At the Computex 2025 exhibition held in Taiwan, Nvidia CEO Jensen Huang unveiled the company’s next-generation artificial intelligence technologies. One of the most notable innovations was the RTX Pro Server, a new system designed for deep learning and generative AI applications. This chip delivers four times the performance of Nvidia’s previous flagship model, the H100, and has demonstrated 1.7 times greater efficiency in tests using Meta’s Llama model. The RTX Pro Server is currently in the production phase.
In addition, Nvidia introduced a new system called NVLink Fusion. This technology enables the creation of custom AI systems by linking multiple chips together. While Nvidia previously offered such capabilities only on its proprietary chip platforms, it now plans to make the technology available to third-party manufacturers.
By introducing NVLink Fusion, Nvidia is paving the way for a more open and modular AI infrastructure. The technology allows various chipmakers—such as Qualcomm and Fujitsu—to combine their CPUs with Nvidia GPUs to build high-performance AI systems. Companies like MediaTek and Marvell also plan to integrate NVLink Fusion into their custom AI chips.
NVLink Fusion opens up Nvidia’s AI platform and rich ecosystem to partners, supporting the development of custom AI infrastructures. It is designed to enhance AI system performance through high-speed data transfer and low latency.
These advancements reinforce Nvidia’s leadership in the field of artificial intelligence and reflect the company’s strategy to make AI infrastructure more expansive and flexible.
