Space computing startup StarOrigin closes seed and angel rounds, bringing GPU power to satellites | Unity Ventures portfolio
The team was assembled from core contributors to the Beidou-3 program and experts from Harbin Institute of Technology, with experience in national-level satellite engineering projects.
Space-based computing satellite system solution provider Interstellar Origin recently completed two consecutive funding rounds — a seed round and an angel round. Unity Ventures was the sole investor in the seed round and doubled down in the angel round.
Interstellar Origin focuses on space-based computing satellites, moving GPU computing boards from ground data centers onto satellites to achieve large-scale in-orbit data processing and inference application deployment. Users on the ground can access satellite-based computing power and related services anytime through mobile phones or other terminals.
The core team was jointly assembled by core members of the Chinese Academy of Sciences' BeiDou-3 program and experts from Harbin Institute of Technology, with national-level satellite engineering experience spanning more than 10 BeiDou-3 medium/high-orbit satellites and over 30 low-orbit internet satellites.
Interstellar Origin has partnered with industry-leading enterprises to advance R&D, and has already made breakthroughs in key products including high-performance computing payloads, low-cost large flexible solar arrays, and active thermal control systems using fluid loops with deployable radiators.
A new player has entered the space-based computing race.
36Kr has learned that Interstellar Origin Aerospace Technology (Shanghai) Co., Ltd. (hereinafter referred to as "Interstellar Origin") recently completed two consecutive funding rounds — seed and angel — with total proceeds in the tens of millions of RMB.
Unity Ventures invested in the seed round. The angel round was jointly led by returning investor Unity Ventures, along with Plum Ventures and Shanghai Science and Technology Innovation Group's Ceyuan Fund (under Shanghai International Capital), with Shanghai Angel Association participating as a co-investor. Yiwei Capital served as exclusive financial advisor.
Founded in 2025, Interstellar Origin focuses on space-based computing satellites. Its core team was jointly assembled by core members of the Chinese Academy of Sciences' BeiDou-3 program and Harbin Institute of Technology experts, with national-level satellite engineering experience spanning more than 10 BeiDou-3 medium/high-orbit satellites and over 30 low-orbit internet satellites. Founder Sun Xiaolei has over 15 years of aerospace engineering experience, having served as chief satellite designer and chief engineer at the Chinese Academy of Sciences' Innovation Academy for Microsatellites and a leading commercial aerospace company.
BeiDou-3 is the first independently developed, globally networked satellite navigation constellation in China's aerospace history. Its 30 satellites operate cooperatively across high and medium Earth orbits. From the project's formal launch in 2009 to the completion of global network services in 2020, the entire endeavor took 11 years.
Sun Xiaolei's team at the Chinese Academy of Sciences at the time undertook the development of nearly half of these satellites. "The Innovation Academy for Microsatellites differs from other research-focused institutes — our team was built for satellite engineering. Plus, the project introduced competitive mechanisms, so everyone was pushing on technology and cost, which helped build out an entirely new supply chain."
The BeiDou-3 development experience represents the ability to deliver a national-level constellation on time and to quality with extremely limited resources under extraordinarily demanding conditions. This is precisely what Interstellar Origin's early investors value most in the team.
In 2023, Sun Xiaolei left his government-affiliated position to become chief satellite engineer at a commercial satellite company, while continuing to track innovation directions in the aerospace sector.
The turning point came in the second half of 2025, when Elon Musk's Starlink revenue exceeded $11.3 billion, with 60% as profit, validating a mature commercial闭环. In December, Musk posted repeatedly on social media with a core judgment: within the next 36 months, the cheapest place to deploy AI computing power won't be on Earth — it will be in space. The concept of space-based computing instantly ignited markets at home and abroad.
This is an extraordinarily ambitious narrative, facing challenge after challenge both technically and economically. But Sun Xiaolei believes the direction is worth pursuing — "the aerospace field needs people willing to innovate."
More importantly, it's precisely because it's difficult that the opportunity is greater. "We're sufficiently professional at building satellites, and the AI track has sufficiently vast market space." After rigorous analysis, Sun Xiaolei concluded that building large-scale space-based computing infrastructure is already achievable from the perspective of existing technical accumulation and engineering feasibility.
From its founding, Interstellar Origin decided to focus its core business on space-based computing satellites. Simply put, this means moving GPU computing boards from ground data centers onto satellites to deploy and run large language models in space, allowing users on the ground to access satellite-based computing power and related services anytime through mobile phones or other terminals.
Interstellar Origin's flagship product, the "Xuanji-1", is a 500 kg-class low-orbit satellite with 7–10 kW power capacity, capable of carrying 9–16 computing payloads. The preliminary satellite design has been completed.
