Cerebras

Cerebras designs and manufactures wafer-scale silicon chips and cloud inference services, positioning itself as an alternative to GPU-based AI infrastructure.

Updated 8 days ago Reviewed by 7wData

Visit Website

Profile

AI inference hardware and services using wafer-scale silicon chips, sold to cloud providers and enterprises seeking ultra-low latency for real-time AI applications.

Cerebras designs and manufactures wafer-scale silicon chips and cloud inference services, positioning itself as an alternative to GPU-based AI infrastructure. Founded in March 2016 by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie, and Jean-Philippe Fricker—all former SeaMicro engineers—the company spent nearly a decade in stealth developing its Wafer-Scale Engine (CS-3). The CS-3 is a single processor carved from a 300mm silicon wafer, packing 4 trillion transistors and 900,000 cores onto one contiguous die with 44GB of on-chip memory and ultra-high memory bandwidth.

Cerebras claims the chip delivers 2,000+ inference tokens per second, roughly 8 times faster than Nvidia's H200 GPU for single-user latency workloads.\n\nAfter launching its inference-as-a-service platform in August 2024, Cerebras's business accelerated sharply. 2025 revenue jumped 76% to $510 million, and the company swung to an $88 million profit from a $481.6 million loss in 2024. In January 2026, OpenAI signed a landmark $20 billion multiyear deal for 750 megawatts of Cerebras capacity through 2028—a transformational customer commitment that reduced prior concentration risk. AWS signed a multiyear binding term sheet in March 2026 to distribute Cerebras inference globally.\n\nCerebras went public on May 14, 2026, raising $5.55 billion at a $95 billion post-IPO valuation—the largest U.S. tech IPO since Uber in 2019.

Stock opened 68% above the $185 IPO price but fell roughly 10% the following week, signaling investor caution around valuation and execution.\n\nA critical structural risk persists: 86% of 2025 revenue came from two interlinked UAE entities—Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) at 62% and G42, an Abu Dhabi AI conglomerate, at 24%. This concentration nearly derailed the company's initial IPO filing in late 2024, when the Committee on Foreign Investment in the United States (CFIUS) reviewed G42's historical ties to Huawei. Though CFIUS cleared the path in March 2025, customer concentration has not materially improved: MBZUAI alone held 77.9% of accounts receivable as of year-end 2025. The company expects Q1 2026 revenue growth to decline due to warrant accounting adjustments, not demand weakness.

Who buys this

Cloud AI infrastructure providers scaling inference capacity (OpenAI, AWS)
Large AI model makers and research institutions (Meta, IBM, Mistral, MBZUAI)
Enterprise AI applications in finance, search, and coding (AlphaSense, Notion, Cognition)
Life sciences and genomics research using AI (GSK, Mayo Clinic)
Government and sovereign AI initiatives in the Middle East (G42, MBZUAI)

Publicly disclosed clients

OpenAI
Amazon Web Services
Meta
IBM
Mistral
Cognition
AlphaSense
Notion

Strengths and what to watch

Strengths

Wafer-scale architecture delivers ultra-high memory bandwidth and single-user latency far exceeding GPU clusters, critical for real-time inference and agentic AI workloads
Landmark OpenAI deal ($20B over 10 years) and AWS global distribution partnership reduce prior customer concentration risk and establish mainstream AI infrastructure credibility
Profitable growth: $510M revenue (76% YoY) with $88M net income in 2025; dominant position in emerging latency-sensitive AI inference market

Watch for

Customer concentration: 86% of 2025 revenue from two linked UAE entities (MBZUAI 62%, G42 24%); 77.9% of accounts receivable from MBZUAI alone; CFIUS regulatory and reputational risk persists despite 2025 clearance
Q1 2026 revenue growth expected to slow due to warrant accounting adjustments; market will scrutinize actual customer diversification progress beyond headline deals; valuation multiples risk correction if OpenAI/AWS ramp slower than 2025 growth rates
Post-IPO stock volatility ($95B valuation implies steep growth expectations); intense competition from Nvidia (86% AI data center share), SambaNova, Groq; employee retention concerns noted in Glassdoor reviews citing fast-paced environment and burnout risks

Recent moves

Key Information

Industry: AI Hardware
Founded: 2016
Headquarters: United States

Frequently Asked Questions

What is Cerebras and what does it manufacture?

Cerebras designs and manufactures wafer-scale silicon chips for AI inference. Founded in 2016 by former SeaMicro engineers, the company offers the CS-3 processor—a single chip with 4 trillion transistors—plus cloud inference services to cloud providers and enterprises needing ultra-low latency AI.

How fast is the Cerebras CS-3 compared to Nvidia GPUs?

