Chinese GPU startup Iluvatar CoreX has unveiled a multi-generation product roadmap that it says could see its fourth-generation architecture surpass Nvidia's upcoming Rubin platform by 2027, marking one of the latest efforts by domestic Chinese chipmakers to advance their AI computing capabilities.
According to the company's disclosed timeline, Iluvatar CoreX plans to introduce its "Tianshu" architecture in 2025, targeting performance beyond Nvidia's Hopper platform, including the H200 series. In 2026, the company expects its "Tianxuan" architecture to compete with Nvidia's Blackwell generation, including the B200, followed later the same year by the "Tianji" architecture, which aims to exceed Blackwell-class performance.
By 2027, Iluvatar CoreX plans to roll out its "Tianquan" architecture, which is designed to surpass Nvidia's Rubin platform. Beyond 2027, the company said it will shift toward developing more breakthrough computing chip architectures.
Chinese companies have previously made similar commitments to compete with Nvidia's future AI platforms, including Huawei Technologies, as domestic players seek to strengthen local alternatives.
Edge products deliver competitive performance
Alongside the roadmap, Iluvatar CoreX also unveiled its "Tongyang" series of edge computing products. The company said tests conducted in real-world scenarios, including computer vision, natural language processing, and inference for large language models such as DeepSeek 32B, showed its TY1000 chip delivering performance above Nvidia's AGX Orin.
Positioning strategy sets it apart from rivals
Iluvatar CoreX said it aims to position itself as a supplier of high-end general-purpose GPUs and large-scale computing systems, distinguishing its approach from domestic rivals such as Biren Technology and Moore Threads.
Biren focuses on developing general-purpose GPUs designed for training, inference, and scientific computing based on unified architectures. Moore Threads, by contrast, has pursued a full-function GPU strategy covering AI acceleration, professional graphics, desktop GPU,s and intelligent system-on-chip products.
Put simply, Moore Threads is building a full-function GPU ecosystem, while Iluvatar CoreX and Biren are more clearly segmenting their product lines around computing workloads, though with different technical approaches.
As of June 30, 2025, Iluvatar CoreX had delivered more than 52,000 GPUs to approximately 290 enterprise customers, with applications spanning sectors including finance and healthcare.
Supply chain hurdles persist
Like other Chinese AI chip developers, Iluvatar CoreX faces challenges related to manufacturing capacity and supply-chain access. Further performance gains will depend on continued advances in process technology and packaging.
China's push to develop domestic alternatives to Nvidia continues as demand for AI computing grows across cloud, enterprise, and edge deployments.
Credit: MyDrivers
Article edited by Jerry Chen
