Cerebras raises $5.5 billion in IPO, stock surges 108% on debut to reach $66 billion valuation

Cerebras Systems went public on May 14, 2026, raising $5.5 billion in what became the largest tech IPO of the year to date. The AI inference chip maker priced its shares at $185, above the already-raised expected range, and opened its first day of trading at $385 — a 108% premium to the IPO price. The stock closed at $311, giving Cerebras a market capitalization of approximately $66 billion at closing prices.

What's new

Cerebras had been working toward an IPO for years, facing successive delays: a national security review of its investment from UAE-based Group 42, concerns from regulators about revenue concentration, and repeated filings and refiling cycles. The company refiled for its Q2 2026 IPO after addressing those concerns, and the final offering was priced above the expected range — a signal of strong institutional demand.

Key financials from the offering:

• Amount raised: $5.5 billion
• IPO price: $185 per share
• Opening trade: $385 per share (+108%)
• Closing price: $311 per share
• Fully diluted valuation at IPO price: $56.4 billion
• Fully diluted valuation at closing: ~$66 billion

Cerebras reported 2025 revenue of $510 million — up 76% year over year — and turned its first net profit: $237.8 million in 2025, compared to approximately $500 million in net losses in 2024. The path from near-bankruptcy-level losses to profitability in a single year was a central selling point in the IPO roadshow.

Major customers disclosed include OpenAI, Group 42, Mohamed bin Zayed University of Artificial Intelligence, and Amazon Web Services. The OpenAI partnership — a $10 billion+ compute deal announced in January 2026 committing to 750 megawatts of Cerebras capacity through 2028 — was central to the company's narrative about demand visibility.

Founder stakes at IPO price: CEO Andrew Feldman at approximately $1.9 billion; CTO Sean Lie at approximately $1 billion.

Context

Cerebras makes wafer-scale processors — its CS-3 is built on a single 46,000 mm² die, compared to the 800 mm² of a typical Nvidia GPU. The architecture is designed to minimize the inter-chip communication overhead that limits throughput when large models are partitioned across thousands of conventional GPUs. For inference workloads in particular, Cerebras claims substantially better throughput per unit of compute and power.

The chip's architecture has proven especially effective for long-context and high-throughput inference tasks. The deal with OpenAI specifically targets Codex and other agentic workloads that generate large amounts of tokens per session. As agentic AI usage scales — Codex usage grew 137-fold between August 2025 and June 2026 according to OpenAI's own research — the economics of inference hardware become increasingly important.

Why it matters

Cerebras' successful IPO is a significant moment for the AI chip market beyond Nvidia. The company's path — from regulatory limbo to a $66 billion public market debut in 12 months — reflects how quickly institutional conviction about alternative AI silicon has shifted.

For the broader chip ecosystem, the Cerebras IPO signals that the market now believes inference-optimized silicon can be a standalone public company at scale, not just an acquisition target for larger chip makers. That has implications for other inference chip startups — Groq (acquired by Nvidia in December 2025), SambaNova, d-Matrix, and others — and for the competitive dynamics around Nvidia's dominance in AI compute.