# Cerebras 2024

Cerebras Systems, founded in 2015, develops wafer-scale AI semiconductors and supercomputers, including the WSE-3 and CS-3, and provides cloud AI services. In 2024, it filed for an IPO, aiming to expand its offerings amid growing demand for large-scale AI compute.

Cerebras Systems Inc. is a semiconductor and supercomputer company headquartered in Sunnyvale, California. It designs and manufactures some of the largest AI processors in the world, including the wafer-scale engine (WSE) series, and provides cloud computing services that allow customers to access its hardware remotely. In 2024, the company filed for an initial public offering (IPO) to raise capital for further expansion, marking a significant step in its growth trajectory.

The company's core products include the WSE-3 semiconductor, the CS-3 supercomputer, and APIs for AI inference and training clouds. These systems are designed to accelerate deep-learning and large-language-model workloads, offering performance that often exceeds traditional GPU-based clusters. Cerebras's technology is notable for using entire silicon wafers as a single chip, a technique known as wafer-scale integration, which reduces interconnect latency and power consumption compared to multi-chip alternatives.

## Founding and Early Development

Cerebras was founded in 2015 by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie, and Jean-Philippe Fricker, all of whom had previously worked together at SeaMicro, a server technology company acquired by AMD in 2012. The founders recognized that GPUs were not ideal for all AI tasks, prompting them to design a new architecture from scratch. Early challenges included developing unique cooling methods to manage the heat from massive chips, creating software to route around microscopic manufacturing defects, and inventing a machine to safely install the wafer onto a motherboard. The company spent roughly $200 million over several years to solve these packaging issues, burning through about $8 million per month, before achieving a working product in July 2019.

## Product Evolution: WSE-1 and WSE-2

In August 2019, Cerebras announced its first-generation Wafer-Scale Engine (WSE-1) and the CS-1 supercomputing system. The WSE-1 featured 400,000 processing cores, 1.2 trillion transistors, and 18 gigabytes of on-chip memory. The CS-1 was a 19-inch rack-mounted appliance with twelve 100-gigabit Ethernet connections, designed to plug into existing data centers. Early customers included pharmaceutical companies like GlaxoSmithKline and AstraZeneca, as well as national laboratories such as Lawrence Livermore and the Pittsburgh Supercomputing Center, which used the systems for drug discovery, genomic research, and simulation.

In April 2021, Cerebras released the CS-2, based on the WSE-2 chip, which doubled the core count to 850,000 and expanded on-chip SRAM to 40 gigabytes. Manufactured on TSMC's 7-nanometer process, the CS-2 was smaller (26 inches tall) and could support AI models with over 120 trillion parameters. Customers for the CS-2 included TotalEnergies, the National Center for Supercomputing Applications, and the Leibniz Supercomputing Centre. In June 2022, Cerebras set a record by training models with up to 20 billion parameters on a single CS-2 system, including versions of GPT-3 and GPT-NeoX.

## Scaling Up: Andromeda and Wafer-Scale Clusters

In September 2022, Cerebras introduced the concept of Wafer-Scale Clusters, which connect multiple CS-2 systems into a single computing cluster. A 16-system cluster could achieve 13.6 million cores, and up to 192 CS-2s could be linked for massive AI training workloads. In November 2022, the company unveiled Andromeda, a 16-chip cluster delivering up to 1 exaflop of AI computing power while consuming only 500 kilowatts, substantially less than comparable GPU-based supercomputers. This milestone highlighted Cerebras's efficiency advantages in energy and scalability.

## Recent Achievements and Partnerships

In 2023, Cerebras partnered with G42, an AI company based in the United Arab Emirates, which agreed to purchase up to nine supercomputers in a deal valued at around $100 million for the first unit. The company also expanded its customer base to include OpenAI and Amazon Web Services (AWS), with contracts signed in 2026, according to later announcements. By 2024, Cerebras had four major customers: Mohamed bin Zayed University of Artificial Intelligence (62% of 2025 revenues), G42 (24%), OpenAI, and AWS. The company also operates offices in San Diego, Toronto, and Bangalore, India, and uses TSMC as its sole semiconductor manufacturer due to the specialized requirements of wafer-scale chips.

## Technology and Competitive Landscape

Cerebras's chips are the largest AI semiconductors ever built, measuring 215 millimeters squared and covering an entire silicon wafer. They use static random-access memory (SRAM) instead of the dynamic RAM found in many GPUs, which contributes to their speed. The primary competitors for its hardware include Nvidia, AMD, Intel, and Broadcom, while its cloud computing services compete with Amazon Web Services, Microsoft Azure, Google Cloud, Oracle Cloud, and CoreWeave. Despite its advantages, Cerebras faces drawbacks such as a 25-kilowatt power draw and a cost of up to $3 million per node, limiting its appeal to a niche market of high-performance AI users.

## IPO Filing and Corporate Outlook

In 2024, Cerebras filed for an IPO to go public, aiming to raise funds for research and development, manufacturing capacity, and global expansion. The filing came amid growing demand for AI infrastructure, driven by the proliferation of [large language models](https://www.wikiprompt.org/wiki/large-language-model) and [generative AI](https://www.wikiprompt.org/wiki/generative-ai) applications. As of 2024, the company had not yet completed its IPO, and its financial details remained confidential. The move was seen as a strategic effort to capitalize on its position as a differentiated player in the AI hardware space, though it also faced scrutiny over its heavy reliance on a few key customers and the high costs associated with its technology.

## Applications and Ecosystem

Cerebras systems are used in a variety of scientific and commercial applications, including drug discovery, genomics, fluid dynamics, and nuclear stockpile stewardship. For example, the National Energy Technology Laboratory and Argonne National Laboratory have employed Cerebras hardware for COVID-19 research leveraged [deep learning](https://www.wikiprompt.org/wiki/deep-learning). The company also supports efforts in fields like computational chemistry and climate modeling, though its primary focus remains on [machine learning](https://www.wikiprompt.org/wiki/machine-learning) and [AI](https://www.wikiprompt.org/wiki/artificial-intelligence) training and inference. Through its cloud APIs, Cerebras allows developers to access its technology without purchasing hardware, broadening its reach beyond large enterprises and research institutions.

## Future Directions

Looking ahead, Cerebras aims to continue pushing the boundaries of wafer-scale computing, with plans for larger chips and more efficient cooling mechanisms. The company is also exploring new markets, such as edge computing and on-device AI, though these remain in early stages. With the AI industry rapidly evolving, Cerebras's success will depend on its ability to maintain technological leadership while managing costs and scaling production. The IPO, if successful, could provide the financial resources needed to achieve these goals, but the competitive landscape remains intense, with major cloud providers investing heavily in custom silicon like [AWS Trainium](https://www.wikiprompt.org/wiki/aws-trainium) and [Google Cloud](https://www.wikiprompt.org/wiki/google-cloud) TPUs.

Throughout its history, Cerebras has demonstrated a capacity for innovation, overcoming unprecedented engineering challenges to deliver products that redefine what is possible in AI hardware. As of 2024, the company continues to be a key player in the push toward more powerful and efficient AI systems, with its wafer-scale technology offering a distinct alternative to traditional GPU-based architectures.

---
Source: https://www.wikiprompt.org/wiki/cerebras-2024
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-13T03:55:39.373234+00:00