Space computing is no longer just for aerospace missions—it has started to serve the AI industry.

Over the past decade or more, competition in the AI industry around computing power has almost all centered on a single question: how to build increasingly larger terrestrial computing clusters.

Whether it’s NVIDIA GPUs, ultra-large data centers, or intelligent-computing centers spread across the globe, together they form the most important infrastructure in today’s AI industry. In the past, nearly all the computing resources used for training and inference of large models came from the ground. Now, that landscape is beginning to change.

On July 18, during the 2026 World Artificial Intelligence Conference (WAIC), Guoxing Aerospace and SenseTime announced their joint initiative to build the SenseTime computing constellation. According to the plans released by both parties, following the development path of “single-satellite validation—platform buildout—constellation networking,” they will gradually build a space-computing network on the scale of thousands of satellites and “ten thousand P,” and further construct a new generation of globalized AI infrastructure.

Previously, if you wanted to deploy computing capability into orbit, it was mostly engineering exploration carried out by aerospace agencies, satellite manufacturers, and space AI companies around onboard computing.

Today, as space computing gradually moves from single-satellite validation toward constellation networking, space-ground coordination, and unified scheduling, more AI infrastructure participants—such as large-model companies and intelligent-computing platforms—are entering this space, jointly driving the evolution of space computing from a specialized capability for aerospace missions toward a more systematic and open infrastructure.

So a new question is slowly coming into view: when leading large-model companies begin to access space computing in a systematic, end-to-end way, does that mean space computing is entering the AI industry supply chain and becoming a new component of AI infrastructure?

1｜Why is space computing moving toward openness?

If we look back at the global development of space computing, it’s not hard to see that for a long time, the focus in this field has been how to enable satellites to have computing capability.

The earliest satellites mostly served functions such as data collection and forwarding. Whether it’s remote-sensing imagery or communication data, it needs to be transmitted back to the ground first, and then processed by terrestrial computing centers. Although this model has matured, as the number of satellites continues to grow, downlink bandwidth, communication latency, and terrestrial processing capacity have gradually become important factors limiting overall system efficiency.

So more and more research has begun trying to deploy some computing capability directly into orbit. From early onboard AI and onboard preprocessing of remote-sensing data, to later support for AI model deployment, multi-satellite cooperative computing, and space-ground joint scheduling—over the past few years, global space computing has gradually moved from concept validation into the engineering exploration stage.

This means the industry’s focal questions are changing. In the past, people cared about whether satellites could perform computation; today, more people are asking: how can these space-computing resources better connect with the industry so as to continuously serve more applications?

In other words, as space-computing capabilities continue to improve, the industry’s attention is shifting from how to equip a single satellite with computing capability to how to build sustainable, open, and serviceable space-computing infrastructure.

This is also the industrial context behind this cooperation between Guoxing Aerospace and SenseTime. After the two companies announced the cooperation, SenseTime became the first large-model company to connect to Guoxing’s space-computing system at a system level. At the same time, it sends an important signal: space computing has begun trying to form synergy with mature large-model systems and terrestrial AI infrastructure.

From this perspective, if previously the development focus of space computing was continuously enhancing onboard computing capability, then the next stage is very likely to focus on enabling space computing to truly become a new type of computing resource that enters the AI industry supply chain.

2｜A computing satellite alone is not yet a space cloud

However, to enable space computing to truly enter the AI industry supply chain, launching just a few satellites with computing capability is far from enough.

The development path disclosed by both sides is “single-satellite validation—platform buildout—constellation networking.” They plan to advance the construction of an integrated space-and-ground AI infrastructure in phases. Along this path, space-computing capability is also evolving—from the initial computing payloads to computing satellites, and further to computing constellations. The latter requires large numbers of satellites to work together to form a computation network that can be uniformly scheduled and operates continuously. This is also why the two sides proposed the concept of a hybrid space-ground cloud.

