As the commercial space industry enters its second half, the satellite applications market is increasingly driven by cost efficiency and is, in turn, pushing upstream remote sensing satellite manufacturers toward a new paradigm of integrated sensing and computing. China has now entered the era of intelligent satellites.
Why has onboard intelligence become a necessity for the industry? How will AI reshape the remote sensing satellite sector? And how will the criteria for evaluating remote sensing satellite capabilities change in the future?
Today, we take a closer look at how AI is transforming remote sensing satellites.
Remote Sensing Satellites Enter the Era of Intelligence
On August 5, 2026, Dongfang Xinglian launched two intelligent remote sensing satellites—Dongfang Xinglian 05/06, also known as the Dongfang Huiyan hyperspectral AI twin satellites—into orbit. Just two weeks earlier, CAS Xiguang Aerospace had launched its first “quantitative hyperspectral intelligent computing satellite,” Xiguang-2 01, also known as Caiyun Hyperspectral-01.
These two launches reveal a profound shift in China’s remote sensing satellite industry: the development paradigm is moving from being primarily technology-driven to increasingly application-driven, signaling that China has formally entered the era of intelligent remote sensing satellites.
Although the two satellite systems use different names and technical approaches, both the Dongfang Xinglian hyperspectral AI twin satellites and CAS Xiguang Aerospace’s hyperspectral intelligent computing satellite share the same fundamental concept: adding onboard computing platforms and AI capabilities to conventional remote sensing satellites.
By deploying dedicated onboard AI models, satellites can become more autonomous. The industry is moving away from the traditional “collect data, downlink it, and process it on the ground” model toward an intelligent architecture in which part of the data is processed directly in orbit.
For the past six decades, the basic function of Earth observation satellites has essentially been that of a “data relay”: collect, store, and downlink data, then wait for ground systems to analyze it. This process often takes hours or even days. Yet many remote sensing applications evolve within minutes or even seconds—typhoon trajectories, wildfire spread, algal blooms, and flood peaks, for example. By the time decision-makers receive and analyze the data, the situation may already have entered its next stage.
With onboard computing and AI models, conventional remote sensing satellites can perform autonomous target recognition, information extraction, and anomaly detection in orbit. They can also autonomously plan onboard tasks according to changing requirements.
This shifts the core value of satellites from “acquiring data” to “generating actionable intelligence.”
As China’s commercial remote sensing constellations expand rapidly, while onboard AI models, satellite computing chips, and lightweight algorithms continue to mature, intelligent processing capabilities that previously could only run on the ground are increasingly being deployed in space. The transition to the intelligent satellite era is therefore becoming a natural next step.
Putting AI in Orbit Is Becoming a Necessity
Why do remote sensing satellites need to become AI-enabled? There are three major reasons.
1. Resolving the conflict between explosive data growth and limited downlink bandwidth
This is the most immediate requirement.
A single high-resolution remote sensing image can reach several gigabytes, while a single hyperspectral dataset can exceed tens of gigabytes. A large constellation can generate data at the petabyte scale every day.
Under the traditional architecture, more than 90% of acquired imagery may reportedly be unable to reach the ground because of insufficient downlink bandwidth, eventually being overwritten or deleted.
With onboard computing, satellites can screen and process data in orbit and downlink only critical target information. This can potentially improve transmission efficiency by orders of magnitude, breaking the bandwidth bottleneck that has long constrained satellite applications.
2. Meeting the demand for minute-level real-time response
This is where the commercial value of remote sensing satellites becomes particularly evident.
Under the traditional “space collection, ground processing” model, the satellite data workflow is lengthy. From uploading commands and acquiring imagery to downlinking data and generating usable information, the process can take hours or even days.
This may be acceptable for applications that are relatively insensitive to time, such as mapping and large-scale surveys. But in time-sensitive scenarios—including emergency response, agricultural monitoring, and maritime surveillance—traditional satellites can often only play the role of a post-event analysis tool.
Once computing moves into orbit, AI can potentially support autonomous task planning, image acquisition, data processing, and intelligent decision-making directly onboard the satellite.
The entire workflow can be compressed to minutes, enabling a new operational model: “capture, recognize, and alert.”
This is a capability that conventional ground-based processing architectures cannot easily achieve.
3. Supporting the operation of large-scale satellite constellations
As constellations expand to hundreds or even thousands of satellites, the complexity of constellation operations and mission planning increases dramatically.
The traditional approach—manually uploading commands, processing data on the ground, and making decisions based on the processed information—not only consumes substantial space-to-ground bandwidth but also becomes increasingly difficult to sustain in terms of time and operating costs.
The greatest advantage of onboard computing, therefore, may be that it enables satellites to operate with greater autonomy.
Satellites can evolve from tools that simply “take pictures” into autonomous space nodes capable of sensing, analyzing, and executing tasks.
This represents a fundamental shift from “people operating around satellites” to “satellites proactively serving people.”
