As record-breaking heat waves sweep across parts of the world, another challenge is confronting the artificial intelligence boom: keeping the high-performance chips inside data centers cool enough to keep running.

The challenge extends far beyond seasonal heat. As AI models become larger and more computationally intensive, demand for computing power is rising at a pace that is testing the physical limits of conventional data centers.

According to the International Energy Agency (IEA), data centers accounted for about 1.5% of global electricity consumption in 2024, with power demand expected to rise sharply as AI deployment accelerates. 

Rising electricity use, cooling costs and increasing exposure to climate-related disruptions are prompting governments and technology companies to rethink where future computing infrastructure should be built.

Translating this background into action, a multi-city layout is taking shape in China. Beijing serves as the anchor for research and development and standard setting through its new innovation center. Tianjin focuses on space-ground computing integration and hardware development. Shanghai is building out the broader industrial ecosystem. Hangzhou leans into intelligent applications, anchored by the "Three-Body Computing Constellation," which has already been deployed in orbit. Led by Zhejiang Lab (ZJ Lab), it aims to build a network of thousands of satellites with a total computing power of 1,000 peta operations per second. The constellation will enable real-time in-orbit data processing with intelligent payloads.

Why move computing power into space?

Traditional satellites act like "cameras that only take pictures": they collect raw data in orbit, beam it all down to ground stations and wait for Earth-based supercomputers to process it. Industry estimates suggest less than 10% of collected data ever gets used, with latency running to hours or days.

Space computing flips the model. Satellites are equipped with radiation-hardened chips, onboard servers and storage, linked into constellations that can collect, analyze and decide in orbit, sending back only high-value results. 

According to industry insiders, space computing mainly comprises three operating models: "in-orbit processing of space-collected data," which processes data gathered in space on board and downlinks only high-value results; "space-based processing of ground-uploaded data," which leverages the advantages of space computing to process data uploaded from the ground and "integrated space-ground collaborative computing," which builds a global distributed computing network to achieve efficient complementarity of space and terrestrial resources.

Bandwidth consumption can drop by more than 90%; response times shrink from hours to seconds. Applications range from real-time wildfire and flood alerts to crop-stress tracking, illegal-fishing detection in remote oceans and polar-ice monitoring – all without waiting for a downlink window.

Accelerating tech breakthroughs to fortify space intelligent infrastructure

Progress across computing, energy and communications is reinforcing the foundations of China's space-based intelligent infrastructure.

On the computing front, the country is prioritizing constellation-scale in-orbit clusters and homegrown anti-radiation technologies. The Three-Body Computing Constellation serves as a reference case: the 12-satellite network, equipped with domestically produced space-borne AI servers, delivers a peak of 744 TOPS per satellite and 5 POPS in aggregate, with core components fully localized through native aerospace-grade chips and tailored processes.

On the energy front, efforts center on high-efficiency space power generation and radiative thermal management. Galactic Aerospace's Lingxi 03, China's first flat-stack satellite with flexible solar wings, illustrates the direction: the "flexible wings" are compact, lightweight and modular, easier to stow and per unit mass, larger in area to capture more solar energy. On the thermal side, the company has broken through in pump-driven fluid-loop heat dissipation, technologies that Zhang Shijie, chief scientist at Galactic Aerospace, says will underpin high-power space computing payloads.

On the communication front, the focus is on a regionally resilient architecture. With dense ground-station coverage domestically and selective nodes overseas, inter-satellite laser links have stabilized at 100 Gbps and satellite-to-ground laser tests have reached 120 Gbps, while round-trip latency within China and the Asia-Pacific is held at 30 to 50 milliseconds, sufficient for time-sensitive scenarios such as emergency communications, with core devices 100% domestically controlled, according to Xu Zifan, a prominent technology analyst and the deputy director of the Advanced Computing Research Office at the Electronics Research Institute, which is part of the China Center for Information Industry Development.

The global race intensifies

The global race is intensifying. SpaceX has outlined a long-term vision of deploying up to one million low-Earth orbit (LEO) satellites as orbital data centers, while the European Union is advancing its Space Data Center initiative. Russia is enhancing onboard computing capabilities for its Sfera constellation, and Japan is investing in in-orbit processing for Earth-observation satellites. Some recent studies estimate that the LEO region between 300 and 2,000 km could safely accommodate only around 175,000 satellites under current collision-avoidance assumptions. 

Meanwhile, under the International Telecommunication Union's spectrum filing framework, satellite operators generally have seven years to bring frequency filings into use, followed by phased deployment milestones for non-geostationary constellations.

As AI models become larger and Earth observation grows exponentially, the competition is no longer about who can launch more satellites, but who can process information first. The next frontier of the digital economy may therefore be measured not only by computing power on Earth, but also by intelligence deployed in orbit.