2026-09-10 · Fuente：Nvidia Blog
NVIDIA says its NVLink Fusion ecosystem is adding NVHBM custom high-bandwidth memory as AI agents and trillion-parameter workloads go mainstream.
NVIDIA has moved to broaden its NVLink Fusion program with the introduction of NVHBM, a custom high-bandwidth memory offering aimed at the builders of large-scale AI systems. The announcement, published on the company's blog, frames the move as a response to what NVIDIA describes as a next wave of AI that is placing new demands on the infrastructure layer. The company positions NVHBM as an extension of NVLink Fusion rather than a standalone product line.
The demand picture NVIDIA sketches is one in which AI agents and trillion-parameter workloads become mainstream. That shift, in the company's framing, changes what infrastructure must deliver. As agentic systems and extremely large models move toward mainstream use, the demands placed on the hardware layer grow accordingly — and NVIDIA's answer is to treat memory as a first-class design element alongside processors, through a custom high-bandwidth memory option integrated into the NVLink Fusion ecosystem.
Central to the announcement is a systems-level argument. NVIDIA contends that the performance of AI infrastructure depends not only on compute, but on how compute, memory, storage, networking and software are designed together as a unified system. That framing has become a recurring theme in the company's infrastructure messaging, and NVHBM extends it into the memory tier: rather than treating high-bandwidth memory as a commodity component, NVIDIA is folding it into the same co-design program that governs its interconnect strategy. For data center architects, the implication is that memory choices are increasingly inseparable from interconnect and system choices.
The stated audience is hyperscalers and AI innovators building the next generation of AI infrastructure. By extending NVLink Fusion with custom high-bandwidth memory, NVIDIA is signaling that the program is intended for customers who want tailored system building blocks rather than purely off-the-shelf configurations. The expansion also underscores how far the AI hardware conversation has shifted: with trillion-parameter workloads becoming mainstream, the company is arguing that competitive advantage will accrue to platforms designed as unified systems, not to individual components in isolation.
Etiquetas: NVIDIA · NVLink Fusion · NVHBM · High-Bandwidth Memory · AI Infrastructure
萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado