China's Nayuta Space Unveils Alaya AI Compute Satellite Constellation

China's Nayuta Space has introduced the 'Alaya' orbital computing constellation, intending to deploy 12,500 artificial intelligence compute satellites. This GW-class system integrates satellites and rockets for operation in sun-synchronous orbit, addressing terrestrial data centre limitations.

Credit: NAYUTA SPACE

Terrestrial data centres face increasing pressure from power supply, cooling requirements, and land availability as demand for artificial intelligence training and inference grows. JP Morgan data indicates that large language model token call volume increased twentyfold year-on-year by June. Goldman Sachs projects global monthly token consumption will reach 120 quintillion by 2030, a twenty-four-fold rise from the figures seen in the current year. The search for new compute infrastructure capacity is increasingly leading organisations, including many Big Tech firms, to consider space-based solutions as a viable alternative.

Nayuta Space publicly presented its "Alaya" system at the Global Digital Economy Conference during June. The GW-class orbital computing constellation integrates satellites and rockets. It plans for 12,500 compute satellites in a sun-synchronous orbit, designed to use the inherent benefits of space. Space offers near-continuous solar power along the terminator orbit, with very brief eclipse periods. So, extreme low-temperature environments also improve heat dissipation efficiency significantly, adhering to the Stefan-Boltzmann law.

Big Tech firms are also working on similar space-based initiatives. SpaceX proposed its Starmind plan to the FCC in Jan., suggesting millions of orbital data centre satellites. Blue Origin launched Project Sunrise, targeting 51,600 data centre satellites. Google announced its "Sun Catcher" programme to send proprietary TPU chips into orbit. NVIDIA released the Space-1 Vera Rubin space computing module.

Nayuta Space's approach centres on its proprietary "Xuan Niao-R" (玄鸟-R) reusable rocket, employing Aerodynamic Deceleration-Horizontal Landing (ADHL) technology. This system distinguishes itself from SpaceX Falcon 9's retropropulsive vertical landing. ADHL utilises aerodynamic drag during atmospheric reentry to decelerate the rocket. This method reduces the fuel penalty for recovery to below 3 per cent, which is a considerable decrease compared to the significant payload reduction seen with traditional propulsive landing techniques. Company estimates suggest this approach can deliver a 30 per cent overall payload improvement, adding approximately 4 to 6 tonnes per heavy-lift launch.

The cost considerations for orbital computing are extremely sensitive to launch expenses. Nayuta Space actively pursues aggressive cost reduction strategies through industrial supply chain integration. The firm works with automotive-grade cable manufacturers to introduce mass manufacturing efficiency directly into rocket production processes. But, with stable first-stage recovery of the Xuan Niao-R and subsequent full-rocket recovery of the planned Xuan Niao-FR heavy liquid rocket, the company projects launch costs dropping to hundreds of yuan per kilogram. This development could make space-based AI computing economically viable at a large scale, according to Pandaily.

• Nayuta Space plans to deploy 12,500 AI compute satellites in orbit.

• The "Alaya" constellation seeks to address increasing demands for AI computing beyond terrestrial data centre limitations.

• The company's "Xuan Niao-R" reusable rocket uses Aerodynamic Deceleration-Horizontal Landing technology to reduce launch costs and increase payload capacity.

• Other Big Tech organisations are also pursuing orbital data centre projects.

• Nayuta Space aims to reduce launch costs to hundreds of yuan per kilogram through supply chain integration.

Source: Pandaily