Shanghai-based chipmaker Iluvatar CoreX has unveiled a multi-year GPU architecture roadmap that lays out when it expects to surpass Nvidia’s platforms, a rare level of specificity in the semiconductor sector.
Rather than leaning on conventional benchmark comparisons, Iluvatar has tied its ambitions to a public timeline, inviting direct comparison with Nvidia’s generational cadence.
According to the roadmap, Iluvatar said its current Tianshu architecture has already met the first target. Citing data reported by 36Kr, the company said that in workloads related to DeepSeek’s V3 large model, Tianshu delivered about 20% higher average performance than Nvidia’s Hopper platform, achieving more than 90% utilization of computing power. The company described these results as measured performance in real-world workloads rather than peak theoretical figures.
The roadmap extends this trajectory through successive designs. Following Tianshu, the Tianxuan architecture is intended to compete with Nvidia’s upcoming Blackwell generation, while Tianji is positioned to surpass it in 2026. Tianquan, the fourth architecture in sequence, is projected to outperform Nvidia’s Rubin platform in 2027. Beyond that, Iluvatar said it plans to move toward what it called a “breakthrough” architectural design, according to SCMP.
All four architectures are named after Chinese terms associated with the Big Dipper constellation. Iluvatar said it will release multiple products under each generation over the coming years.
Alongside its roadmap, the company highlighted commercial milestones to reinforce its claims. By June 2025, it reported delivering more than 52,000 general-purpose GPUs (GPGPUs) to over 290 customers across industries. Company disclosures indicated that its 2024 shipments ranked third among domestic GPGPU suppliers in China. Iluvatar has also expanded into edge computing, introducing four products under its Tongyang (TY) series with performance ranging 100–300 TOPS (tera operations per second). The company said its entry-level TY1000 outperformed Nvidia’s AGX Orin in several testing scenarios.
The announcement follows Iluvatar’s Hong Kong IPO on January 8. Its shares recently closed at HKD 179.2 (USD 23), just below the debut price of HKD 190.2 (USD 24.4), giving it a market capitalization of about HKD 45.5 billion (USD 5.8 billion).
According to its prospectus, Iluvatar’s revenue rose from RMB 189 million (USD 26.5 million) in 2022 to RMB 540 million (USD 75.6 million) in 2024, though net losses also widened over the same period. The figures reflect the challenge of balancing expansion and profitability faced by many chipmakers.
The timing of the roadmap coincides with ongoing uncertainty over Nvidia’s ability to sell its most advanced GPUs in China, a backdrop that has encouraged domestic suppliers to position themselves as alternatives. After years of emphasizing local breakthroughs, mass production, and deployment, Iluvatar is now anchoring its ambitions to a public schedule, raising the stakes on execution.