{"id":5854,"date":"2025-12-31T08:59:06","date_gmt":"2025-12-31T15:59:06","guid":{"rendered":"https:\/\/novus2.com\/righteouscause\/?p=5854"},"modified":"2025-12-31T16:40:28","modified_gmt":"2025-12-31T23:40:28","slug":"has-beijing-has-pulled-into-an-early-lead-in-the-contest-to-build-the-first-true-space-supercomputer","status":"publish","type":"post","link":"https:\/\/novus2.com\/righteouscause\/2025\/12\/31\/has-beijing-has-pulled-into-an-early-lead-in-the-contest-to-build-the-first-true-space-supercomputer\/","title":{"rendered":"Has Beijing pulled into an early lead in the contest to build the first true \u201cspace supercomputer?"},"content":{"rendered":"
\r\n
\r\n \"\"<\/a>\r\n <\/div>\r\n <\/div>\n
\n

China is no longer just racing to dominate AI on Earth; it is now openly laying the foundations for a supercomputer network in orbit that could redefine who controls the world\u2019s computing power. While the United States is experimenting with similar concepts, a cluster of recent Chinese launches, policy moves, and industrial consortia suggests that Beijing has pulled ahead in the contest to build the first true \u201cspace supercomputer.\u201d\u200b<\/span><\/p>\n

What China Is Actually Building<\/strong><\/span><\/h2>\n

China\u2019s vision goes far beyond a few smart satellites.<\/strong><\/em><\/span><\/h3>\n

Beijing municipal authorities have endorsed plans for a centralized space<\/strong> data center in a dawn\u2013dusk low-Earth orbit roughly 700\u2013800 kilometers above the planet, with an eventual power capacity above one gigawatt and the ability to host millions of server cards distributed across multiple space \u201csub\u2011centers.\u201d Officials describe a three-phase buildout, with an experimental satellite already developed and slated for launch by late 2025 or early 2026, and a large-scale orbital data center targeted for completion by 2035.\u200b<\/span><\/p>\n

In parallel, Chinese commercial and state-linked firms are already assembling the computing building blocks.
\nA collaboration between Guoxing Aerospace and Zhejiang Lab has placed 12 AI-enabled satellites into low-Earth orbit, forming what is being billed as the first in\u2011space \u201ccomputing constellation,\u201d capable of 5 peta operations per second and running an 8\u2011billion\u2011parameter model directly in orbit to process Earth-observation and other data in real time. Zhongke Tiansuan, a spin\u2011off from the Chinese Academy of Sciences\u2019 Institute of Computing Technology, has kept an upgraded space computer operating on a satellite for more than 1,000 days, giving engineers rare long\u2011duration experience with advanced chips in the radiation-soaked conditions of orbit.\u200b<\/span><\/p>\n

How the US (and Big Tech) Are Responding<\/strong><\/span><\/h2>\n

If China\u2019s play is state-led and centrally framed as a strategic project, the American answer so far looks like a loose coalition of tech titans and startups experimenting at the edge of feasibility.
\nCompanies tied to Elon Musk\u2019s Starlink, Jeff Bezos\u2019 Blue Origin, and Google are all testing ways to mount high-performance accelerators on satellites and link them with laser-based free-space optical networks robust enough to serve as a distributed AI data center in orbit. Google\u2019s internal \u201cProject Suncatcher\u201d is exploring constellations of solar-powered satellites equipped with TPUs that could be scaled up if early trials prove technically and economically viable.\u200b<\/span><\/p>\n

The most aggressive American move so far may be coming from a small player.<\/strong><\/em><\/span><\/h3>\n

Startup Starcloud, backed by Nvidia, has launched its Starcloud\u20111 satellite carrying an Nvidia H100 GPU\u2014an 80\u2011gigabyte accelerator roughly 100 times more powerful than any chip previously flown in space\u2014and used it to train a lightweight NanoGPT model from orbit, then run Google\u2019s Gemma model in space, marking the first time a modern large language model has been powered by an in\u2011orbit GPU. Starcloud\u2019s own white paper argues that \u201cgigawatt-scale orbital centers\u201d are not science fiction but a necessary next step if AI is to grow without blowing through planetary limits on water, energy, and emissions.\u200b<\/span><\/p>\n

Why Put a Supercomputer in Space at All?<\/strong><\/span><\/h2>\n

At first glance, putting racks of fragile high-end processors into orbit\u2014where launch vibrations, vacuum, extreme temperature swings, and charged particles all conspire to kill electronics\u2014sounds like a wildly inefficient way to run AI. But three converging pressures are driving both China and the US toward the same improbable answer:\u200b<\/span><\/p>\n