NVIDIA Announces Full Production of Blackwell GPUs and Servers

At CES 2025, NVIDIA CEO Jensen Huang unveiled that NVIDIA and its partners have initiated full-scale production of Blackwell GPUs, designed for artificial intelligence (AI) and high-performance computing (HPC) applications. Blackwell-based servers, powered by these GPUs, are now fully deployed across all major cloud service providers, with systems designed to support data centers worldwide.

While Blackwell GPUs offer significant improvements in computational performance and performance per watt compared to the previous generation of Hopper processors for AI and HPC applications, they also bring a notable increase in power consumption. This challenge has made installation in data centers more complex, as these systems require enhanced cooling and power infrastructure. For example, a rack based on Hopper with 40 kW of power consumption would see a Blackwell-equipped rack with 72 GPUs consume up to 120 kW.

Dell became the first company to begin delivering Blackwell-based servers to select cloud service providers in mid-November last year. However, it is no longer the only company offering these advanced server solutions. With over a dozen server manufacturers now providing more than 200 different configurations, the Blackwell system is becoming highly adaptable to a wide range of data center needs.

Jensen Huang highlighted, "We now have systems from around 15 manufacturers, with roughly 200 different SKUs, including liquid cooling, air cooling, x86 processors, NVIDIA Grace CPU variants, and NVL36×2, NVL72×1 configurations. This extensive variety ensures that we can support nearly every data center worldwide."

He further emphasized the growing adoption of AI, noting, "These systems are being produced in approximately 45 factories, demonstrating the widespread adoption of AI and how the entire industry is shifting towards AI-driven computing."