Amazon Unveils Trainium3, A Budget-Friendly AI Chip Rivaling Nvidia

Amazon said its servers equipped with Trainium3 chips are four times faster and more energy efficient than those with its previous-generation chips.

" Trainium already represents a multibillion-dollar business today and continues to grow really rapidly," Amazon Web Services CEO Matt Garman said during the tech giant's annual event, called re:Invent, Tuesday.

The tech giant is among several Nvidia customers developing their own AI chips, posing one of the biggest competitive challenges to the leading chipmaker.

Read Also - Evertrust Raises €10M To Fuel Expansion In Europe’s Digital Trust Market

Google (GOOG) introduced its seventh-generation Ironwood TPU (tensor processing unit) in early November and is reportedly negotiating to supply Meta (META) with billions of dollars’ worth of TPUs, in addition to a recent multibillion-dollar deal with Anthropic (ANTH.PVT). Microsoft (MSFT) also plans to rely on custom in-house chips instead of Nvidia’s, though its development timeline has experienced delays.

“Diversity of chips in the AI market is a good thing,” said Dave Brown, AWS vice president of compute and machine learning, in an interview with Yahoo Finance.

Amazon has been rapidly expanding its custom AI hardware over the past year, recently completing a major AI data center initiative called Project Rainier. AI developer and OpenAI competitor Anthropic is expected to use 1 million of Amazon’s custom chips from this project and other data centers by the end of 2025. According to Dave Brown, Anthropic has collaborated closely with Amazon in developing its AI chips.

Amazon’s chips are significantly more cost-effective than Nvidia’s, offering AI developers savings of 30% to 40%, Brown noted.

”That's what our customers are looking for, is to constantly get more compute and more performance, and then super importantly, at a lower price,” he said.

Nvidia CEO Jensen Huang has argued that customers would choose the company’s GPUs even if alternatives were free, citing their superior performance and robust software ecosystem.

Amazon is a major Nvidia customer, spending over 10% of its capital expenditures on Nvidia products, which generate 7.5% of the chipmaker’s revenue, according to Bloomberg. OpenAI recently signed a $38 billion agreement to access Nvidia GPUs via Amazon’s cloud platform.

When asked about OpenAI transitioning to use Amazon’s chips, Brown left open the possibility, speaking broadly of AWS customers: “They may say, ‘Hey, we're going to stay on Nvidia.’ But if they can see a meaningful price performance benefit … then normally we see that they'll definitely work on moving [to Amazon’s chips].”

Amazon also said Tuesday it's working on developing its next-generation Trainium4 AI chips that, notably, will be designed for compatibility with Nvidia's networking technology, NVLink Fusion. The coveted tech connects chips within AI server racks.