OFICIAL NVIDIA Newsroom

NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory

What happened
Based on NVIDIA Newsroom · Aug 26, 2026

NVIDIA introduced NVLink Fusion with NVHBM, a custom high-bandwidth memory technology, to enhance AI infrastructure performance and efficiency for hyperscalers and AI innovators.

🚩 Report a correction ↗ Open source
NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory
NVIDIA Newsroom — NVIDIA
Key points
·
The next wave of AI is placing new demands on infrastructure.
·
As AI agents and trillion-parameter workloads become mainstream, 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.
·
It will be validated and offered by leading memory partners, extending this advanced memory capability to NVLink Fusion customers.
·
Traditional HBM architectures place the memory controller on the XPU die, consuming valuable silicon area that could otherwise be dedicated to compute.
Key numbers
·
NVHBM integrates NVIDIA’s custom memory controller into the HBM base die, improving memory bandwidth by up to 30% and reducing power consumption by 15% compared to standard HBM4E.
·
The technology also frees up to 25% more area on XPU compute dies, enabling more efficient chip design for AI infrastructure.

NVIDIA announced an expansion of its NVLink Fusion platform with NVHBM, a next-generation high-bandwidth memory technology designed to meet the demands of AI workloads such as trillion-parameter models. NVHBM integrates NVIDIA’s custom memory controller into the HBM base die, improving memory bandwidth by up to 30% and reducing power consumption by 15% compared to standard HBM4E. The technology also frees up to 25% more area on XPU compute dies, enabling more efficient chip design for AI infrastructure. NVIDIA is standardizing NVHBM implementations across multiple memory suppliers to simplify integration and qualification for customers.

Amazon’s Annapurna Labs will be the first partner to adopt NVHBM as part of its collaboration with NVIDIA on NVLink Fusion, focusing on enhancing performance and efficiency for AI workloads. The partnership includes support for NVHBM technology and the NVLink scale-up architecture, building on AWS’s existing support for NVLink Fusion. Annapurna Labs plans to integrate NVLink Fusion with its next-generation Trainium chips, starting with Trainium4, enabling Amazon’s chips to work alongside NVIDIA GPUs within a common rack-scale architecture.

NVLink Fusion provides a platform for partners to connect custom XPUs and CPUs to NVIDIA’s rack-scale infrastructure, offering access to NVLink chiplets, NVLink-C2C, NVLink Switches, and NVIDIA MGX systems. The platform supports a broad ecosystem of CPU partners, ASIC designers, system manufacturers, and technology providers, facilitating collaboration across the AI infrastructure supply chain. By leveraging NVIDIA’s proven technology stack, partners can focus engineering resources on XPU innovation while reducing risks associated with scale-up and scale-out networking.

The expansion of NVLink Fusion with NVHBM aims to accelerate the deployment of semi-custom AI infrastructure for hyperscalers and AI-native companies. NVIDIA states that this approach provides a faster, lower-risk path to market by combining custom memory solutions with a standardized, validated technology stack. The initiative reflects NVIDIA’s broader strategy to address the evolving performance and efficiency requirements of next-generation AI workloads.

Original source → Deals on Clipraptor.com →