It also allows the efficiency of constellation operations to scale with the rapid growth in the number of satellites being deployed.
AI Is Redefining Remote Sensing Satellites
Dongfang Xinglian 05/06 and Xiguang-2 01, or Caiyun Hyperspectral-01, follow different technical approaches, but they embody the same underlying trend:
bringing computing capabilities into satellites and giving them significantly greater information-processing capabilities.
Take Xiguang-2 01, also known as Caiyun Hyperspectral-01, as an example.
The satellite carries advanced hyperspectral and panchromatic optical payloads, with 70 spectral channels. Leveraging an interferometric architecture, it provides Earth observation capabilities in the visible-to-near-infrared spectrum with 2-meter panchromatic spatial resolution, 10-meter hyperspectral spatial resolution, and 7-nanometer spectral resolution.
It can precisely capture the spectral fingerprints of materials on the Earth’s surface, enabling three core capabilities: material identification, compositional inversion, and quantitative monitoring.
This allows it to distinguish material characteristics such as crop growth conditions and water pollution with greater precision. In effect, it is like performing a “CT scan” of the Earth, moving remote sensing from qualitative imaging toward quantitative analysis.
The satellite also carries the Zhijia NX4 onboard intelligent computing unit, jointly developed with Zhejiang Lab. The computing platform provides up to 248 TOPS of computing performance and high-speed data interfaces, enabling a workflow in which data is acquired, processed, and analyzed in space.
This can significantly shorten monitoring response times and reduce the pressure on space-to-ground data transmission, shifting remote sensing monitoring from post-event analysis toward real-time warning.
The Dongfang Huiyan hyperspectral AI twin satellites take the integration of hyperspectral remote sensing and onboard AI a step further.
Their primary payload is a hyperspectral remote sensing camera featuring a 22-dimensional spectral fingerprint. It provides 5-meter spatial resolution in the visible and near-infrared (VNIR) band, a 300-kilometer swath width, and 22 spectral channels—17 in VNIR and five in short-wave infrared (SWIR)—covering the 0.4–1.7 μm spectral range.
On the computing side, the two satellites carry 400 TOPS of onboard computing power and 10 TB of onboard storage.
By moving AI inference to orbit, the satellites are no longer limited to simply taking images. During a satellite pass, they can potentially perform segmentation, recognition, classification, change detection, information analysis, and strategy generation.
These capabilities transform the raw parameters acquired by remote sensing payloads into a higher-dimensional sensing capability. Every part of the Earth’s surface can potentially acquire a unique spectral identity—from vegetation stress and water composition to soil minerals and urban materials—which can then be analyzed directly in orbit.
The difference is clear.
Traditional remote sensing satellites are primarily designed to “capture images of the Earth.”
Intelligent remote sensing satellites are beginning to “understand the Earth.”
This is not simply an efficiency upgrade for remote sensing satellites. It represents a fundamental shift in the application paradigm.
How Will Remote Sensing Satellites Be Evaluated in the Future?
Historically, the two core metrics used to evaluate remote sensing satellite performance have been spatial resolution and temporal resolution.
Spatial resolution describes the level of detail a satellite can capture, while temporal resolution reflects how frequently it can revisit a given area.
However, as remote sensing moves toward the integration of sensing and computing, these two metrics can no longer fully describe a satellite’s overall service capability.
A new dimension is emerging:
AI resolution.
Today, conventional remote sensing satellites are approaching technological limits in spatial and temporal resolution. Sub-meter spatial resolution and minute-level revisit capabilities are increasingly becoming standard for some advanced systems.
Yet two satellites with similar spatial and temporal resolution can have dramatically different practical value in real-world applications.
The concept of “AI resolution” attempts to describe the application efficiency of an individual remote sensing satellite.
The greater the onboard computing power and the more capable the AI models, the more high-value information a single satellite can potentially generate—and the greater its commercial value.
AI resolution encompasses factors including:
- Onboard computing capability
- AI model capability
- Data-processing efficiency
- Autonomous mission planning and execution
In the future, AI resolution could stand alongside spatial and temporal resolution as one of the three core dimensions for evaluating remote sensing satellites.
It could even become a key factor in reshaping the business model of satellite-based remote sensing.
The traditional satellite service model is essentially to “sell raw data.” Downstream users must then invest additional resources to process that data themselves.
High-AI-resolution intelligent satellites, by contrast, could increasingly “sell solutions”—delivering information that can be directly used for decision-making, or even generating strategies for subsequent action.
Conclusion
The launches of Xiguang-2 01, also known as Caiyun Hyperspectral-01, and the Dongfang Huiyan hyperspectral AI twin satellites send an important signal.
Satellites are beginning to move from “seeing” to “understanding.”
The future competition in space may not simply be about who owns more satellites, but about who can turn more satellite data into valuable, actionable decision-making information.
China’s commercial space industry is taking significant strides into the era of intelligent satellites.
