models can operate off-planet. Google Wants to Train AI Where Sun Never Sets Now, Google is taking that idea further. It turns out the search giant has bigger ambitions for space itself. This week, Google announced Project Suncatcher—a new research initiative exploring whether AI compute can be scaled in space, using solar-powered satellite constellations equipped with TPUs. It is basically 'Jadoo' from the cult film Koi Mil Gaya, soaking up sunlight to get supercharged and do mind-blowing feats. "Inspired by our history of moonshots, from quantum computing to autonomous driving, Project Suncatcher is exploring how we could one day build scalable ML compute systems in space, harnessing more of the Sun's power (which emits more power than 100 trillion times humanity's total electricity production)," said Google CEO Sundar Pichai. The company's researchers have already tested the limits of what's possible. In the lab, they've achieved 1.6 terabits per second optical transmission using a single transceiver pair—the kind of bandwidth that could make in-orbit model training or inference viable. Each Suncatcher satellite would orbit just a few hundred meters from its neighbours, close enough for multi-terabit interlinks yet stable enough to avoid frequent manoeuvring. And remarkably, Google's Trillium v6e Cloud TPU chips have passed radiation tests under a 67 MeV proton beam, surviving nearly three times the expected five-year dose. If that sounds like science fiction, Google's economics team has done the math: by the mid-2030s, when launch costs drop below $200 per kilogramme, the company estimates space-based AI compute could rival Earth's data centres in cost. Elon Musk Enters the Chat And of course, no space tech story is complete without Elon Musk jumping in.  Musk has a new plan for the cloud, and it's not on Earth. The SpaceX CEO said the company could transform its Starlink satellites into fully fledged data centres in orbit, bringing computing power closer to the stars. Musk revealed that SpaceX's next-generation Starlink V3 satellites could be scaled up to act as the foundation for orbital compute systems. Built for gigabit-level Internet speeds, they use laser-based inter-satellite links to transfer data across orbit—no ground relay needed. Each unit is larger, smarter and more powerful, designed to form a mesh of high-performance nodes that circle the planet. They'll hitch a ride aboard Starship, SpaceX's still-in-development super-heavy rocket, which is expected to deploy dozens of V3 satellites in a single launch. If timelines hold, the first batch could be in orbit by early 2026. Amazon's Jeff Bezos also predicted that within the next 10 to 20 years, gigawatt-scale data centres will be constructed in orbit. "These giant training clusters, those will be better built in space, because we have solar power there, 24/7. There are no clouds and no rain, no weather," he said. [AIM Network Exclusive] In a recent episode of the Simulated Reality podcast, host Sandeep Sharma discusses the solution to solving the Digital Shelf dilemma with Debobroto Ghosh, VP of products and IPs at C5i. Ghosh unpacks how C5i's Compete platform is giving e-commerce an AI-powered makeover. [Webinar Alert] AIM x Polestar Innovator Session: Are You Agent-Ready? Join Ankit Rana, chief innovation officer at Polestar Analytics, and Siddharth Poddar, chief product officer at the company, as they break down what it truly means to be "agent-ready". Click here to know more. Join us at the Dentsu Global Services Developers' Meet Up 2025 to explore this powerful synergy that's reshaping innovation. Engage with industry leaders, discover cutting-edge tools and gain insights into how this convergence is powering the next era of digital transformation. 📍 November 22 – Bengaluru 📍 December 6 – Pune   AI Projects Can't Run like R&D Experiments in Pharma AI promises faster drug discovery, smarter clinical trials and more personalised patient care. But as regulators like the FDA roll out detailed guidance on AI and machine-learning-based medical devices, the industry faces a sharper question: how to innovate at speed without tripping over compliance? According to Manish Mittal, managing principal and India business head at Axtria, the answer lies in embedding compliance into the DNA of AI programmes rather than treating it as an afterthought. Read the full story here. The List Doesn't END HERE  Turns out, everyone wants a place in the orbit. Axiom Space, best known for building commercial space station modules, is developing Orbital Data Centre nodes in partnership with Red Hat, testing how cloud workloads can run in low Earth orbit. The company's first Data Centre Unit 1 (AxDCU-1) will operate aboard the International Space Station, supporting both commercial and national-security customers. Meanwhile, Seattle-based Sophia Space recently raised $3.5 million to develop orbital computing clusters—small, modular data nodes designed to operate as a mesh network in space. That said, back on home turf, Hyderabad-based TakeMe2Space is building what it calls the world's first open low Earth orbit (LEO) satellite infrastructure—an effort to make space computing accessible to everyone, not just deep-pocketed institutions. The company's Indian-built satellites function as miniature computers in orbit, offering an open, AI-enabled platform for developers and researchers to run real-time applications directly in space. While others look to orbit, Lonestar Data Holdings is thinking even bigger. In March, the Florida-based company said it had tested a small data centre on the Moon, calling the mission a success even though the device toppled sideways and shut down prematurely. Madari Space in Abu Dhabi is partnering with Thales Alenia Space to launch compute demonstration satellites that will handle on-orbit data processing for Earth observation and other mission-critical applications. In a sense, the evolution of cloud computing has always been about distance. We moved from servers under our desks to data centres halfway across the world. The future of computing may not be underground, underwater or even at the edge—it might just be above us, glinting faintly in the sunlight. Now, subscribe to our Digital & Print Editions > For Brand collaborations, reply to this email or write to info@aim.media You received this email because you signed up to the updates from AIM. Click here to unsubscribe if you do not want to receive emails from us.