Realtime AI News
China's Answer for Space Computing: Photons Over Electrons — Optical Computing Startup Partners for Space-Grade AI Chips
Shanghai-based optical computing company Guangbenwei (Optical Computing Tech) has partnered with Dongfang Tianxuan to develop space-based photonic computing payloads, using photons instead of electrons to bypass radiation, heat dissipation, and power challenges in orbit.
The space computing race is intensifying. Elon Musk has predicted that by 2032, solar-powered AI satellites will become the most cost-effective computing solution globally, while NVIDIA CEO Jensen Huang stated that intelligence must exist wherever data is generated. However, space computing faces harsh engineering challenges — no air convection for cooling, and high-energy particles that can cause chip errors.
Shanghai-based optical computing company Guangbenwei (Optical Computing Tech) is taking a different approach. In partnership with Dongfang Tianxuan, they have initiated development of a space-based photonic computing payload, marking the first time optical computing has been applied to space engineering. Photons, unlike electrons, carry no charge and are naturally immune to high-energy particle interference. Light propagating through waveguides generates almost no heat, solving the cooling challenge in vacuum environments. Additionally, photonic chips have near-zero static power consumption, making them inherently suited for satellites with limited energy.
The same payload weight can deliver higher total compute with photonics. Guangbenwei is currently the world's only company to have simultaneously achieved in-memory photonic computing. Their photonic-electronic fusion computing card delivers 300 TOPS, supports INT8 and FP8 inference, and has already begun in-orbit environmental testing. A second-generation card is planned for release this year.
For its technical approach, Guangbenwei chose glass over silicon as the chip substrate, fundamentally overcoming the size, warpage, and interconnect limitations of silicon photonics platforms. According to the company's research institute vice president Pu Huanan, photonic computing does not rely on EUV lithography — existing 45nm or even sub-micron processes are sufficient.
The project, reported by QbitAI, aims to create the world's first space-based photonic computing satellite, pushing space computing from a theoretical concept to a verifiable, iterative engineering pathway.
Why it matters
Photonic computing could become a differentiated technological route in the space computing race, providing a new computing foundation for on-orbit AI inference and large model operations while lowering the engineering barriers to space-based computing.