Amazon Web Services has introduced a new version of its AI training chip, known as Trainium3, with impressive specifications. The cloud provider made the announcement Tuesday at its AWS re:Invent 2025 conference. AWS also teased the next product on its roadmap, Trainium4, which is already in development and will be designed to work alongside Nvidia’s chips.
At its annual tech conference, AWS formally launched the Trainium3 UltraServer, a system powered by the company’s state-of-the-art, 3-nanometer Trainium3 chip and its homegrown networking technology. According to AWS, this third-generation chip and system offer significant performance improvements for both AI training and inference compared to the second-generation chip.
AWS states the new system is more than four times faster and has four times more memory, not only for training but also for delivering AI applications during peak demand. Furthermore, thousands of these UltraServers can be linked together to provide an application with access to up to 1 million Trainium3 chips, which is ten times the capacity of the previous generation. Each UltraServer can host 144 chips.
Perhaps more importantly, AWS claims the chips and systems are also 40% more energy efficient than the previous generation. As the world races to build larger data centers with enormous power demands, AWS is focusing on creating systems that consume less electricity. This is obviously in AWS’s direct interest, but in Amazon’s cost-conscious manner, it also promises these systems will save its AI cloud customers money.
AWS customers, including Anthropic, Japan’s LLM Karakuri, SplashMusic, and Decart, have already been using the third-generation chip and system and have significantly reduced their inference costs, according to Amazon.
AWS also presented a roadmap for the next chip, Trainium4, which is already in development. The company promised the chip will provide another major step up in performance and will support Nvidia’s NVLink Fusion high-speed chip interconnect technology. This means AWS Trainium4-powered systems will be able to interoperate and extend their performance with Nvidia GPUs while still utilizing Amazon’s homegrown, lower-cost server rack technology.
It is worth noting that Nvidia’s CUDA has become the de facto standard that major AI applications are built to support. The Trainium4-powered systems may make it easier to attract large AI applications, originally built for Nvidia GPUs, to Amazon’s cloud.
Amazon did not announce a timeline for Trainium4. If the company follows previous rollout schedules, more details will likely be shared at next year’s conference.