Broadcom to lead back-end design for Amazon Trainium 3.5 AI chips using 3nm process
News

Broadcom to lead back-end design for Amazon Trainium 3.5 AI chips using 3nm process

Friday, March 20, 2026 at 07:38 AM

Broadcom has reportedly secured a role in Amazon Web Services' (AWS) Trainium 3.5 chip project using the 3nm process node. While AWS subsidiary Annapurna handles front-end design, Broadcom will manage back-end design and integration services, specifically to test their ethernet switch solutions. Production for Trainium 3.5 is projected to reach 750,000 units with a rollout between late 2027 and early 2028. This shifts the supplier landscape for AWS, which also utilizes Alchip for Trainium 3 and Marvell for Trainium 3 Lite.

Context

AWS has reportedly selected Broadcom to lead the back-end design and integration for its upcoming Trainium 3.5 AI accelerator. This project marks a significant expansion for Broadcom within the Amazon ecosystem, utilizing a cutting-edge 3nm process. While Amazon’s internal Annapurna Labs handles the front-end design, Broadcom’s involvement provides AWS a strategic platform to test high-performance ethernet switch solutions alongside its custom silicon. This shift follows a competitive landscape where Alchip and Marvell have previously secured major design wins for the Trainium 3 and 3 Lite iterations. The deployment of Trainium 3.5 is projected for late 2027 to early 2028, with an estimated total shipment volume of 750,000 units. This aligns with AWS's aggressive strategy to reduce reliance on Nvidia by scaling its internal ASIC production, which is expected to represent a significant portion of the global AI chip market by 2026. For investors, Broadcom’s capture of this high-volume socket reinforces its dominance in the custom AI silicon market, where it already co-develops flagship accelerators for Google and Meta.

Related Companies

Broadcom
Broadcom
AVGO
US
Marvell
Marvell
MRVL
US
Alchip
Alchip
3661
TW
Amazon
Amazon
AMZN
US