As the deployment of commercial aerospace constellations speeds up, massive industry dividends are on the way.

On August 11, the commercial aerospace sector on China's A-share market witnessed drastic volatility, with stocks such as Tianli Composite and Srui New Materials posting intraday drops of more than 10%.

Just the night before, the launch mission of the Long March-7A carrier rocket ended in failure, and subsequently, the Zhuque-3 Y2 carrier rocket of Landspace announced the postponement of its launch plan.

I believe the sharp market reaction stems from the fact that most investors and observers still anchor the valuation of commercial aerospace to the single metric of "launch success or failure".

A failed rocket launch is seen as casting a shadow over the entire industry, while a successful launch immediately ignites a new round of speculative fervor.

This mindset has a critical blind spot.

When people are talking about the Qianfan constellation having more than 200 satellites in orbit, about Firestar Aerospace signing a series of "massive constellation deployment" agreements with multiple satellite companies, and about China submitting frequency and orbit resource applications for more than 200,000 satellites to the International Telecommunication Union, I wonder if anyone has noticed that the underlying logic behind these events is far more complex than simply "who sends satellites into space". Why is that? Today we will discuss this topic.

Accelerated Constellation Deployment, No Room for Retreat

The current accelerating pace of commercial aerospace is closely linked to the "first-come, first-served" rule of the International Telecommunication Union.

This rule stipulates that after any country or entity submits a frequency and orbit resource application to the ITU, it must launch the first satellite within seven years, and ensure that the satellite operates continuously in orbit for at least 90 days.

If this requirement is not met, the application will be automatically invalidated.

Orbital positions and frequency spectrums in space are scarce resources, whose scarcity is identical to that of land: once occupied, latecomers will lose the right to use them forever.

China's submitted applications for more than 203,000 satellites are essentially a timetable presented to the international community.

The seven-year window behind the 200,000 satellites is a countdown timer.

This countdown forces the entire industrial chain to shift from a "technology verification-driven" track to a "delivery capability-driven" track.

As a result, it is not difficult to understand many of the phenomena currently taking place in the industry.

Firestar Aerospace's launch of the "massive constellation deployment" service is driven by the rigid demand from downstream satellite operators. According to reports from CLSA, Gao Yufeng, co-founder and CEO of Firestar Aerospace, explicitly stated that the current domestic one-off rocket launch cost is about 50,000 to 100,000 yuan per kilogram, and their goal is to reduce the cost to around 10,000 yuan per kilogram to realize "airline-style" transportation. The underlying logic is that only by spreading the cost of each launch through high-frequency and large-scale launches can downstream constellation plans become financially viable.

Professor Jin Zhonghe, director of the Micro-satellite Research Center of Zhejiang University, also pointed out in an interview that domestic rocket companies and satellite companies currently generally lack large-scale commercial revenue that can cover their costs.

Although mobile phone direct-to-satellite connectivity and the Internet of Things are highly anticipated, limited by communication coverage and user payment habits, it is difficult to see explosive application scenarios in the short term.

This reality of "high input and low output" forms the background of the constellation deployment race.

The frequent signing of contracts, bulk packaging, and deep binding between rockets and satellites appear to be commercial cooperation on the surface, but in essence, they are paths for upstream and downstream players in the industrial chain to jointly seek survival under intense time pressure.

With the seven-year deadline hanging over the industry, the "step-by-step" approach is no longer a viable option.

In addition, there are even deeper industrial changes behind this trend.

The core logic of traditional satellite constellation deployment is function-oriented: a communication satellite is launched to forward signals, a remote sensing satellite is launched to capture ground images, and every satellite is regarded as a "payload" with a specific function.

However, when the scale of constellations jumps from the level of hundreds of satellites to tens of thousands or even hundreds of thousands of satellites, a qualitative transformation is taking place.

The manufacturing cost of individual satellites is dropping sharply. For example, companies such as GalaxySpace and Spacety have reduced the cost of a single satellite to the level of millions of yuan, which means satellites are evolving from sophisticated "customized instruments" to mass-producible "standardized hardware".

As hardware becomes commoditized, the connotation of constellation deployment changes accordingly. Every satellite that enters orbit is no longer just an executor of a specific function, but a "network node" equipped with computing, sensing, and communication capabilities. The interconnection, collaboration, and data flow between these nodes form the prototype of a distributed system.

This is exactly why Firestar Aerospace proposed the "one rocket, one orbital plane" massive launch model: this is a qualitative change that deploys a complete "computing plane" at one go. It is similar to building a data center: adding servers piecemeal can never create cluster effects, and servers must be delivered in full units of cabinets and planes.

Constellation deployment follows a similar logic: only when satellites on one orbital plane work in coordination can seamless coverage of a certain region and continuous data acquisition be achieved.

According to Gao Yufeng, this is a "breakthrough in business thinking". In the past, the industry was accustomed to "ride-sharing" launches, sending one or two experimental satellites slowly; now what is needed is "exclusive express delivery", which sends satellites into orbit in batches directly to complete the full deployment of an orbital plane.

