Listen to the article
Amazon Web Services and NVIDIA deepen their strategic alliance by integrating NVLink Fusion into AWS infrastructure, promising faster AI training, expanded cloud workloads, and new AI factory services leveraging cutting-edge GPU and interconnect technologies.
Amazon Web Services and NVIDIA have deepened a multi‑year strategic partnership unveiled at AWS re:Invent, with AWS agreeing to integrate NVIDIA NVLink Fusion into its custom silicon and cloud stack to accelerate large‑scale AI training and inference. [1][2][3]
NVLink Fusion is being adopted by AWS to link NVIDIA’s vertical‑scale interconnect, the MGX rack architecture and AWS’s bespoke chips, enabling much higher bandwidth between accelerators and custom ASICs. According to the announcement, the platform supports scale‑up interconnects and NVLink Switches that can connect many ASICs at terabyte‑scale bandwidths. [1][4][5]
AWS says it is designing its next‑generation Trainium4 around NVLink and MGX integration, marking the first step in a multi‑generation collaboration intended to speed the rollout of cloud‑scale AI capabilities. Reuters reported that the move aims to improve inter‑chip communication to faster train very large models. [1][2][5]
The partners will also extend NVLink Fusion support to AWS Graviton CPUs and the Nitro System virtualization layer, so the technology can be used across inference, agent‑style model training and general‑purpose cloud workloads while remaining compatible with AWS infrastructure. [1][3]
AWS is expanding its accelerated compute portfolio on the back of NVIDIA Blackwell GPUs , including HGX B300 and GB300 NVL72 , and expects RTX PRO 6000 Blackwell server GPUs to appear on AWS in the coming weeks; these GPUs will be a core element of the new AWS AI Factories service. [1]
According to the companies, AWS AI Factories will deliver dedicated, AWS‑managed infrastructure in customer premises and sovereign‑compliant clouds, combining NVIDIA Blackwell accelerators, Spectrum‑X networking and AWS services to give organisations local control of data while accessing advanced AI capabilities. The partners said the service targets regulated and public‑sector use cases. [1]
On software, NVIDIA’s Nemotron open models have been integrated with Amazon Bedrock so developers can deploy high‑efficiency models and agent capabilities on a serverless platform, with early adopters such as CrowdStrike and BridgeWise already cited as customers. The firms also highlighted tighter integration across NVIDIA NIM microservices, NeMo toolkits and Bedrock/SageMaker workflows. [1]
AWS said it has added GPU‑accelerated, serverless vector indexing to Amazon OpenSearch Service using NVIDIA’s cuVS library, noting early adopter gains of up to 10× index speed and cost reductions to roughly one quarter of previous expenses , a change the companies say accelerates retrieval‑augmented generation and other dynamic AI workloads. [1]
NVIDIA described the collaboration as part of a virtuous cycle of compute and AI innovation. NVIDIA founder and CEO Jensen Huang said at re:Invent that rising GPU compute demand is driving smarter AI and broader applications, and that integrating NVLink Fusion with Trainium4 will create a “new generation of accelerated compute platforms” to carry advanced AI to enterprises worldwide. AWS executives framed the move as the next milestone in a 15‑year partnership that will bring greater performance, efficiency and scale to customers. [1][3]
Industry observers say the expansion combines hyperscaler silicon customisation with NVIDIA’s interconnect and GPU roadmaps, potentially shortening time‑to‑market for very large models while giving enterprises more options for sovereign and on‑prem deployments; adoption by Arm and other ecosystem partners was also highlighted as broadening the NVLink Fusion footprint. [4][5][6]
📌 Reference Map:
##Reference Map:
- [1] (Newtalk) – Paragraph 1, Paragraph 2, Paragraph 3, Paragraph 4, Paragraph 5, Paragraph 6, Paragraph 7, Paragraph 8, Paragraph 9
- [2] (Reuters) – Paragraph 1, Paragraph 3
- [3] (NVIDIA Blog) – Paragraph 1, Paragraph 2, Paragraph 9
- [4] (NVIDIA News) – Paragraph 2, Paragraph 10
- [5] (NVIDIA Developer Blog) – Paragraph 2, Paragraph 3, Paragraph 10
- [6] (Tom’s Hardware) – Paragraph 10
Source: Noah Wire Services
Access Our Exclusive Research
In-depth analysis and actionable data to keep you ahead of the curve.
Never Miss an Update. Get our latest research delivered straight to your inbox.