Nvidia Unveils Revolutionary AI Chips at GTC 2025

At the 2025 GPU Technology Conference (GTC) in San Jose, California, Nvidia CEO Jensen Huang unveiled the company's latest advancements in artificial intelligence (AI) hardware, introducing the Blackwell Ultra and Vera Rubin AI chips. These developments signify substantial progress in AI processing capabilities, aiming to meet the escalating computational demands of modern AI applications.

The Blackwell Ultra, an enhancement of Nvidia's existing Blackwell architecture, is slated for release in the second half of 2025. This chip is designed to deliver 1.5 times the performance of its predecessor, featuring 20 petaflops of FP4 inference and an expanded 288GB of HBM3e memory. It will be integrated into Nvidia's GB300 NVL72 system, which connects 72 Blackwell Ultra GPUs and 36 Grace CPUs, offering substantial improvements in AI performance.

Following this, the Vera Rubin GPU is scheduled for launch in late 2026. Named after astronomer Vera Rubin, this chip is expected to provide twice the performance of the Blackwell Ultra, with 50 petaflops of FP4 performance. It will be part of the Vera Rubin NVL144 system, delivering 3.3 times the training and inference performance of the GB300 NVL72.

Nvidia also introduced the Rubin Ultra NVL576 platform, slated for release in the second half of 2027. This platform will connect 576 Rubin Ultra GPUs with Vera CPUs using a seventh generation of Nvidia’s NVLink, capable of 15 exaflops of FP4 inference performance and 5 exaflops of FP8 training performance, making it 14 times faster than the GB300 NVL72 platform.

To support these high-performance systems, Nvidia is launching advanced networking technologies, incorporating silicon photonics into their Quantum InfiniBand and Spectrum Ethernet switches. These innovations aim to enhance AI data centers by reducing energy use and costs, supporting up to 1.6 Tbps per port, and connecting millions of GPUs. The Quantum 3450-LD InfiniBand switch is expected in late 2025, with the Spectrum SN6810 and SN6800 Ethernet switches following in 2026.

These advancements have significant social and economic implications. Enhanced hardware capabilities will accelerate AI research and development, enabling more complex and efficient AI models. Industries reliant on AI, such as healthcare, finance, and autonomous vehicles, will benefit from improved processing power and efficiency. However, the increased power requirements of these advanced systems, with racks consuming up to 600kW, raise concerns about energy consumption and sustainability. Major tech companies are advocating for increased nuclear output by 2050 to meet these demands.

Nvidia Corporation, founded in 1993, is a leading technology company specializing in graphics processing units (GPUs) and AI hardware. Jensen Huang, co-founder and CEO of Nvidia, has been instrumental in steering the company towards AI and high-performance computing innovations. Vera Rubin, an American astronomer known for her pioneering work on galaxy rotation rates, provided evidence for dark matter. Naming the GPU after her honors her contributions to science.

"AI has made a giant leap — reasoning and agentic AI demand orders of magnitude more computing performance," said Jensen Huang, founder and CEO of Nvidia.

Nvidia's previous GPU architectures, such as Ampere and Hopper, have set benchmarks in AI processing. The Blackwell Ultra and Vera Rubin chips represent a continuation and significant advancement in this trajectory, offering unprecedented performance levels.

These developments highlight Nvidia's pivotal role in shaping the future of AI technology, with far-reaching implications across various industries and society at large.