NVIDIA’s Rubin GPUs start production as HBM4 samples arrive

NVIDIA has begun producing its next-generation Rubin graphics processors while taking delivery of HBM4 memory samples from all major suppliers. These GPUs are set to become a cornerstone for AI in 2026, with volume shipments targeted for the third quarter of next year. The synchronized push on production and memory validation hints at a bid to compress timelines and lock in supply early.

Earlier, CEO Jensen Huang showcased the Vera Rubin superchip at GTC 2025, presenting two powerful GPUs alongside a next-generation CPU and LPDDR modules. Rubin is positioned to anchor AI compute platforms in data centers, and TSMC is actively preparing its lines to meet the expected surge in demand. By the look of the design, NVIDIA is signaling a tightly integrated approach to training and inference.

Alongside Rubin, NVIDIA has sourced HBM4 samples from multiple manufacturers to equip the new GPUs with modern high-speed memory modules. Strong demand for today’s Blackwell and Blackwell Ultra chips has already prompted TSMC to boost 3 nm wafer output by 50%. That surge suggests the market isn’t cooling—and that Rubin’s runway is being cleared well in advance.