The “hybrid space-ground cloud” here is not simply moving terrestrial data centers into space; rather, it means combining orbital compute nodes with terrestrial intelligent-computing centers into a unified scheduling computation system. Different computing tasks can be dynamically allocated between orbit and ground based on the location of the data, latency requirements, and compute scale: tasks better completed in orbit remain in space for processing, while large-scale training and other tasks still rely mainly on terrestrial intelligent-computing centers.

In this system, terrestrial intelligent-computing centers and orbital compute nodes are not independent of each other. Instead, task allocation, model updates, data synchronization, and resource scheduling must be completed dynamically between orbit and ground according to the characteristics of each task.

This is not an isolated case. In recent years, more and more similar ideas have emerged internationally. For example, as onboard AI gradually expands into Orbital Data Center (orbital data centers), the underlying thinking is changing: the discussion is no longer about a single satellite, but about a set of computing infrastructure that can run stably over the long term.

This trend is also increasingly reflected in the layouts of more companies. For example, Google proposed “Project Sunfish” to explore space solar power; SpaceX is trying to introduce AI computing capability into the Starlink ecosystem; and Starcloud, supported by NVIDIA, is targeting orbital data centers. Although these explorations follow different technical paths, they all point to the same direction: low Earth orbit is gradually evolving from a place where satellites operate into a new kind of infrastructure that can continuously provide computing services.

From a single satellite to a computing network, what truly changes is not the number of satellites, but how resources are organized. The challenges at this stage are real and substantial: in an environment where communication links, energy supply, and satellite operating conditions keep changing, achieving reliable task scheduling, data exchange, and cooperative computing among a large number of satellites.

Only by solving these issues can space computing potentially become an infrastructure that continuously provides computing services—like today’s data centers—rather than just a space technology capability.

3｜The first time space computing enters the AI industry supply chain at a system level

In the cooperation between Guoxing Aerospace and SenseTime, the most worth paying attention to is not the “thousands of satellites and ten thousand P” goal proposed by both parties, but the fact that SenseTime became the first large-model company to connect to Guoxing’s space-computing system at a system level. This shift indicates that the development focus of space computing is moving.

According to publicly available information, Guoxing Aerospace is a company dedicated to building space-computing infrastructure that serves artificial intelligence, providing technical products and services worldwide such as space computing and artificial intelligence.

Looking at the company’s key progress in recent years, it has continued to build out its space AI roadmap. In September 2024, it developed and launched the world’s first AI satellite, and subsequently carried out the world’s first in-orbit running of AI large models. In May 2025, Guoxing Aerospace’s “Star Computing” Plan 01 successfully placed a space computing center into orbit; as a result, the world’s first space computing center was established, with in-orbit cluster compute reaching 5POPS, ranking first globally.

According to publicly disclosed information, Guoxing Aerospace has completed end-to-end commercial closed-loop validation of “deployment—inference—calling.” From the in-orbit deployment of Alibaba Qwen3 in November 2025, to the in-orbit inference of ByteDance UI-TARS in January 2026, and then in March 2026 when a ground robot called space computing power after passing 19 tests by the China Academy of Information and Communications Technology. In addition, it has also accumulated commercialization practices in areas such as remote-sensing data processing and in-orbit intelligent computing.

The core goal of all this work is to answer two questions: can space computing be built? And after it is built, can it operate stably? Today, these two questions have been gradually validated.

By contrast, this cooperation answers a different question: can already-built space computing become a new type of computing resource that the AI industry can call?

The difference in this cooperation is that Guoxing is starting to open the already-built space-computing system to leading large-model companies as a system-level capability. Previously, whether it was AI satellites or large models deployed into orbit, most validation focused on whether models could run in orbit, update, and complete tasks—fundamentally still part of building space AI capability.

In this cooperation, based on the existing space-computing system, SenseTime’s large-model capabilities, intelligent-computing platform, and resource scheduling system are further introduced to explore a computation model of space-ground synergy. Put another way, what used to be “sending the model into space” is now beginning to “open up the already-built space computing resources.”

