Iluvatar CoreX unveils Tiangai 300 supernode packing 144 GPUs into a single rack

Chinese company Iluvatar CoreX has unveiled the Tiangai 300 IXL144 — its first supernode, setting a new industry record for GPU integration. The system packs 144 GPUs into a single rack, surpassing every known alternative on the market. The official debut took place at the China International Industry Fair (CIIF) in Shanghai on Monday, October 12, 2026.

Technical specifications of the Tiangai 300 IXL144

The Iluvatar CoreX supernode is configured with the following hardware:

• 144 graphics processing units (GPUs) — the primary compute resource for parallel processing • 18 central processing units (CPUs) — for system management and auxiliary task handling • Approximately 10.1 terabytes of RAM — sufficient capacity for working with large AI models

Concentrating 144 GPUs in a single system is the highest figure in the history of the computing systems industry. This achievement marks a notable step forward in hardware integration and optimization for artificial intelligence workloads.

Purpose and architecture

The Tiangai 300 IXL144 is designed for two primary scenarios: large-scale AI model training and subsequent inference in production. The supernode is a deeply integrated system that combines critical components:

• Compute core — performing mathematical operations during training and model operation • Data storage system — managing large volumes of training datasets and results • Network infrastructure — synchronizing and exchanging data between components • Management and monitoring — overseeing hardware performance and status • Software and frameworks — simplifying model development and deployment

This architecture allows all components to be deeply integrated into a single supercomputer, ensuring optimal interaction between compute, storage, networking, and software.

Context and strategic significance

The launch of the Tiangai 300 IXL144 coincides with two key trends. The first is the worldwide growing demand for AI computing resources and the sharp rise in model costs. The second is the tightening of US export restrictions on advanced semiconductors, particularly high-performance Nvidia chips.

Under these conditions, supernodes are becoming a critical alternative. Instead of importing individual chips with the risk of logistical disruptions, companies are deploying local interconnected systems, scaling compute power by integrating multiple rack-based systems without dependence on American suppliers.

Iluvatar CoreX claims the Tiangai 300 IXL144 cuts token costs by 40% — a significant reduction in operating expenses when working with generative models. This advantage makes the supernode an attractive solution for Chinese companies actively investing in local AI infrastructure and seeking alternatives to American systems.