According to Sun Xiaolei, over the past six months, leveraging the team's accumulated research resources at Harbin Institute of Technology and the Chinese Academy of Sciences, Interstellar Origin has partnered with industry-leading enterprises to advance R&D, and has already made breakthroughs in key products including high-performance computing payloads, low-cost large flexible solar arrays, and active thermal control systems using fluid loops with deployable radiators.
"We're building a data center in space," Sun Xiaolei said. "How we operate it in the future can draw from mature ground-based models — right now, the core is to build the data center well and drive costs down."
Below is an edited excerpt of 36Kr's interview with Interstellar Origin founder Sun Xiaolei:
What does the BeiDou-3 experience mean?
Sun Xiaolei: This was our country's first large-scale networked constellation. Before BeiDou-3, domestic satellite engineering generally followed the traditional path of "single-satellite development, single-satellite deployment." BeiDou-3 was the first to achieve integrated networked design and systematic construction across 30 satellites, breaking the previous single-satellite paradigm.
During BeiDou-3's construction, our team, with a relatively lean staffing configuration, undertook the development work for nearly half of the constellation satellites.
As healthy competition mechanisms took hold across the industry, all participating units actively pushed on technology iteration and process efficiency. This not only achieved reasonable optimization of overall satellite costs, but gradually cultivated a new, fully comprehensive supply chain with extremely high self-sufficiency — that hard-fighting cohesion was the precious combat effectiveness our team developed.
What exactly is the 10 PFLOPS computing satellite you're building? How does it differ from other computing satellites currently on the market?
Sun Xiaolei: Once this satellite is in orbit, it won't need to transmit data back to Earth — it can run lightweight large models directly in space. This isn't edge computing.
But this 10 PFLOPS figure wasn't pulled out of thin air. It's technically achievable now — we've already completed preliminary designs with our partners.
However, getting from 10 PFLOPS to 100 PFLOPS requires solving numerous technical challenges: flexible radiators, large-inertia satellite control, low-cost ultra-high-power flexible solar arrays, satellite-ground laser communications — none of these are easy. First we land the 10 PFLOPS, get the engineering to work, then iterate upward.
This sounds like a very long-term plan. What's the commercialization path for the near term?
Sun Xiaolei: We're currently preparing proposals for potential customers, aiming to secure at least one order within the year.
What's the customer logic right now? Once computing satellites are deployed in space, remote sensing data can be processed in real time on orbit without being transmitted back to Earth for computation — this creates continuity with their existing investments.
Our planned rhythm is: we'll compete for large computing satellite orders with the goal of first getting the technology to work, then driving costs down in the future, and eventually operating ourselves. The operational vision is building our own constellation, including hundreds of low-orbit computing satellites that can later scale to thousands, providing computing services.
There's another issue people don't talk about much: even if I launch computing satellites, how do users access them? So we're also planning medium-orbit communication satellites — in the future, users will connect through these communication satellites to access large models deployed in space.
How big is the market for space-based computing? Do the costs work out right now?
Sun Xiaolei: Current commercial rocket pricing is generally 60,000–70,000 RMB per kilogram. Once rocket reusability is achieved, we're hoping to get that down to 10,000 RMB per kilogram.
Satellite costs also need to come down. At 30 million RMB per satellite now, we need to drive that to around 10 million through technology iteration.
We're building a calculation model — factoring in satellite costs, launch costs, and projected AI computing demand growth over the next 5 years, assessing whether ground infrastructure can sustain it, then back-calculating the economics of space-based data centers. We're running separate calculations for domestic and international markets to prioritize market direction.
We don't face the same energy constraints as the United States — is putting data centers in space really more cost-effective than building them on the ground?
Sun Xiaolei: Back when satellite internet first emerged, nobody could clearly explain how it would make money — Musk just went ahead and did it. Space-based computing is the same. We believe the direction is right, so we move forward first, and the path becomes clearer as we go.
Look at how AI has exploded over five years — who predicted this five years ago? If it develops for another five years with deeper penetration, computing demand could be dozens or hundreds of times what it is now. When ground infrastructure can't sustain that, space-based computing becomes commercially viable.
There's another point we can't ignore: security. The Russia-Ukraine war made everyone see how significant Starlink's role was. Ground data centers are the first things destroyed in war — move them to space, and only countries with real capability can reach them. This is also why the government has been pushing this direction.
What's the operating model for a space-based data center? What ultimately generates revenue?
Sun Xiaolei: We can reference ground data center models — however they operate on the ground, we operate in space, with similar billing structures. Either charge large model companies, or buy models ourselves and offer direct-to-consumer services. Fully operating on our own might require a 10-year development horizon.
Right now the key isn't how to operate — it's building the "data center" itself, driving costs down through technology iteration. Once costs reach a certain level, operations will follow naturally.
Everyone knows this needs early positioning — being 2 years late doesn't matter, being 10 years late means it's too late.
Source: 36Kr Future Industries
Author: A Zhi
Generation and Judgment: My "Sparring" Notes with AI | Unity Ventures Perspectives
