Amazon partnering with Nvidia for new AI chips

Amazon Web Services said Tuesday it will use a key Nvidia technology in future generations of its artificial intelligence chips as it seeks to draw major AI customers to its cloud platform.

AWS said it plans to add Nvidia’s NVLink Fusion to a forthcoming chip called Trainium4. The company did not provide a release date. NVLink creates high-speed connections between different types of chips and is one of Nvidia’s most valuable technologies.

The announcement was made during AWS’s annual cloud conference in Las Vegas, an event that draws about 60,000 attendees. Nvidia has been working to sign more chip developers to NVLink, with Intel, Qualcomm and AWS now participating.

AWS said the technology will enable construction of larger AI servers that can communicate more quickly across thousands of machines during training of large models. As part of the partnership, customers will gain access to AI Factories, dedicated AI infrastructure hosted inside their own data centers to support faster development.

Amazon separately introduced new servers built with its Trainium3 chip. The servers, available Tuesday, each carry 144 chips. AWS said they deliver more than four times the computing power of the previous generation while using 40% less energy. Dave Brown, AWS vice president of compute and machine learning services, said the company aims to compete on price but did not provide specific performance or power figures.

Brown said the goal is to show customers that AWS can offer the performance they need at competitive pricing.

AWS also unveiled updated versions of its AI models under the Nova brand. Amazon said Nova 2 offers faster responses and includes a version that handles text, images, speech and video prompts. Another model, Sonic, generates speech responses to spoken inputs. AWS CEO Matt Garman described the interaction as “human-like” in his keynote.

Amazon has faced difficulty gaining broad adoption of Nova in a market dominated by ChatGPT, Claude and Gemini. Still, AWS reported a 20% sales increase in its most recent quarter, driven by cloud and AI infrastructure demand.

At the conference, Amazon also launched Nova Forge, a service that lets companies build custom AI models using their own data.