Cerebras Systems

What Cerebras Systems does

Cerebras Systems is a US-based AI semiconductor and infrastructure company founded in 2015 and headquartered in Sunnyvale, California. The company designs and manufactures the Wafer Scale Engine (WSE-3), the world's largest commercially available AI processor at 46,225 mm² with 4 trillion transistors, 900,000 AI-optimized cores, and 125 petaflops of AI compute, and packages it into the CS-3 system for customer deployment. Its core innovation is replacing distributed GPU clusters with a single device that uses a weight streaming architecture to enable linear scaling from 1 to 192 nodes with no code changes, supporting training of models from 1 billion to 24 trillion parameters on identical code.

Name

Cerebras Systems

Legal name

Cerebras Systems

Website

https://cerebras.ai

Company type

Public

Founded year

2015

Operating status

Ipo

Headcount range

1,001–5,000 employees

Short description

Cerebras Systems designs wafer-scale AI processors (WSE-3) and the CS-3 system, delivering cloud inference and training services to AI labs, hyperscalers (OpenAI, AWS, Meta), national labs, and enterprises in healthcare, pharma, and market intelligence. The company went public on NASDAQ in May 2026 under ticker CBRS.

Ownership category

akta.pro rank

Product category

AI Semiconductor Hardware

NAICS

Semiconductor and Other Electronic Component Manufacturing (3344), Electronic Computer Manufacturing (334111), Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (51821)

SIC

Electronic Computers (3571), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

Compute & Acceleration ICs (CPU/GPU/AI/FPGA/ASIC) (HDAHABAC)

akta.pro secondary industries

AI Accelerators (GPUs/TPUs/NPUs/ASICs) (HDAAAAAA), AI/Neural Processing Units (NPUs) & ML Accelerators (HDAHAJAF), AI Server Systems & HGX/Accelerator Platforms (HDAAAAAB), AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI)

Keywords

• AI accelerator chips
• Wafer-scale processors
• AI inference services
• AI training cloud
• AI supercomputing

Headquarters

HQ city

Sunnyvale

HQ country

United States

HQ region

North America

Offices3 records

Markets served

GTM type

B2B

Offering type

Hardware or Manufacturing

Cost components

Technology or R&D, Personnel, Infrastructure, Operations, Supply Chain, Marketing or Sales

Revenue model

1. AI Hardware Systems: Sale of CS-3 systems and WSE-3 processors to enterprises, research institutions, and governments. Hardware revenue represented approximately $111.6M in Q1 2026.
2. Cloud Inference Services: Pay-per-token and pay-per-hour inference API running on Cerebras Wafer Scale Engines. Cloud revenue surged 167% to $79.8M in Q1 2026, with expansion through AWS partnership.
3. Training Cloud: Pay per hour and pay per model training services. Allows training models of any size without distributed computing expertise.
4. Professional Services: AI strategy development, model design and build, organizational upskilling services provided by 50+ PhDs with ML expertise.

Pricing tiers

Go-to-market motion3 records

Distribution channels5 records

Marketing channels6 records

Core offering

Cerebras Systems designs and manufactures the world's largest AI processor (Wafer Scale Engine 3 / WSE-3) and the CS-3 system that houses it, delivering 125 petaflops of AI compute per chip. The company sells these hardware systems directly to enterprises and research institutions and operates the Cerebras Cloud platform offering pay-per-token AI inference and pay-per-hour AI training services via an OpenAI-compatible API.

Product overview

Cerebras Systems offers a comprehensive AI infrastructure portfolio anchored by its WSE-3 (Wafer Scale Engine 3) chip—the world's largest and fastest AI processor—and the CS-3 system that houses it. The company's product lineup spans two primary cloud services: Cerebras Inference for fast model serving (up to 15x faster than GPUs via OpenAI-compatible API) and Cerebras Training Cloud for model training and fine-tuning. Enterprise customers can access these capabilities through the Cerebras Cloud platform or via AWS integration combining Trainium and CS-3 chips. Cerebras also provides AI Model Services—consulting delivered by 50+ ML PhDs—to help organizations develop custom models. The ecosystem includes open-source models (Cerebras-GPT, BTLM-3B-8K, Jais) and healthcare-specific models (Med42, Mayo Clinic Genomic Foundation Model). Condor Galaxy supercomputers, built with partner G42, extend capacity to exaFLOP scale.

Differentiator

Problem solved

Functional benefit

Brands

• Condor Galaxy: AI supercomputer series built in partnership with G42, offering up to 36 exaFLOPS of AI compute
• Cerebras Cloud
• WSE-3 (Wafer Scale Engine 3)

Products and services

• WSE-3 (Wafer Scale Engine 3) The world's largest and fastest AI chip, measuring 46,225 mm² with 4 trillion transistors and 125 petaflops of AI compute through 900,000 AI-optimized cores, sold to enterprises, research institutions, and governments.
• CS-3 System The world's fastest and most scalable AI accelerator system powered by WSE-3, designed for training and inference of large AI models, sold to AI labs, enterprises, and research institutions.
• Cerebras Inference Cloud-based inference service providing AI processing up to 15x faster than NVIDIA GPUs, with OpenAI API-compatible endpoints and support for leading open-source models including Llama, Mistral, Qwen, GLM, and GPT-OSS.
• Cerebras Training Cloud Cloud platform for training and fine-tuning AI models from 1 billion to 24 trillion parameters with simplified scaling and no DevOps required, offered under pay-per-hour and pay-per-model pricing.
• Condor Galaxy Series of AI supercomputers built in partnership with G42, ranging from 4 to 36 exaFLOPS of AI compute, using multiple CS-2/CS-3 systems connected via Weight Streaming architecture; available through G42 Cloud in the UAE and Middle East region.
• Cerebras Cloud Cloud platform providing access to Cerebras hardware and models, including pay-per-hour and pay-per-model pricing tiers, with preconfigured environments and secure storage.
• AI Model Services Consulting services including AI strategy development, model design and build, and organizational upskilling, delivered by 50+ ML PhDs.

