Cerebras Systems is an artificial intelligence (AI) company founded in 2015 with the mission to revolutionize AI compute. Cerebras Systems aims to accelerate generative AI by building a new class of AI supercomputers from the ground up. Its core goal is to develop and provide technology that pushes the boundaries of what is possible in AI, enabling advancements across various industries. Cerebras Systems designs and manufactures large-scale AI-specific processors, known as Wafer-Scale Engines (WSE), and the computer systems that house them, like the CS-3 system.
Cerebras Systems is recognized for producing some of the largest and fastest AI processors commercially available. The company has established a reputation for innovation in the AI hardware sector, specifically for AI training and inference. Its solutions are utilized by corporations, research institutions, and governments for developing proprietary models and training open-source models. Cerebras Systems is considered a challenger to established players in the AI chip market, offering powerful alternatives for demanding AI workloads.
Offerings, Capabilities, and Integrations
Cerebras Systems specializes in developing and deploying artificial intelligence (AI) hardware and software solutions engineered for large-scale AI workloads, particularly in large language models (LLMs) and scientific computing. Its core offering revolves around its Wafer-Scale Engine (WSE) technology, the largest commercially available AI processor. This technology provides Cerebras Systems with a competitive edge by enabling significantly faster processing speeds and computational power for complex AI tasks compared to traditional chip designs. Cerebras Systems offers its solutions via the Cerebras Cloud and on-premises AI supercomputer setups, enhancing accessibility to its AI capabilities. The company’s ability to deliver massive acceleration for training and inference, simplify the development workflow for large models, and reduce programming complexity contributes to its reputation as a pioneer in accelerating generative AI.
Cerebras Systems integrates its technology with popular machine learning frameworks like TensorFlow and PyTorch. It has also partnered with companies like Hugging Face to make its inference capabilities accessible through their platforms, and with ZS to integrate its CS-3 systems into ZS’s MAX.AI platform for fine-tuning, training, and deploying LLMs and agentic AI solutions. These integrations allow developers and enterprises to leverage Cerebras Systems’ high-performance computing for their AI models and applications.
Products and Services
- Wafer-Scale Engine (WSE) Processors: These are the largest chips ever built for AI. The latest generation, the WSE-3, features 900,000 AI-optimized cores and 4 trillion transistors, designed to accelerate AI training and inference. The WSE technology is central to all Cerebras Systems’ hardware products.
- CS Systems (CS-1, CS-2, CS-3): These are AI supercomputers powered by the WSE processors. The flagship CS-3 system, powered by the WSE-3, is designed for demanding generative AI workloads, offering significant speed and efficiency for training and deploying large models. These systems can be clustered to create even larger supercomputers.
- Cerebras Software Platform (CSoft): This is a comprehensive software stack that enables users to deploy AI models on WSE hardware efficiently. It integrates with familiar ML frameworks and includes tools for debugging and profiling.
- Cerebras Inference: This service provides high-speed AI inference, reportedly significantly faster than GPU-based solutions. It is available as a cloud service via an API and is designed for easy deployment of large AI models. Cerebras recently launched six new AI inference datacenters to expand this service.
- Cerebras Cloud: This service allows customers to access WSE-powered systems without needing on-premises hardware, offering flexibility for high-performance AI training and inference.
- AI Model Services: Cerebras Systems offers services to help organizations train and deploy state-of-the-art AI models. This includes assistance with data preparation, model design, training oversight, and optimization.
- Condor Galaxy: In partnership with G42, Cerebras Systems has developed Condor Galaxy, a network of interconnected supercomputers powered by Cerebras systems, offered as a cloud service. CG-1, the first supercomputer in this network, links 64 CS-2 systems.
Target Customers
Cerebras Systems targets organizations with substantial AI computing needs, including those training large models or working with vast datasets, often spending millions on AI compute annually. Its customers span various industries and market segments:
- Research Institutions and National Laboratories: Organizations like Argonne National Laboratory, Lawrence Livermore National Laboratory, and the National Energy Technology Laboratory (NETL) use Cerebras systems for scientific computing and research.
- Pharmaceuticals and Healthcare: Companies such as GlaxoSmithKline and Mayo Clinic leverage Cerebras technology for drug discovery, genomics research, and enhancing medical diagnostics.
- Energy: The energy sector, including companies like TotalEnergies, utilizes Cerebras systems for complex modeling and simulations.
- Government: Government entities use Cerebras technology for sovereign clouds and to meet domestic AI requirements.
- Financial Services: The finance industry employs Cerebras solutions for complex modeling and risk assessment.
- Tech Companies and AI Developers: Companies like Perplexity, Mistral AI, AlphaSense, and developers on platforms like Hugging Face use Cerebras for high-speed inference and model development.
- Enterprises: Cerebras focuses on enterprises needing extremely fast inference for applications like agent-based AI.
These target customers benefit from Cerebras Systems’ products and services through significantly reduced AI model training times, faster inference speeds, the ability to train larger and more complex models, and simplified deployment of AI workloads. The technology enables them to tackle problems that were previously computationally prohibitive, accelerating innovation and time-to-solution in their respective fields.
