News

Nvidia CEO Jensen Huang says Blackwell B200 ramp is on track for H1 2025 with significant inference gains

Tuesday, March 17, 2026 at 03:40 PM

Nvidia CEO Jensen Huang has confirmed that the Blackwell B200 chip ramp is scheduled for the first half of 2025. The new architecture is expected to provide four times the training performance of the H100 and thirty times the inference performance, marking a significant shift toward AI inference applications.

Context

At the GTC 2026 conference on March 18, 2026, Nvidia CEO Jensen Huang confirmed that the high-volume ramp of the Blackwell B200 GPU is on track for the first half of 2025. While the company previously navigated technical challenges and supply chain reworks in late 2024, Huang emphasized that the architecture is now the "engine of the new industrial revolution." He specifically flagged an "inference inflection," noting that Blackwell delivers up to 30x the performance of the previous Hopper generation for certain workloads. This transition is critical as inference now accounts for over 40% of Nvidia's revenue. The B200 platform is designed to reduce large language model (LLM) inference operating costs and energy consumption by up to 25x. With the launch of NVIDIA Dynamo 1.0 alongside this hardware, the company aims to sustain its dominant market position by offering a 15x better inference performance profile and 2.2x higher training throughput compared to the H100, targeting trillion-parameter scale AI models.

Related Companies

Nvidia
Nvidia
NVDA
US