Nvidia Powers the Future of AI Factories
Nvidia will introduce a new lineup of AI chips every year, according to CEO Jensen Huang. The next release, slated for 2026, will be called Rubin and will feature networking ICs, a GPU, and a CPU named Versa.
Companies such as ASRock Rack, ASUS, GIGABYTE, and others will use Nvidia GPUs and networking to offer AI systems for cloud, on-premises, embedded, and edge applications.
Huang mentioned that we are at the beginning of a new industrial revolution, with companies and nations collaborating with Nvidia to transform traditional data centers into AI factories, thereby creating artificial intelligence as a new commodity.
The new offerings will include various configurations, ranging from single to multiple GPUs, using both x86 and Grace-based processors, and supporting both air and liquid-cooling technologies.
To speed up system development, Nvidia’s MGXTM modular reference design platform now supports Nvidia Blackwell products. This includes the Nvidia GB200 NVL2 platform, designed for mainstream large language model inference, enhanced data retrieval, and data processing tasks.
The GB200 NVL2 is ideal for data analytics, a market where companies invest tens of billions annually. Leveraging NVLink-C2C interconnects and the specialized decompression engines in the Blackwell architecture, it boosts data processing speeds by up to 18 times, with an 8-fold improvement in energy efficiency compared to x86 CPUs.
Nvidia’s MGX reference architecture allows manufacturers to quickly and cost-effectively create over 100 system design configurations. These manufacturers can start with a base system architecture for their servers and then choose the appropriate GPU, DPU, and CPU for different workloads. Currently, more than 90 systems using the MGX architecture are being developed, up from 14 systems from six partners last year.
AMD and Intel are also supporting the MGX architecture, with plans to introduce their own CPU host processor modules, like AMD’s next-generation Turin platform and Intel’s Xeon 6 processor (formerly known as Granite Rapids).
The GB200 NVL2 platform, part of Nvidia’s latest offerings, uses MGX and Blackwell technologies for versatile system configurations and networking options, making it easier to integrate accelerated computing into current data centers. It joins the Blackwell lineup that includes Blackwell Tensor Core GPUs, Grace Blackwell Superchips, and the GB200 NVL72.
Nvidia’s ecosystem partners include TSMC, the world’s leading semiconductor manufacturer, and other global electronics companies that provide essential components for AI factories, such as server racks, power supplies, and cooling solutions from companies like Amphenol, AVC, Cooler Master, and more.
This collaborative effort ensures the rapid development and deployment of new data center infrastructure to meet enterprise demands, further enhanced by Blackwell technology, Nvidia’s Quantum-2 and Quantum-X800 networking, SpectrumTM-X Ethernet networking, and BlueField-3 DPUs, as featured in servers from Dell, Hewlett Packard Enterprise, and Lenovo.
Enterprises can also leverage the Nvidia AI Enterprise software platform, which includes Nvidia NIMTM for inference tasks.