The fundamental reason why this model shift is accepted by the market is that the downstream pressure of constellation deployment has become large enough to reshape the upstream service form.

When the launch pace is compressed from "one launch every few months" to "two launches a week", and the scale of constellations expands from dozens of satellites to hundreds and even tens of thousands as planned, a qualitative change occurs: these satellites will work in coordination in a cluster, establish links between each other, make data flow horizontally across the orbital plane, and make their coverage areas overlap seamlessly.

The capability boundary of a single satellite is limited, but the coordination of dozens of satellites on a complete orbital plane forms a distributed in-orbit data acquisition and processing system.

At that point, will you still think it is just a single "communication network" or "remote sensing network"? It should be the prototype of a new species.

What Is the True Value of Space?

I call this new species "space-based cognitive infrastructure". Its underlying architecture can be broken down into three functional layers: perception, connection, and computing. Corresponding industrial actions are taking place at each layer right now, though most people have not yet put them into the same analytical framework.

From overseas developments, SpaceX and Tesla jointly announced an investment of 16.8 billion US dollars to build the large-scale chip manufacturing center Terafab, whose application scenarios are clearly targeted at autonomous driving, humanoid robots, and space data centers.

NVIDIA launched a computing module for space scenarios at the GTC conference last March, whose main directions include orbital data processing, geospatial information analysis, and autonomous operation of spacecraft.

SpaceX then announced a cooperation with NVIDIA to design satellite AI computing payloads, exploring the extension of AI computing capabilities from the ground to low Earth orbit.

These actions show that the next piece of the puzzle for the space economy is computing.

All kinds of constellations that are being deployed rapidly in China, whether the communication network of the Qianfan constellation or the remote sensing sector represented by Mofang Satellite and Sixiang Tech, are essentially laying the foundation for this positioning.

To better understand this process, we can break it down from the three functional layers.

First, perception.

The most direct change brought by the intensive deployment of remote sensing satellite constellations is the revolutionary improvement of temporal resolution.

In the past, the revisit period of a remote sensing satellite for the same location could be several days or even weeks, and what it obtained was a static "snapshot". But when the number of satellites in a constellation reaches hundreds or even thousands, the sampling frequency for any point on the Earth's surface can approach real-time or near-real-time levels.

This means the Earth has transformed from a physical space that can only be observed from a static perspective to an object that can be continuously and dynamically modeled.

The growth of crops, urban expansion, traffic flow, ocean temperature changes, and the evolution process of natural disasters — all these phenomena that were difficult to track accurately in the past due to data delays — will be converted into structured data flows in the time dimension by the space-based perception network.

In the training of large AI models, the "quality" and "quantity" of data directly determine the upper limit of the model's capability. The training data of an AI that can understand the spatio-temporal laws of the physical world cannot only come from texts and images on the Internet, it needs continuous observation data of the "real world" as nourishment.

The denser the remote sensing constellation is and the more frequent its revisit, the more developed the "sensory nerves" that feed this AI brain will be.

The space situational awareness demand represented by the Kaiyuan Group and the commercial remote sensing sector represented by Sixiang Tech, which is highly dependent on low-cost and high-frequency launches, seem to belong to different tracks. But under the framework of the "perception layer", they jointly form a high-density, high-frequency, multi-dimensional space-based data acquisition network.

The fulcrum of its commercial value does not lie in "selling images", but in providing the raw information entry for larger-scale computing systems.

Second, connection.

The Ministry of Industry and Information Technology of China approved the first commercial test of satellite IoT services, and the industry's continuous investment in the long-term vision of mobile phone direct-to-satellite connectivity, all these actions point to a common direction: the space-based network is evolving from "people connected to the network" to "everything connected to the network".

In the traditional narrative of satellite communication, the core metric is bandwidth, that is, the maximum amount of data that can be transmitted. But at the connection layer of the "cognitive infrastructure", the more critical metrics become certainty and low latency.

For example, a large number of self-driving cars are running on roads, drones are carrying out logistics deliveries over cities, and ocean-going cargo ships are sailing in sea areas with no ground signal coverage. When all these mobile terminals rely on the space-based network for connectivity, the network needs to ensure not only "can connect", but also "can connect within a certain time limit" and "whether the latency after connection is predictable".

These features are the foundation for the coordinated operation of complex ground systems. Self-driving vehicles need to maintain millisecond-level communication latency with the cloud decision system, otherwise safety cannot be guaranteed. The global supply chain tracking system needs to be able to locate the position and status of goods at any time, and IoT devices that monitor environmental data in no-man's lands need to transmit data stably without packet loss.

The cooperation between NVIDIA's space computing module and SpaceX is also supposed to seek breakthroughs at this layer: deploying computing capabilities to orbit, so that data processing can be as close to the source of data generation as possible, thus shortening the physical length of the decision-making link.

In a sense, the construction goal of this layer is to lay a set of highly efficient "nervous system" for the future space-based AI system, so that the information obtained by the "eyes" of the perception layer can be transmitted and processed in a timely manner.

Third, computing.

