Newly-listed Chinese chipmaker targets beating Nvidia Rubin platform in just two years — Shanghai Iluvatar CoreX unveils multi-year GPU architecture roadmap with 2027 deadline

Chinese chip designer Shanghai Iluvatar CoreX Semiconductor has unveiled a multi-year GPU architecture roadmap that explicitly targets Nvidia’s next-gen Rubin platform as its main competition, as China continues to push hard to establish viable domestic alternatives to Western silicon for AI training and inference hardware. According to reporting by the South China Morning Post, Iluvatar CoreX’s roadmap outlines four successive GPU architectures, named after Chinese terms associated with the Big Dipper constellation.

The company has said its current architecture, Tianshu, has already outperformed Nvidia’s Hopper generation, while a follow-on design, Tianxuan, has been pitted against Blackwell. A third generation, Tianji, is earmarked as being capable of surpassing Blackwell this year, with a fourth, Tianquan, projected to exceed Rubin by 2027, claims the company. Following this, CoreX will attempt what it describes as a “breakthrough” architectural redesign.

According to Iluvatar CoreX, Tianshu achieves more than 90% effective utilization of compute resources through architectural features that reduce redundant memory access and dynamically allocate workloads to ease resource contention. The company claims a roughly 20% higher average performance than Hopper on DeepSeek V3 through its own testing, a claim that comes conveniently soon after Nvidia’s China business was dealt a new blow by the rejection of H200 GPUs by Chinese officials. Nvidia CEO Jensen Huang arrived in China on Friday, January 23, for meetings with partners and customers, amid efforts to stabilize the company’s position in the market.

Beyond data-center GPUs, Iluvatar CoreX has also unveiled four edge-focused components under its Tongyang (TY) series, spanning 100 to 300 TOPS. The company claims its TY1000 exceeded Nvidia’s Jetson AGX Orin in a range of test scenarios, but no detailed third-party benchmarks have been published — a suspiciously common theme when Chinese manufacturers make such lofty performance claims.

Founded in 2015, Iluvatar CoreX introduced its first general-purpose training GPU, TG Gen 1, which it described as China’s first domestically mass-produced AI training GPU. A second-gen component followed in 2023, while TG Gen3 is scheduled for mass production later this year. The company also launched an inference-oriented ZK product line in 2022.

The company was listed in Hong Kong on January 8, joining a wave of Chinese companies going public all at the same time, and is currently valued at around HK$46.3 billion based on current market capitalization. That places Iluvatar CoreX well below its larger domestic peers but is still significant for a firm of its age and size. In the first half of 2025, the company reported revenue of 324 million yuan and reportedly shipped more than 52,000 general-purpose GPUs.

Whether the company’s ambitious Rubin-era claims come to fruition remains to be seen, but, at least on paper, its roadmap demonstrates just how aggressively China’s domestic tech companies are organizing their ambitions against Nvidia.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.