Chinese AI chipmaker Iluvatar CoreX debuted its first supernode system, a computing system that squeezes 144 graphics processing units (GPUs) into a single cabinet, breaking an industrial record.
The new product, dubbed the Tiangai 300 IXL144, was unveiled on Monday at the 2026 China International Industry Fair (CIIF) in Shanghai. Designed for large-scale AI model training and inference, the system packs 144 GPUs and 18 central processing units (CPUs) into a single cabinet and boasts around 10.1 terabytes of memory, according to the company.
Supernodes deeply integrate computation, storage, networking, management and software frameworks into a single supercomputer. As US export controls continue to restrict access to advanced chips from Nvidia alongside surging AI demand, the proliferation of interconnected cabinet systems is increasingly seen as a resource to scale computing power.
Iluvatar said that the system had a flexible single-cabinet architecture and that its efforts in improving efficiency could lower the cost per 1 million valid tokens by 40 per cent compared with “mainstream international alternatives”.
