AWS did not bother offering a launch date for the new silicon but said the Nvidia tech will knit mixed chips together at high speed, which is vital when you are stitching thousands of servers into one giant model-chewing cluster.
The reveal came during AWS’s Las Vegas shindig, where around 60,000 punters gather to hear about cloud wizardry and watch executives try to look relaxed on stage. Nvidia has been hustling hard to get other firms to adopt NVLink and now has Troubled Chipzilla, Qualcomm and AWS tagging along.
The linkup should enable AWS to build larger AI servers that can recognise and chat with each other faster. As part of the deal, customers will be offered something AWS calls AI factories, which are exclusive AI chunks inside their data centres built for speed and readiness.
Nvidia chief executive Jensen Huang said, “Together, Nvidia and AWS are creating the compute fabric for the AI industrial revolution, bringing advanced AI to every company, in every country, and accelerating the world's path to intelligence.”
AWS compute and machine learning boss Dave Brown announced that Amazon pushed out new servers based on a Trainium3 chip. They pack 144 chips each and offer more than four times the grunt of the previous generation while sipping 40 per cent less power.
Brown did not offer absolute numbers on power or performance. He said AWS aims to compete with rivals, including Nvidia, on price by offering strong performance at a lower bill.
He said: “We've got to prove to them that we have a product that gives them the performance that they need and get the right price point so they get that price-performance benefit.”
Amazon trotted out fresh versions of its Nova AI models. Nova 2 is faster and more responsive, and ships with a variant that handles text-to-text, image, speech, or video prompts. Another version called Sonic replies to speech prompts with speech outputs, which AWS chief executive Matt Garman claimed were “human-like.”
Amazon has struggled to get Nova accepted as being in the same league as ChatGPT, Claude, and Gemini. Even so, AWS saw a 20 per cent sales increase in its last quarter, driven by demand for cloud and AI infrastructure.
At the Las Vegas show, Amazon announced Nova Forge, a service that lets companies craft their own AI models using their own data.
Garman said: “This allows you to produce a model that deeply understands your information, all without forgetting the core information that the thing has been trained on."