Skip to main content

Gemma 4 31B runs at over 1,800 tokens per second on Cerebras Inference. This is the world’s fastest multimodal model. >>

Scaling American manufacturing
of AI

Flex and Cerebras are expanding their partnership to build the CS-3 in Milpitas, California — supporting an anticipated 7× increase in production as demand for AI infrastructure accelerates.

Video thumbnail
Loading player...

Building AI infrastructure in Silicon Valley

The CS-3 is one of the world’s most advanced AI systems — designed, assembled, integrated, and tested in the heart of Silicon Valley.

7x

~7x

Anticipated increase in CS-3 production
capacity in 2026

Milpitas, CA

Built in the heart of Silicon Valley

58x

WSE-3 is 58x larger than the leading GPU

Up to 15x

faster Inference than GPU

What the expansion delivers

Multiple parallel lines bring more CS-3 systems through integration and testing at once.

The CS-3 is unlike any computer system ever built, and scaling its production requires an extraordinary manufacturing partner. Flex brings the technical depth, operational rigor, and manufacturing expertise needed to support that scale.

Dhiraj Mallick

Chief Operating Officer, Cerebras

With the ideas and vision that Cerebras brings, and the manufacturing expertise, power, and integration skills that Flex brings, this partnership will last well into the future.

Rob Campbell

President, Communications, Enterprise and Cloud Business

Unlike any super computer ever built

Built on Cerebras' wafer-scale engine architecture, the CS-3 uses a processor physically larger than any conventional AI chip — integrated into a platform designed for large-scale AI training and inference.

Challenges rarely seen in server production

Flex engineers worked closely with Cerebras to develop dedicated assembly flows, automated test stations, and new manufacturing methodologies tailored to wafer-scale computing.

  • Specialized handling processes for wafer-scale hardware
  • Custom tooling and precision calibration
  • Dedicated assembly flows and automated test stations
  • Extensive system-level validation and qualification

Every stage of the manufacturing process — from mechanical integration to thermal validation and final system qualification — required deep collaboration between our engineering teams.

Rob Campbell

President, Communication, Enterprise & Cloud, Flex

Build the fastest & smartest apps

Get started in <30 seconds

Performance comparisons are based on third-party benchmarking or internal testing. Observed inference speed improvements versus GPU-based systems may vary depending on workload, configuration, date and models being tested.

1237 E. Arques Ave
 Sunnyvale, CA 94085

© 2026 Cerebras.
All rights reserved.