Nvidia’s upcoming GPU Technology Conference in the USA is anticipated to center on advancements in artificial intelligence (AI) servers, with critical updates regarding the Blackwell chip, the NVL72 AI server system, and various data center networking solutions.
According to reports from Investing and analyst Ming-Chi Kuo of TF International Securities, the conference could significantly influence the AI market, especially in light of recent declines in technology stocks. A key highlight of the event is expected to be the introduction of the Blackwell chip, which serves as the successor to the previous Blackwell series. This new chip line will feature two variations: dual-die (CoWoS-L) and single-die (CoWoS-S), with HBM memory capacity expected to increase from 192 GB to 288 GB, resulting in a projected performance boost of 50% over its predecessor. Nvidia plans to commence testing production of the Blackwell chip in the second quarter of 2025, with mass production set to begin in the third quarter.
The conference is also likely to provide updates on the NVL72, a product generating considerable interest in the market. Should Nvidia present real-world data on NVL72 deployment in data centers and elaborate on the upgrade transition from the older chip versions to Blackwell, it could significantly reshape investor expectations regarding this new product line.
In addition to the Blackwell chip, Nvidia is set to unveil the NVL72 AI server system, which is designed to match the existing rack size and power requirements of the previous NVL32 model. This compatibility will allow for seamless integration into current data center setups, with a prototype of the NVL72 expected to be launched in June 2025.
Furthermore, Nvidia may announce several innovative solutions within the data center networking domain, including the Quantum-3, Quantum-X800, Spectrum-5, and ConnectX-8 (CX8). The CX8 aims to double the speed of its predecessor while enabling the use of SuperNICs and PCIe switches, contributing to a reduction in power consumption by approximately 30%.
While Edge AI continues to be a significant trend in development, the primary focus of the conference is likely to remain on the AI server segment. Solutions related to AI PCs, such as N1X and N1, may be introduced later at Computex 2025 in Taiwan.