NVIDIA's next-gen B300 AI GPU expected to launch in Q3 2025, mass production in Q4 2025

NVIDIA's next-gen B300 AI GPU is expected to launch in Q4 2025 with mass production expected in Q4 2025 according to the latest reports.

In a new article published by DigiTimes, we're to expect NVIDIA's unveiling of its new Blackwell B300 AI GPU at GTC 2025 (its in-house GPU Technology Conference) in March. NVIDIA will reportedly mention the first CPO technology-based InfiniBand switch that will debut in Q3 2025, supporting 36 x 3.2T CPO optical modules and 4 x 28.8T application-specific integrated circuits (ASICs) that will reach a blistering 115.2T signal transmission.

NVIDIA's new B300 AI GPU will reportedly chow down on up to 1400W of power, with 1.5x the performance in FP4 performance on a single AI GPU over the B200 AI GPU. We can expect the HBM capacity of each GPU to increase from the max of 192GB HBM3E on the B200, to a much bigger 288GB of HBM3E on B300.

The current-gen NVIDIA GB200 AI GPU features 192GB of HBM3E memory on an 8-Hi stack configuration, but the new GB300 will have a larger 288GB pool of HBM3E memory on an improved 12-Hi memory stack. Not only that, but the number of fast connector components and network cards has also been upgraded on GB300 AI server platforms, with the optical module improved again from 800G to an ultra-fast 1.6T.

The performance and equipment have been improved in all aspects according to recent reports, that B300 is NVIDIA's next "market-grabbing weapon'. All eyes are now on NVIDIA GTC 2025 in March, where we'll be introduced to the new B300 family of AI GPUs.