# Cerebras

Cerebras Systems Inc. is a Sunnyvale, California-based company that develops wafer-scale AI semiconductors, supercomputers, and cloud services for 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 the wafer scale engine (WSE)-3 semiconductors, the CS-3 supercomputers, and the "AI inference cloud" and "AI training cloud" APIs, which allow users to access the company's computing power without buying its hardware. The company also builds data centers using its processors and supercomputers to provide cloud computing services directly to clients.

Measuring 215 mm (8.5 in) squared, the company's WSE-3 semiconductors are currently the largest AI semiconductors ever built. They take up entire silicon wafers and use wafer-scale integration and switched fabric. This reduces latency and interconnect bottlenecks compared to GPU clusters. They use static random-access memory, as opposed to dynamic random-access memory. Cerebras semiconductors and computer systems are much more powerful than those of competitors; however, they have disadvantages due to their large size, 25kW power draw, and cost of as much as $3 million per node.

## Founding and Early Challenges

Cerebras Systems was founded in 2015 by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie, and Jean-Philippe Fricker. These five founders worked together at SeaMicro, which was started in 2007 by Feldman and Lauterbach and sold to AMD in 2012 for $334 million. The founders knew that GPUs were not the optimal semiconductors for high-level processes. However, they had to design unique cooling methods to prevent a "massive" semiconductor from burning when drawing power, unique software to route around usual microscopic manufacturing defects, and they had to invent a machine that could drill 40 screws into the wafer simultaneously without it cracking.

The company had difficulty solving the problem of integrated circuit packaging: adhering the silicon to a motherboard, receiving power, and dealing with heating and cooling and the pipes to deliver and return data. It was burning through $8 million per month and spent $200 million trying to solve the problem. In July 2019, after exhaustive trial and error, the company finally produced a product that worked.

## First-Generation Products

In August 2019, Cerebras announced WSE-1, its first-generation Wafer-Scale Engine (WSE) semiconductors and its CS-1 supercomputing system. The CS-1 is a 19-inch rack-mounted appliance and includes a single WSE primary processor with 400,000 processing cores, 1.2 trillion transistors (twelve 100-gigabit ethernet connections), and 18 gigabytes of memory.

The company's first customers were educational institutions and life-sciences companies that were building supercomputers for purposes of drug discovery, computational fluid dynamics, genetic and genomic research, to predict response to drugs, and for COVID-19 research. Early customers included GlaxoSmithKline, AstraZeneca, the National Energy Technology Laboratory, Lawrence Livermore National Laboratory, the Pittsburgh Supercomputing Center, and Edinburgh Parallel Computing Centre.

## Expansion and CS-2

In September 2020, the company opened an office in Japan and partnered with Tokyo Electron. In April 2021, the company released its CS-2 system, based on the company's Wafer Scale Engine Two (WSE-2), which has 850,000 cores. The CS-2 is manufactured by the 7 nm process of TSMC. It is 26 inches (660 mm) tall and fits in one-third of a standard data center rack. The WSE-2 has 850,000 cores and 2.6 trillion transistors. It enables a single system to support AI models with more than 120 trillion parameters. The WSE-2 expanded on-chip SRAM to 40 gigabytes, memory bandwidth to 20 petabytes per second, and total fabric bandwidth to 220 petabits per second. Customers included TotalEnergies, nference, the National Center for Supercomputing Applications (NCSA), and the Leibniz Supercomputing Centre.

In August 2021, Cerebras announced a partnership with Peptilogics on the development of AI for peptide therapeutics. In June 2022, Cerebras set a record for the largest AI models ever trained on one device - a single CS-2 system with one Cerebras wafer trained models with up to 20 billion parameters. The Cerebras CS-2 system can train multibillion-parameter natural-language-processing (NLP) models including GPT-3XL 1.3 billion models, as well as GPT-J 6B, GPT-3 13B, and GPT-NeoX 20B with reduced software complexity and infrastructure.

## Recognition and Clusters

In August 2022, the Computer History Museum in Mountain View, California unveiled a new display featuring the WSE-2, named "The Biggest Chip In the World". Also in August 2022, Cerebras opened an office in Bangalore, India. In September 2022, Cerebras announced that it can patch its chips together to create what would be the largest-ever computing cluster for AI computing. A Wafer-Scale Cluster can connect up to 192 CS-2 AI systems into a cluster, while a cluster of 16 CS-2 AI systems can create a computing system with 13.6 million cores for natural-language processing. It uses data parallelism to train.

In October 2022, Sandia National Laboratories of the National Nuclear Security Administration began using the CS-2 in nuclear stockpile stewardship computing, to determine if nuclear weapons will work as intended. In November 2022, Cerebras unveiled the Andromeda supercomputer, which combines 16 WSE-2 chips into one cluster with 13.5 million AI-optimized cores, delivering up to 1 exaflop of AI computing horsepower, or at least one quintillion (1018) operations per second. The entire system consumes 500 kW, which was a drastically lower amount than somewhat-comparable GPU-accelerated supercomputers.

## Partnerships and Recent Developments

In November 2022, Cerebras announced a partnership with Cirrascale Cloud Services to provide a flat-rate "pay-per-model" compute time for its Cerebras AI Model Studio. In November 2022, the National Energy Technology Laboratory (NETL) set milestones using Cerebras products. In November 2022, Argonne National Laboratory won the 2022 Gordon Bell Special Prize for COVID-19 research by using the CS-2 as well as products from Nvidia and Hewlett-Packard to transform large language models to analyze and predict variants of SARS-CoV-2.

In July 2023, G42 agreed to pay around $100 million to purchase the first of potentially nine supercomputers from Cerebras. The first system was delivered in 2024. The company has 4 major customers: 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).

## Manufacturing and Market Position

Cerebras has offices in Sunnyvale, San Diego, Toronto, and Bangalore, India. Its semiconductors are manufactured by TSMC, currently the only company that has the ability to manufacture Cerebras chips. The company's primary competitors for its hardware are Nvidia, AMD, Intel, and Broadcom and the company's primary competitors for its cloud computing services are Amazon Web Services, Microsoft Azure, Google Cloud Platform, Oracle Corporation, and CoreWeave.

Cerebras's approach to [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) hardware diverges from the GPU-centric model used by many competitors. By using wafer-scale integration, the company aims to reduce the communication overhead that plagues multi-chip systems, which is particularly relevant for training [large-language-model](https://www.wikiprompt.org/wiki/large-language-model)s and other [deep-learning](https://www.wikiprompt.org/wiki/deep-learning) workloads. The company's focus on [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) infrastructure has positioned it as a niche player in the broader [generative-ai](https://www.wikiprompt.org/wiki/generative-ai) ecosystem, with applications ranging from [neural-network](https://www.wikiprompt.org/wiki/neural-network) research to commercial cloud offerings.

Despite its technical advantages, the high cost and power requirements of Cerebras systems limit their adoption to specialized use cases. The company's reliance on [tsmc](https://www.wikiprompt.org/wiki/tsmc) for manufacturing and its relatively small customer base compared to giants like [nvidia](https://www.wikiprompt.org/wiki/nvidia) and [amd](https://www.wikiprompt.org/wiki/amd) highlight the challenges of competing in the semiconductor industry. However, its partnerships with entities like [openai](https://www.wikiprompt.org/wiki/openai) and [amazon-web-services](https://www.wikiprompt.org/wiki/amazon-web-services) signal growing interest in alternative architectures for AI compute.

---
Source: https://www.wikiprompt.org/wiki/cerebras
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-07T02:31:22.095854+00:00