# Cerebras Systems

Cerebras Systems is a Sunnyvale, California-based company that designs wafer-scale semiconductors and supercomputers for artificial intelligence deep-learning applications, known for producing the largest AI chips ever built.

Cerebras Systems Inc., headquartered in Sunnyvale, California, develops semiconductors, supercomputers, and related software to power artificial intelligence deep-learning applications such as inference engines. Its products include wafer-scale engine (WSE) semiconductors, CS-series supercomputers, and cloud APIs for AI training and inference, allowing users to access its computing power remotely. The company also builds data centers using its processors to provide direct cloud computing services.

Cerebras's WSE-3 semiconductors, measuring 215 mm (8.5 in) squared, are currently the largest AI semiconductors ever built, occupying entire silicon wafers. They use wafer-scale integration and switched fabric to reduce latency and interconnect bottlenecks compared to GPU clusters, and employ static random-access memory (SRAM) instead of dynamic random-access memory. While significantly more powerful than competitors' offerings, they face drawbacks including large size, a 25 kW power draw, and costs up to $3 million per node.

## History

Cerebras Systems was founded in 2015 by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie, and Jean-Philippe Fricker, who had previously worked together at SeaMicro, a company started by Feldman and Lauterbach in 2007 and sold to AMD in 2012 for $334 million. The founders recognized that GPUs were not optimal for high-level AI processes, but they had to overcome significant engineering challenges, including designing unique cooling methods to prevent a massive semiconductor from burning, developing software to route around microscopic manufacturing defects, and inventing a machine to drill 40 screws into the wafer simultaneously without cracking it.

The company struggled with integrated circuit packaging - adhering silicon to a motherboard, receiving power, and managing heating, cooling, and data delivery. Burning through $8 million per month and spending $200 million on the problem, Cerebras finally produced a working product in July 2019 after exhaustive trial and error.

In August 2019, Cerebras announced its first-generation WSE-1 semiconductors and the CS-1 supercomputing system, a 19-inch rack-mounted appliance with a single WSE primary processor containing 400,000 processing cores, 1.2 trillion transistors, twelve 100-gigabit Ethernet connections, and 18 gigabytes of memory. Early customers included educational institutions and life-sciences companies building supercomputers for drug discovery, computational fluid dynamics, genetic research, drug response prediction, and COVID-19 research, such as GlaxoSmithKline, AstraZeneca, the National Energy Technology Laboratory, Lawrence Livermore National Laboratory, the Pittsburgh Supercomputing Center, and Edinburgh Parallel Computing Centre.

In September 2020, Cerebras opened an office in Japan and partnered with Tokyo Electron. In April 2021, it released the CS-2 system based on the WSE-2, manufactured by TSMC's 7 nm process. The WSE-2 has 850,000 cores, 2.6 trillion transistors, 40 gigabytes of on-chip SRAM, 20 petabytes per second memory bandwidth, and 220 petabits per second fabric bandwidth, enabling support for AI models with more than 120 trillion parameters. Customers included TotalEnergies, nference, the National Center for Supercomputing Applications, and the Leibniz Supercomputing Centre.

In August 2021, Cerebras partnered with Peptilogics on AI for peptide therapeutics. In June 2022, it set a record for the largest AI models trained on a single device - a CS-2 system trained models up to 20 billion parameters, including GPT-3XL 1.3B, GPT-J 6B, GPT-3 13B, and GPT-NeoX 20B, with reduced software complexity. In August 2022, the Computer History Museum unveiled a display featuring the WSE-2, and Cerebras opened an office in Bangalore, India.

In September 2022, Cerebras announced Wafer-Scale Clusters, connecting up to 192 CS-2 systems, with a 16-system cluster providing 13.6 million cores for natural-language processing using data parallelism. In October 2022, Sandia National Laboratories began using the CS-2 for nuclear stockpile stewardship computing. In November 2022, Cerebras unveiled the Andromeda supercomputer, combining 16 WSE-2 chips into a cluster with 13.5 million AI-optimized cores, delivering up to 1 exaflop of AI computing while consuming 500 kW - drastically less than comparable GPU-accelerated systems. That month also saw partnerships with Cirrascale Cloud Services for a flat-rate "pay-per-model" compute service, and milestones at the National Energy Technology Laboratory. Argonne National Laboratory won the 2022 Gordon Bell Special Prize for COVID-19 research using the CS-2 alongside Nvidia and Hewlett-Packard products.

In July 2023, G42 agreed to pay around $100 million for the first of potentially nine supercomputers from Cerebras.

## Technology and Products

Cerebras's core technology is the wafer-scale engine, a single semiconductor that spans an entire silicon wafer, eliminating the need for multiple chips and interconnects. The WSE-3, announced in 2024, continues this approach with enhanced performance. The CS-3 supercomputer, introduced in 2024, houses the WSE-3 and is designed for both training and inference of large AI models. The company also offers cloud APIs for AI training and inference, allowing customers to access its hardware without purchasing it.

The use of SRAM instead of DRAM provides extremely high memory bandwidth and low latency, which is critical for deep-learning workloads. The switched fabric architecture enables efficient communication within the chip, reducing bottlenecks common in GPU clusters.

## Manufacturing and Customers

Cerebras's semiconductors are manufactured by TSMC, currently the only company capable of producing its chips. The company has offices in Sunnyvale, San Diego, Toronto, and Bangalore, India. Its major customers include the Mohamed bin Zayed University of Artificial Intelligence (62% of 2025 revenues), G42 (24% of 2025 revenues), OpenAI (signed in 2026), and Amazon Web Services (signed in 2026).

## Market Position and Competition

Cerebras's primary hardware competitors are Nvidia, AMD, Intel, and Broadcom, while its cloud computing services compete with Amazon Web Services, Microsoft Azure, Google Cloud Platform, Oracle Corporation, and CoreWeave. The company's wafer-scale approach offers advantages in performance and energy efficiency for certain AI workloads, but its high cost and power requirements limit its market to specialized applications.

## Applications and Research

Cerebras systems have been used in diverse fields, including drug discovery, genomics, computational fluid dynamics, and national security. The company's partnership with G42 has expanded its reach into the Middle East, and its collaboration with OpenAI and Amazon Web Services signals growing adoption in the AI industry. Cerebras has also been involved in large-scale AI research, such as training models with billions of parameters on a single system.

## Recent Developments

In 2024, Cerebras introduced the WSE-3 and CS-3, further pushing the boundaries of chip size and performance. The company has continued to expand its cloud offerings, enabling broader access to its technology. As of 2025, Cerebras remains a niche player in the AI hardware market, but its unique approach has attracted significant attention and investment.

## See Also

- [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence)
- [deep-learning](https://www.wikiprompt.org/wiki/deep-learning)
- [large-language-model](https://www.wikiprompt.org/wiki/large-language-model)
- [tsmc](https://www.wikiprompt.org/wiki/tsmc)

---
Source: https://www.wikiprompt.org/wiki/cerebras-systems
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-09T01:54:26.410772+00:00