Quantifiable outcome

• Up to 15x faster inference than NVIDIA GPUs
• +5 more outcomes

Named customers12 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration8 records

AI capability13 records

Feature5 records

Partnerships

Eight partnerships are on record, tiered core, strategic and minor.

• Amazon Web Services (AWS)coreStrategic or Co-development Partner · 1 March 2026Multi-year partnership for disaggregated inference services combining AWS Trainium chips with Cerebras CS-3 systems. First hyperscaler cloud partnership with revenue expected to ramp in 2027.
• OpenAIcoreStrategic or Co-development Partner · 1 January 2026$20 billion multi-year agreement to deploy 750 megawatts of high-speed AI inference compute. OpenAI runs GPT-5.4 on Cerebras hardware with GPT-5.5 in porting process. Flagship customer anchor for Cerebras cloud infrastructure expansion.
• AlphaSensestrategicStrategic or Co-development Partner · 1 March 2025Strategic partnership integrating Cerebras WSE-3 and Llama AI models for 10x faster generative search. AlphaSense uses Cerebras inference for market intelligence platform.
• Mayo ClinicstrategicStrategic or Co-development Partner · 1 January 2025Collaboration on genomic foundation model achieving 96% accuracy in cancer prediction. Model trained on Cerebras Wafer Scale Cluster combining Mayo patient data with human reference genome.
• G42coreStrategic or Co-development Partner · 1 July 2023Condor Galaxy partnership building 9 AI supercomputers totaling 36 exaFLOPS. CG-1 (4 exaFLOPS) operational in Santa Clara. G42 Cloud distributes Cerebras capacity in UAE and Middle East.
• MetastrategicTechnology or IntegrationLlama 4 Scout model running on Cerebras achieving 2,000+ tokens/second. Meta partnership for fastest inference on open-weight models.
• Aleph AlphaminorStrategic or Co-development PartnerSovereign AI model development partnership using Cerebras wafer-scale technology for European AI independence.
• M42strategicStrategic or Co-development PartnerHealthcare AI joint venture between G42 and Mubadala. Collaborated on Med42 model achieving 72% on US Medical License Exam, fine-tuned Llama-70B in one week.

Scale indicators9 records

Recent moves7 records

Expansion highlights6 records

Direct peers

• SambaNova Systems: SambaNova builds custom AI accelerators (RDU) and offers DataScale systems for enterprise AI training and inference. Directly comparable to Cerebras as a fabless AI chip company offering alternative architectures to GPUs with a similar enterprise sales motion.
• Groq: Groq develops LPU (Language Processing Unit) inference accelerators focused on low-latency AI inference at high tokens-per-second. Directly comparable to Cerebras as a fast-inference AI chip company targeting similar developer and enterprise customers with cloud APIs.

Broad incumbents

• NVIDIA: NVIDIA is the dominant GPU provider for AI training and inference with the CUDA software moat. Directly comparable as a competing AI accelerator platform targeting the same AI labs, hyperscalers, and enterprises with H100/B200/GB200 systems versus Cerebras' WSE-3/CS-3.
• AMD: AMD produces MI300/MI325X AI accelerators competing in the same data center AI compute market as Cerebras. Comparable as a broader semiconductor incumbent with an AI accelerator portfolio, though focused on GPU-style architectures versus Cerebras' wafer-scale approach.
• Intel (Habana Labs): Intel's Habana Labs division produces Gaudi AI accelerators for training and inference. Comparable as a broad semiconductor incumbent offering AI accelerator alternatives, though packaged within Intel's broader portfolio rather than specialized like Cerebras.
• Microsoft (Azure Maia): Microsoft's Azure Maia 100 AI accelerator is designed for internal Azure AI workloads. Comparable as a broad hyperscaler-incumbent developing custom AI silicon for inference and training, representing both a potential customer and competitive threat to Cerebras.

Others

• TSMC: TSMC is Cerebras' foundry partner manufacturing the WSE-3 chip. Comparable as an ecosystem enabler/supplier rather than a competitor, but a critical dependency for Cerebras' wafer-scale production given the unique 46,225 mm² die requirements.

Emerging players

• Tenstorrent: Tenstorrent develops RISC-V-based AI accelerators for training and inference, led by Jim Keller. Comparable as an emerging fabless AI chip company pursuing alternative architectures to NVIDIA GPUs, though earlier-stage and smaller-scale than Cerebras.
• Etched: Etched develops Sohu, an ASIC specialized for transformer model inference. Comparable as an emerging AI chip company focused on accelerating inference workloads with custom silicon, similar to Cerebras' inference-first positioning.
• Positron: Positron is an AI inference-focused chip startup building memory-optimized accelerators. Comparable to Cerebras as a specialized inference chip company targeting similar fast-inference workloads with custom silicon.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• LinkedInlinkedin.com/company/cerebras-systems
• Xx.com/Cerebras

Revenue estimate

Valuation estimate

Number of profiles

Profiles12 records

Subsidiaries1 record

Funding overview

Funding rounds14 records

Investors31 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Cerebras Systems