A Chinese AI Chip Startup Unveils Ambitious Roadmap to Compete with NVIDIA’s Advanced Vera Rubin by 2027

Chinese AI chip startup Iluvatar CoreX has rolled out an ambitious plan that aims to achieve performance levels comparable to NVIDIA’s Vera Rubin in the coming years. This announcement signals a competitive shift in the AI chip industry, particularly as China escalates its efforts to enhance its computational capabilities.

Iluvatar CoreX Aims for Competitive Edge Against Vera Rubin by Next Year

As part of the broader race against Western technological advancements, China is aggressively pursuing improved compute capabilities, whether through domestically produced technologies, rental solutions, or even chip smuggling. With a significant stake in the global AI landscape, notable Chinese firms such as Huawei, Moore Threads, and BirenTech are striving to equip local hyperscalers with viable domestic technologies. Recently, a report by MyDrivers reveals that Iluvatar CoreX, a new player in the market, is set to challenge NVIDIA’s Blackwell architecture this year, aiming even higher with plans to compete with Vera Rubin by next year.

Iluvatar CoreX distinguishes itself as China’s premier HPC (High-Performance Computing) focused company, unlike its rivals, which cater to both consumer electronics and AI sectors. While specific strategies for competing with NVIDIA, the leading provider of AI infrastructure, have not been disclosed, the startup plans to introduce a native architecture under the “Tianshu Zhixin”brand. Currently, Iluvatar CoreX claims to offer solutions that rival NVIDIA’s Ampere series through models such as TianGai-100 and TianGai-150, although comprehensive details about these products remain scarce.

An Iluvatar CoreX roadmap image shows chips labeled '天数天枢, ' '天数天璇, ' and '天数天权' with milestones from 2025 to 2027 — Image Credits: MyDrivers

This latest announcement is not Iluvatar’s first bid to outpace NVIDIA; similar intentions to rival the Vera Rubin AI lineup were previously voiced by Huawei. Their future Atlas 950 and Atlas 960 ‘SuperPoDs’ are touted for their high-density rack solutions, which can accommodate up to 8, 192 Ascend 950 AI chips. Through this innovative setup, Huawei has expressed its intention to compete with NVIDIA’s Vera Rubin NVL144 configurations. While intriguing, these claims must take into account various challenges, such as power limitations and thermal requirements, which are critical in the performance of such technologies.

A person stands on stage showcasing various open server units with visible cooling systems and hardware components. — Image Credits: NVIDIA

Despite the ambitious plans laid out by Chinese AI chip startups like Iluvatar CoreX, a significant barrier remains: the lack of a robust semiconductor ecosystem comparable to that of Western nations. While innovative architectures may be proposed, the effectiveness of these advancements hinges on solid production capabilities. Without this critical infrastructure, many of these initiatives may remain aspirational rather than practical.

Source & Images