The collaboration will allow companies around the world to access advanced AI systems through AWS.

Amazon Web Services (AWS) is deepening its partnership with Nvidia, announcing that its future generation of Trainium AI chips will incorporate NVLink Fusion, one of Nvidia’s flagship interconnect technologies. The move signals AWS’s intention to strengthen its position in the rapidly intensifying AI infrastructure market and attract enterprises building large-scale AI systems.

The announcement was made at AWS re:Invent 2025 in Las Vegas, where the company said the NVLink-enabled Trainium4 chip will deliver dramatically faster communication between processors, enabling larger, more tightly connected AI training clusters. AWS did not specify when Trainium4 will debut, but emphasized that NVLink Fusion will be foundational to its next wave of AI hardware.

Nvidia has been encouraging chipmakers to adopt NVLink to create unified, high-bandwidth systems capable of scaling to thousands of GPUs. With AWS joining Intel and Qualcomm as adopters, Nvidia is expanding the influence of its interconnect technology across the industry.

Nvidia CEO Jensen Huang described the partnership as a step toward “building the compute fabric for the AI industrial revolution,” saying the collaboration will allow companies around the world to access advanced AI systems through AWS. As part of the tie-up, customers will also gain access to AI Factories, dedicated AI compute clusters that AWS will deploy inside customer data centers for faster, more secure model development.

Challenges to deploying custom AI silicon

AI workloads are getting larger, models are becoming more complex, and the pressure to rapidly deploy AI compute infrastructure that meets the needs of the growing market is higher than ever.

Emerging workloads like planning, reasoning and agentic AI, running on hundreds of billions- to trillion-parameter models and mixture-of-experts (MoE) model architectures require many systems with many accelerators all working in parallel, and connected in a single fabric.

Amazon hopes to entice companies looking for a bargain. Trainium chips are capable of powering the intensive calculations behind AI models more cheaply and efficiently than Nvidia’s market-leading graphics processing units, according to the company. “We’ve been very pleased with our ability to get the right price performance with Trainium,” Brown said.

Amazon is releasing Trainium3 about a year after deploying its predecessor accelerator. That’s a sprint by chip industry standards. “The main thing we’re gonna be hoping for here is just that we don’t see any kind of smoke or fire,” an AWS engineer joked when the chip was first fired up in August.

By - Aaradhay Sharma