Nvidia began shipping samples of accelerators based on Blackwell architecture

Nvidia will soon begin full-scale deliveries of equipment based on Blackwell chips for artificial intelligence systems. At the SIGGRAPH conference in Denver, America on July 29, the company presented several updates to its software solutions and announced that it had begun shipping samples of AI equipment on Blackwell chips, Commercial Times reports.

Image source: NVIDIA

Nvidia CEO Jensen Huang believes that the Blackwell series of accelerators will be the most successful product in the company’s history – it is expected that cloud service providers will begin to build new data centers with servers for artificial intelligence. These chips will also affect other segments of the technology industry: demand for TSMC’s 4 nm process will increase; Water cooling technologies will begin to actively spread – their penetration will reach 10% and will play into the hands of such companies as Asia Vital Components, Auras Technology, Delta Electronics and Cool IT.

The new AI chips are expected to begin shipping to customers in the fourth quarter, with full-scale production planned for 2025. This will benefit the assembly plants of Wistron and Ingrasys (part of Foxconn), which produce substrates, computing and communications boards. Orders for rack-mount systems from Wiwynn, Quanta, Gigabyte, Asus and ASRock will increase – while Quanta, Wiwynn and Inventec have already announced related products with deliveries in the fourth quarter and further increase in volumes in the first half of next year.

In 2025, 60,000 Nvidia GB200 NVL36 rackmount systems will ship, and Blackwell GPU adoption will range from 2.1 million to 2.2 million units, making Blackwell the primary platform that will be present in 80% of Nvidia’s high-end systems, TrendForce analysts predict . The GB200 NVL36 systems will use a combination of air and liquid cooling solutions, while the NVL72 will operate primarily on liquid cooling.