These new Blackwell systems are now implemented by all major cloud service providers, ensuring their widespread availability across global data centers.

However, the power requirements for these systems are significant; a typical rack with Hopper GPUs consumes about 40 kW, while a Blackwell-based rack with 72 GPUs can consume up to 120 kW, necessitating advanced cooling and power solutions.

Dell has taken the lead by commencing shipments of Blackwell-based servers to select cloud providers since mid-November 2024.

Nvidia's manufacturing process for the Blackwell systems involves approximately 45 factories, reflecting the rapid adoption of AI technology in the industry.

Despite earlier reports of Nvidia canceling dual-rack systems due to perceived low value, current production suggests that some companies are now manufacturing or planning to produce these dual-rack configurations.

At the Consumer Electronics Show on January 8, 2025, Nvidia's CEO Jensen Huang announced the full-scale production of Blackwell GPUs, designed for artificial intelligence and high-performance computing.

Huang emphasized that the Blackwell systems are available in various configurations, including liquid-cooled and air-cooled options, as well as versions featuring x86 and Nvidia Grace CPUs.

Nvidia provides over 200 configurations from more than a dozen server manufacturers, allowing for versatility to meet various data center requirements.

While Blackwell GPUs offer substantial improvements in compute performance and efficiency over the previous Hopper generation, their higher power consumption poses challenges for data center installations.