Cerebras claims the CS-3 delivers 2,000+ inference tokens per second, roughly 8 times faster than Nvidia's H200 GPU for single-user latency workloads. The wafer-scale architecture's 44GB on-chip memory and 900,000 cores enable ultra-high memory bandwidth critical for real-time AI applications.

What was Cerebras's IPO valuation and stock performance?

Cerebras went public May 14, 2026, raising $5.55 billion at a $95 billion post-IPO valuation—the largest U.S. tech IPO since Uber in 2019. The stock opened 68% above the $185 IPO price but fell roughly 10% the following week, signaling investor caution around valuation.

What major customers does Cerebras have?

Cerebras serves OpenAI, Amazon Web Services, Meta, IBM, and Mistral. Customer segments include cloud AI infrastructure providers, large AI model makers, enterprises in finance and search, life sciences researchers, and sovereign AI initiatives in the Middle East. Each segment values the chip's low-latency performance.

Why is Cerebras facing customer concentration risk?

Cerebras faces significant concentration risk: 86% of 2025 revenue came from two linked UAE entities—Mohamed bin Zayed University of AI at 62% and G42 at 24%. Accounts receivable concentration from MBZUAI alone reached 77.9%. CFIUS cleared the path in 2025, but reputational and customer-diversification risks persist.

Did Cerebras become profitable after launching its inference service?

Yes. After launching its inference-as-a-service platform in August 2024, Cerebras's growth accelerated. 2025 revenue jumped 76% to $510 million, and the company swung to an $88 million profit from a $481.6 million loss in 2024. Landmark OpenAI and AWS deals further de-risked customer concentration.

How Cerebras compares

Direct head-to-head against 2 competitors. Picked by 7wData.

Positioning: AI inference hardware and services using wafer-scale silicon chips, sold to cloud providers and enterprises seeking ultra-low latency for real-time AI applications.
Customer segments: Cloud AI infrastructure providers scaling inference capacity (OpenAI, AWS)
Strengths: Wafer-scale architecture delivers ultra-high memory bandwidth and single-user latency far exceeding GPU clusters, critical for real-time inference and agentic AI workloads
Watch for: Customer concentration: 86% of 2025 revenue from two linked UAE entities (MBZUAI 62%, G42 24%); 77.9% of accounts receivable from MBZUAI alone; CFIUS regulatory and reputational risk persists despite 2025 clearance
Recent moves: Cerebras closes Series G funding of $1.1B at $8.1B valuation, led by Fidelity and Atreides Management

Positioning: Cloud inference service on LPU custom silicon; repositioned after Nvidia licensed its IP and absorbed its founding engineering team (December 2025).
Customer segments: AI model developers and enterprises needing low-cost, low-latency inference via API; overlaps directly with Cerebras cloud customer profile.
Strengths: GroqCloud developer platform with broad open-source model support and documented sub-second time-to-first-token at commodity pricing.
Watch for: Founder Jonathan Ross and core engineering team departed to Nvidia (December 2025); new CEO Simon Edwards leads a company running on Nvidia-licensed IP.
Recent moves: December 2025: non-exclusive IP and talent deal with Nvidia at approximately $20 billion; founder and core engineering team transferred to Nvidia.

Positioning: Full-stack dedicated AI chip and inference platform targeting enterprise and sovereign deployments, with a strategic Intel manufacturing partnership.
Customer segments: Financial services, telco, energy, and sovereign AI buyers; SoftBank deploying SN50 for Asia-Pacific enterprise inference.
Strengths: SN50 chip supports 10 trillion-parameter models and 10 million-token context lengths via three-tier memory, built for long-context production workloads.
Watch for: Intel partnership emerged only after acquisition talks reportedly failed (CNBC, February 2026); Vista Equity Partners-led round introduces PE exit pressure.
Recent moves: February 2026: SN50 chip launched, $350 million Series E raised led by Vista Equity Partners, and Intel multi-year collaboration announced.

Sources

techcrunch.com — Series G funding details, CEO Andrew Feldman background, inference services launch date (August 2024), new data center locations 2025
www.cnbc.com — IPO pricing, share count, valuation, revenue and profitability metrics for 2025, net income swing from loss
www.techtimes.com — UAE customer concentration (86%), MBZUAI revenue share (62%), G42 revenue share (24%), accounts receivable concentration, CFIUS regulatory history
www.nextplatform.com — OpenAI $20B deal details, AWS binding term sheet, customer concentration issues in prior year, agentic AI market positioning
www.cerebras.ai — Product portfolio (CS-3, cloud/dedicated/on-premise options), customer list, inference performance positioning
www.cerebras.ai — Series H funding amount, valuation, and lead investor