Cloud Integrations and Marketplaces
Cerebras Systems offers its AI supercomputing capabilities through the Cerebras Cloud. This allows customers to access the performance of Cerebras’s AI supercomputers, such as the Condor Galaxy network, without needing to manage the physical systems themselves. The Condor Galaxy is a network of interconnected supercomputers built in partnership with G42. Cerebras and G42 offer Condor Galaxy 1 (CG-1) as a cloud service. Flexible deployment options for Cerebras solutions include serverless API, private cloud, or on-premises.
Cerebras Systems has also been expanding its cloud presence through partnerships and new data centers:
- G42 Partnership and Condor Galaxy: Cerebras, in partnership with G42, has been building out the Condor Galaxy network of AI supercomputers. CG-1, the first supercomputer in this network, is offered as a cloud service, allowing customers to train large AI models. The partnership plans to deliver tens of exaFLOPs of AI compute. Some of these supercomputers are located in the U.S. G42 is also migrating its data platform and other technology infrastructure to Microsoft Azure as part of a strategic investment by Microsoft.
- Cerebras Cloud: Cerebras solutions are available through the Cerebras Cloud, providing access to its AI systems. This includes offerings for both AI model training and high-speed AI inference.
- Data Center Expansion: Cerebras has announced the launch of new AI inference data centers across North America and Europe to expand its high-speed inference cloud capacity. Some of these data centers are operated by Cerebras, while others are jointly operated with G42.
- Hugging Face Partnership: Cerebras has partnered with Hugging Face to integrate its inference capabilities into the Hugging Face Hub. This allows developers on the Hugging Face platform to select Cerebras as their inference provider for various AI models.
- Microsoft Azure: While Cerebras itself is not directly listed as a provider on the Microsoft Azure marketplace in the same way some software vendors are, its partner G42, with whom Cerebras built the Condor Galaxy supercomputers, is migrating its data platform and infrastructure to Microsoft Azure. Additionally, some models trained on Condor Galaxy, like Jais 13B and Jais30B, are available on Azure Cloud. Microsoft has also invested in G42.
- Cirrascale Cloud Services: In November 2022, Cerebras announced a partnership with Cirrascale Cloud Services to offer a flat-rate “pay-per-model” compute time for its Cerebras AI Model Studio.
A search of the Google Cloud Marketplace for “Cerebras Systems” did not yield any direct listings for Cerebras Systems’ products or services.
Cerebras’s strategy involves providing access to its specialized AI hardware through its own cloud services and partnerships, rather than primarily through listings on major public cloud marketplaces like AWS or Google Cloud directly under its own name for purchasing its hardware or base software. However, its systems are used by entities that may leverage these cloud platforms in other ways, and models trained on Cerebras hardware are becoming accessible via cloud platforms through partners.
Key People
- Chief Executive Officer and Co-Founder: Andrew Feldman
- Chief Technology Officer, Emeritus and Co-Founder: Gary Lauterbach
- Chief System Architect and Co-Founder: Jean-Philippe Fricker
- Chief Architect, Advanced Technologies and Co-Founder: Michael James
- CTO and Co-Founder: Sean Lie
- Chief Operating Officer: Dhiraj Mallick
- SVP and Chief Financial Officer: Bob Komin
- SVP, Product and Strategy: Andy Hock
- SVP, Software Engineering: Vinay Srinivas
- Chief Information Security Officer: Naor Penso
- Executive Vice President of Worldwide Sales: Alex Varel
- Senior Vice President of AI Cloud and Inference: Hagay Lupesko
Key Facts
- Headquarters Location: Sunnyvale, California, United States.
- Number of Employees: Approximately 400-673.
- Annual Revenue: $78.7 million for the year ended December 31, 2023. Revenue for the twelve months ending June 30, 2024, was $206.48 million.
- Parent Company: None.
- Subsidiary Companies: Cerebras Systems India Private Limited, Cerebras Wafer Scale ULC (Canada), Cerebras Systems G.K. (Japan).
- Publicly Listed: No, Cerebras Systems is a privately held company. It filed for an initial public offering (IPO) in September 2024.
Analyst Recognition
Cerebras Systems has been recognized by Forrester and IDC in the AI sector.
- Forrester: In “The Forrester Wave™: AI Infrastructure Solutions, Q1 2024,” Cerebras Systems was included as one of 12 vendors evaluated in the AI infrastructure solutions space. Forrester defines AI infrastructure solutions as computer systems and cloud services designed to maximize the performance of AI workloads, including data preparation, model training, and inferencing.
- IDC: Cerebras Systems was listed in the “IDC Market Glance: Generative AI Technologies and Services, 3Q23” and the “IDC Market Glance: Generative AI Technologies and Services in India, 4Q23”. These reports offer a high-level view of the emerging generative AI technology and services landscape, highlighting key technologies in infrastructure, foundation models, GenAI platforms, GenAI applications, and GenAI services.
There is no specific mention in the provided search results of Cerebras Systems being included in technology categories by Gartner or Everest Group for its own products or services. One search result mentions a Gartner analyst, Chirag Dekate, commenting on Cerebras Systems’ position to address opportunities in NLP, speech recognition, and other domains, and another mentions a Gartner analyst, Arun Chandrasekaran, commenting on the shift in the generative AI market towards the cost and speed of inferencing, which Cerebras is addressing. Another result refers to “Cerebra” by Accenture (Flutura) in Gartner Peer Insights, which appears to be a different entity. Similarly, some search results mention Everest Group in the context of other companies or general market reports, not specifically recognizing Cerebras Systems.