Therefore, with SenseTime as the first large-model company to access Guoxing’s space-computing system at a system level, space computing can for the first time form system-level synergy with mature large-model systems, terrestrial intelligent-computing resources, AI development platforms, and model training and inference capabilities. This synergy means that space computing is starting to move from mainly serving aerospace missions toward serving a broader AI industry supply chain.

In the past, large-model companies mainly called data and computing resources from terrestrial intelligent-computing centers. In the future, for some tasks that are more suitable to complete in orbit, space computing may also become a component within the overall computing system, working together with terrestrial computing to complete tasks.

Of course, this does not mean that all AI computation will migrate to space in the future. In fact, constrained by launch costs, energy supply, thermal dissipation conditions, and inter-satellite communication capabilities, large-scale model training will still take place mainly on the ground. A more realistic direction for space computing is to undertake computation tasks that must rely on orbital environments or can be completed more efficiently in orbit, complementing terrestrial intelligent computing rather than replacing it.

Therefore, what is truly worth focusing on in this cooperation is not just the “strong combination” of the two companies, but the fact that system-level synergy for the AI industry is starting to form between space AI enterprises and leading large-model enterprises. As this synergy gradually matures, the value of space computing is also beginning to shift from “successfully completing an in-orbit computation” to “continuously providing computing services.”

4｜Before “thousands of satellites and ten thousand P,” first find space computation that can be scaled and replicated

Although space computing is entering a new stage of development, it’s also important to recognize that there are still many engineering challenges to overcome before it can form a truly mature industrial infrastructure.

First, there is construction cost. The larger the computing scale, the higher the requirements for launch capability, satellite platforms, energy supply, and orbital operations and maintenance. How to control the overall cost while ensuring computing capability directly determines whether a space-computing network can achieve commercial sustainability.

Second, there is operational efficiency. In-orbit computing needs not only to address aerospace issues such as radiation hardening, thermal management, and energy management, but also to establish a stable inter-satellite network to enable continuous, reliable data exchange and resource scheduling among large numbers of satellites. This means that future competition will not be solely about the computing capability of a single satellite, but about the cooperative efficiency of the entire network.

More importantly, space computing needs to find application scenarios that truly fit it. What genuinely needs to be “sent into space” is not all computation, but those computing tasks that must rely on orbital environments and can continuously create value.

The unique value of space computing lies in providing computing services for two kinds of scenarios: first, onboard preprocessing for massive space data to avoid downlink bottleneck congestion; second, low-latency inference support for intelligent devices that are not covered by terrestrial networks. These scenarios are also where space computing’s key value differs from that of traditional data centers.

From Guoxing Aerospace’s engineering practices over the past few years, it has continued to explore in the relevant directions. Looking ahead, the bigger challenge may be how to further turn these capabilities into scalable, sustainably operated business models—not just remain at the level of a single project or pilot demonstration.

Therefore, competition in the next phase of space computing will not simply be about who can send more computing power into orbit earlier; it will be about who can transform existing engineering capabilities into low-cost, scalable, sustainably operated global computing and data services.

5｜Conclusion

From AI satellites and space computing centers to SenseTime becoming the first large-model company to access the Guoxing space-computing system at a system level—through Guoxing Aerospace’s development path, we can see a snapshot of how space computing is evolving into a new stage: in the past, space computing focused more on whether it could compute; today, it is starting to discuss how it can become part of the industry.

The real value of this cooperation that deserves attention is that space computing has begun to truly become part of AI infrastructure. When space computing becomes a new kind of computing resource that leading large-model enterprises can access at a system level, the criteria for measuring the value of space computing in the future may no longer be whether an in-orbit computation succeeds, but whether it can continuously participate in the global AI computing ecosystem—just like today’s data centers.

“ In the wave of investment, there is no shortage of opportunities—only a lack of clear-headed judgment; every turning point may hide Alpha big opportunities; cognition determines returns—follow + like so you don’t miss every industry inflection point; welcome to discuss your judgments and deployment ideas in the comments, and let’s evolve together and seize the next wave of dividend opportunities. ”