Cerebras' $5.55B IPO Unveils the World’s Largest AI Chip Challenging Nvidia

On May 14, 2026 Cerebras Systems listed on Nasdaq at $185 per share, issuing 30 million shares and raising $5.55 billion, making it the largest IPO of the year. The stock surged 108% intraday, pushing the market value past $100 billion, with Bloomberg reporting demand exceeding supply by more than 20‑fold.

Cerebras Founding and Wafer‑Scale Vision

Founded in 2015 in Sunnyvale by Andrew Feldman, Gary Lauterbach, Michael James, Sean Lee and Jean‑Philippe Fricker—formerly of SeaMicro, which AMD bought for $334 million—the company pursued a bold vision: treat an entire silicon wafer as a single chip. Traditional semiconductor practice slices a wafer into hundreds of small dies to avoid defects; Cerebras instead developed a fault‑tolerant architecture that bypasses defective regions, akin to large‑scale data‑center server redundancy.

The engineering challenges were unprecedented: any defect could ruin the whole wafer, requiring a novel redundancy scheme; packaging a 25 kW, plate‑size die demanded a custom water‑cooling system and specialized bonding, power delivery, and data‑pipe solutions. Observer noted the company burned $8 million per month and spent $200 million on packaging R&D, even inventing a machine to drill 40 screws into a wafer without cracking it.

Early Commercial Struggles and Turnaround

Initial S‑1 filings admitted AI was still nascent and the market unreceptive, with 2022 revenue only $25 million from a few life‑science customers. After the 2024 CFIUS‑related withdrawal and a 2025 approval, Cerebras diversified beyond its Emirati partner G42, whose share of revenue fell from near‑100% to about 86%.

Revenue then accelerated: $290 million in 2024, $510 million in 2025 (76% YoY), and projected $800 million in 2026, driven by OpenAI contracts and expanding cloud inference services. GAAP net profit turned positive in 2025 at $237.8 million (≈47% margin) after a $481.6 million loss in 2024.

Funding and Investor Landscape

Prior private rounds raised over $2 billion, including an $1.1 billion Series G round in Sep 2025 led by Fidelity and Atreides, valuing the company at $8.1 billion, and an $11 billion Series H round in Feb 2026 led by Tiger Global, pushing valuation to $23 billion. Major shareholders listed in the S‑1 include Benchmark, Foundation Capital, Eclipse Ventures, Alpha Wave, Feldman, Lee, and G42, with OpenAI holding warrants for 33.4 million shares conditional on purchasing Cerebras compute capacity.

Wafer‑Scale Engine (WSE) Product Line

The WSE series comprises three generations:

WSE‑1 : First commercial wafer‑scale chip, proving feasibility.

WSE‑2 : 2.6 trillion transistors, 850 k AI‑optimized cores, 40 GB on‑die SRAM.

WSE‑3 (current flagship): Largest commercial AI chip, integrated into the CS‑3 rack system delivering 125 PFLOPS per node and scalable to 256 EFLOPS with up to 2 048 nodes.

Artificial Analysis benchmarked CS‑3 as 21× faster than Nvidia’s DGX B200 while consuming only one‑third of its power and cost. In inference tasks, Cerebras claimed >15× speedup over top GPU solutions, and a carbon‑capture model showed up to 210× acceleration versus Nvidia H100.

Shift to Cloud Inference Services

Since 2024 Cerebras has transitioned from pure hardware sales to a cloud inference platform. 2025 cloud‑inference revenue reached $152 million, and the service is available via AWS Marketplace, allowing customers to access compute without owning hardware.

U.S. AI Chip Market Landscape

Nvidia remains the clear leader in training and inference, leveraging its CUDA ecosystem and rapid product cycles (Hopper, Blackwell, Rubin, Rubin Ultra). AMD’s MI300X rivals Nvidia H100 in some workloads and, as an H‑round investor in Cerebras, illustrates AMD’s strategic diversification despite its ROCm software lagging behind CUDA.

Specialized inference players include Groq (LPU architecture) and Google’s TPU (internal to Google services but offered via Cloud). Intel’s Gaudi line provides an alternative but has yet to threaten Nvidia’s market share.

Within this competitive field, Cerebras is the sole company that has successfully commercialized wafer‑scale integration, granting it structural advantages in latency and bandwidth for inference. If wafer‑scale architectures become a dominant AI‑compute paradigm, Cerebras could emerge as a formidable challenger to Nvidia.