NextFin News - On January 26, 2026, Shanghai-based Iluvatar CoreX Semiconductor Co., Ltd. officially unveiled its fourth-generation chip architecture roadmap, setting a bold target to overtake NVIDIA’s high-end performance benchmarks within the next two years. According to Pandaily, the roadmap outlines a phased progression: the Tianshu architecture is slated to surpass NVIDIA’s Hopper in 2025, followed by the Tianxuan and Tianji architectures targeting the Blackwell series in 2026. By 2027, Iluvatar expects its Tianquan architecture to exceed the performance of NVIDIA’s anticipated Rubin architecture, marking a transition toward what the company describes as "breakthrough computing chip architectures."
This strategic announcement follows Iluvatar’s high-profile debut on the Main Board of the Hong Kong Stock Exchange on January 8, 2026. Trading under the stock code 9903.HK, the company saw its shares surge 31.54% on the first day, pushing its market capitalization beyond $5.2 billion. The capital injection from the IPO is being funneled directly into R&D for the Tongyang series of edge computing products. Notably, the company’s TY1000 product has already demonstrated performance metrics exceeding NVIDIA’s AGX Orin in computer vision and natural language processing tasks, specifically when running the DeepSeek 32B large model.
The aggressive timeline presented by Iluvatar reflects a broader trend of "domestic substitution" within the Chinese semiconductor industry, accelerated by the geopolitical landscape under U.S. President Trump. As the first Chinese firm to achieve mass production of general-purpose GPUs (GPGPUs) for both training and inference, Iluvatar is positioning itself as the primary alternative for domestic cloud service providers and data centers. The company’s "TG" series for training and "ZK" series for inference are designed with a "hardware-software integration" philosophy, ensuring compatibility with mainstream programming ecosystems to lower migration costs for enterprise clients.
From an analytical perspective, Iluvatar’s ability to meet these targets depends heavily on its R&D efficiency. Financial data indicates that the company maintains an R&D-to-revenue ratio between 80% and 120%, which is relatively lean compared to global peers. This suggests a highly focused development cycle but also raises questions about the sustainability of such rapid architectural leaps. Surpassing NVIDIA’s Rubin—a platform expected to define the next era of AI supercomputing—requires not just raw FLOPS (floating-point operations per second) but also sophisticated interconnect technologies and memory bandwidth capabilities that have historically been the Achilles' heel of emerging GPU designers.
The market’s reception of Iluvatar, evidenced by its 414-fold oversubscription during the IPO, underscores the scarcity of high-end GPGPU assets in the Asian equity markets. However, the path to 2027 is fraught with execution risks. While the TY1000’s success against the AGX Orin proves Iluvatar can compete in edge computing, the high-end data center market is a different arena. NVIDIA’s Blackwell and Rubin architectures are not static targets; they represent integrated ecosystems of software (CUDA), networking (InfiniBand), and silicon. For Iluvatar to truly "surpass" these benchmarks, it must ensure its software stack remains robust enough to support the increasingly complex requirements of trillion-parameter models.
Looking forward, the success of the Tianquan architecture will likely serve as a bellwether for the entire Chinese AI chip sector. If Iluvatar can deliver on its 2027 promise, it will validate the feasibility of independent GPGPU development paths. Conversely, any delays in the 2025 or 2026 milestones could dampen investor enthusiasm and slow the adoption of domestic computing backbones. As U.S. President Trump continues to emphasize American technological leadership, the competition between Iluvatar and NVIDIA will be viewed not just as a corporate rivalry, but as a critical data point in the global struggle for AI supremacy.
Explore more exclusive insights at nextfin.ai.