The current technological and industrial conditions are not yet sufficient to support the substantial construction of "orbital computing power centers", but the direction is clearly defined: space provides two physical conditions for large-scale AI computing that cannot be replicated on Earth.

The first is energy. The solar energy received on the Earth's surface is first absorbed and scattered by the atmosphere, so its energy density is reduced; in addition, due to the alternation of day and night, photovoltaic power plants cannot work for half of the time.

In low Earth orbit, solar radiation is not interfered by the atmosphere, so its intensity is higher, and near-24-hour uninterrupted sunlight can be achieved through orbital design. For AI training clusters with huge power consumption, energy cost is the primary variable that determines economic feasibility.

The "free and abundant sunlight" in space fundamentally changes the cost structure of computing power production.

The other is heat dissipation. One of the biggest headaches for large data centers is to remove the huge heat generated when chips run. The cooling system of data centers consumes a huge amount of electricity, and the engineering is complex. In space, the temperature of the cosmic background is close to absolute zero, which is a natural heat sink with unlimited capacity. When high-density computing clusters are placed in orbit, heat dissipation barely generates extra energy consumption — heat can dissipate naturally through radiation.

The combination of these two physical conditions makes the concept of a "space AI computing power factory" economically rational in the long term.

Its logic is not "we go to space to do computing just because space is cool", but "if doing computing there is inherently cheaper, the profit-seeking nature of capital will drive the industry to migrate there".

Therefore, every satellite with edge computing capability launched today is actually a prototype node of the future large-scale space-based computing cluster; the inter-satellite link and constellation deployment technologies verified today are the foundation for internal communication of the future orbital computing power cluster; the orbital data backhaul mechanism explored today is the information channel between the future space-based AI and ground applications.

In this sense, the acceleration of constellation deployment has become an upfront investment in future "orbital computing power assets". The contracts signed and capital expenditures in the industrial chain are essentially laying the underlying orbital infrastructure for this long-term vision.

How Long Will Capital Wait?

When the ultimate vision of an industry is stretched long enough and grand enough, frictions and contradictions in the implementation process will emerge.

The current reality facing commercial aerospace can be summed up in one sentence: there is a clear temperature difference between capital preferences and industrial logic.

From the case of Firestar Aerospace's frequent contract signings, we have observed a phenomenon. According to Wang Ze, Chairman of China Star Yao and Chief Strategy Officer of Firestar Aerospace, who spoke to CLSA, the standards by which capital views rocket companies have shifted: "In the past, investors focused on whether there were technical breakthroughs; now they focus on whether downstream customers are locked in, whether there are confirmed launch orders, and whether mass delivery can be achieved within the constellation deployment window."

This shift echoes the overall rhythm of the industry as it moves from the "technology verification period" to the "large-scale realization period".

Investors' emphasis on "confirmed orders" is driven by the instinctive need for risk control — signed contracts and scheduled launch plans are more anchorable for valuation than technical parameters in laboratories.

However, this "certainty preference" comes at a high cost. It makes the flow of capital show a "real estate-like" feature: emphasizing heavy assets, contracts, and short-term visible cash flow, while being relatively cautious about long-cycle, high-tolerance underlying innovations.

The core links required for space-based AI, such as computing chips, inter-satellite communication protocols, and orbital data processing software stacks, belong exactly to the latter category — long return cycle, unestablished technical routes, and high failure risk.

The tension between this "short-term capital" and "long-term cognition" is the deepest structural financial contradiction facing commercial aerospace today.

Its consequences will not be immediately visible, but will appear five or ten years later: when the "quantity" of constellation deployment reaches the target, but the "quality" — that is, the computing and service capabilities that can truly function in orbit — fails to keep pace, the industry's valuation system may face a revaluation.

On the other hand, there is the issue of corporate governance.

When analyzing SpaceX, Neuberger Berman Fund specifically pointed out the pros and cons of the dual-class share structure.

This structure concentrates control in the hands of the founders, freeing the company from the constraints of quarterly profit pressure, and enabling long-term investment in R&D and scale expansion. But at the same time, it is also a mechanism that institutionalizes and even solidifies control. Once established, external shareholders can hardly influence corporate decisions through conventional channels.

This issue has special significance in the commercial aerospace field, because unlike ordinary tech companies, the infrastructure that commercial aerospace companies may master in the future transcends national borders and the scope of traditional judicial jurisdiction. If a private enterprise in the future operates a huge orbital AI computing network large enough to cover the whole world, which can carry out real-time observation and data analysis of any corner on Earth, who will ensure that the decision-making logic of this system is transparent and responsible?

When control is permanently concentrated in the hands of a few people, and the capability of this system is powerful enough to affect public safety and economic operation, corporate governance will rise from a shareholder interest issue to a public issue about power structure.

This is not an urgent problem that needs to be solved right now, but it is worth being included in the analytical framework in the early stage of the industry. Historical experience proves that the more powerful a technological system is, the heavier the consequences brought by defects in its governance structure will be.

The news of the Long March-7A rocket launch failure spread rapidly on the night of August 10,