Amazon Web Services, which has been building its own AI training chips for years, recently introduced a new version called Trainium 3 that comes with some impressive specifications.
The cloud provider, which made the announcement at AWS re:Invent 2025 on Tuesday, also teased the next product on its AI training product roadmap: Trainium 4, which is already in the works and will be able to work with Nvidia’s chips.
AWS used its annual technology conference to formally launch the Tranium3 UltraServer, a system powered by the company’s cutting-edge, 3 nanometer Tranium3 chip as well as its in-house networking technology. As you might expect, according to AWS, the third-generation chip and system offer a big jump in AI training and performance compared to the second-generation chip.
AWS says the system is up to 4 times faster, with 4 times more memory, not only for training, but also for delivering AI apps at peak demand. Additionally, thousands of UltraServers can be linked together to provide an app with up to 1 million Trenium 3 chips – 10 times more than the previous generation. According to the company, each Ultraserver can host 144 chips.
Perhaps more importantly, AWS says the chips and systems are 40% more energy efficient than the previous generation. While the world races to build massive data centers powered by astronomical gigawatts of electricity, data center giant AWS is trying to build systems that consume less, not more.
Obviously, it is in AWS’s direct interest to do so. But in its classic, Amazon cost-conscious way, it promises that these systems also save its AI Cloud customers money.
Amazon said AWS customers such as Anthropic (in which Amazon is also an investor), Japan’s LLM Karakuri, SplashMusic and Descartes are already using third-generation chips and systems and have seen their estimated costs cut significantly.
techcrunch event
san francisco
,
October 13-15, 2026
AWS also presented a roadmap for the next chip, Tranium4, which is already in development. AWS promised that the chip would take another big step up in performance and support Nvidia’s NVLink Fusion high-speed chip interconnect technology.
This means that AWS Trenium4-powered systems will be able to interoperate and increase their performance with Nvidia GPUs while using Amazon’s homegrown, low-cost server rack technology.
It’s also worth noting that Nvidia’s CUDA (Compute Unified Device Architecture) has become the de facto standard that all major AI apps are built to support. The Tranium4-powered system could make it easier to lure larger AI apps built with Nvidia GPUs to Amazon’s cloud.
Amazon did not announce any timeline for Tranium4. If the company sticks to previous rollout timelines, we’ll hear more about Tranium4 at next year’s conference.
Follow all of TechCrunch’s coverage of the annual enterprise tech event here.
<a href