During the release of NVIDIA’s latest financial report, CEO Jensen Huang confirmed that the company’s next-generation AI chip architecture, Blackwell Ultra, remains on track for a second-half 2025 launch. Additionally, he revealed that Vera Rubin, the architecture following Blackwell Ultra, is already in development and is expected to go into production in 2026.
The rollout of the current Blackwell architecture for data centers faced delays late last year due to chip design issues, which forced NVIDIA to make adjustments. However, despite this setback, as Huang emphasized, it wasn’t significant enough to derail NVIDIA’s long-term roadmap.
“Blackwell Ultra is scheduled for the second half. The next train is Blackwell Ultra with new networking, new memory and, of course, new processors. (…) We have already revealed and we are working closely with all our partners in the next click. The next click is called Vera Rubin and all our partners are preparing for the transition to him. Big, huge advance. “
More details about architectures at GTC 2025
NVIDIA CEO Jensen Huang remains highly optimistic about the future of AI chips, the segment that has propelled the company to become one of the most valuable in the world. He has promised to unveil more details during GTC 2025, scheduled for March 17–21 in California, USA—an event focusing heavily on AI advancements. NVIDIA will also share insights on the upcoming Vera Rubin architecture during this conference.
Among the key innovations in NVIDIA’s next-generation AI chips, one major highlight is the implementation of HBM3E 12H memory in Blackwell Ultra and the transition to HBM4 memory in Vera Rubin, promising significant performance and efficiency improvements.
