Cerebras IPO Pops 108% on Day One — A Signal That AI Inference Is Breaking Nvidia's Grip

What Happened

On May 14, AI chip company Cerebras Systems opened at $385 — a 108% pop from its $185 IPO price. The stock closed at $311, putting Cerebras at a $66B market cap by end of day. With 30 million shares priced at $185, the company raised $5.5B, making it the largest tech IPO of 2026 so far. The IPO price itself was revised upward twice during the roadshow (original range: $115–$125).

What’s striking is the speed of the turnaround. Cerebras lost $500M in 2024. In 2025, it posted $510M in revenue (76% YoY growth) and $237.8M in net income — its first profitable year. The same company whose 2024 IPO attempt was derailed by CFIUS scrutiny over Group 42’s Abu Dhabi backing is now the most-watched listing of the season.

Layered on top of this: SpaceX is preparing its own IPO at a $170B valuation. University of Washington’s $50M stake reportedly grew 30×. Capital is flowing back into mega-private tech.

What This Means for Founders

Cerebras makes chips for AI inference, not training. While Nvidia still dominates training, the inference market — where models are served to actual users — just got institutional validation as a multi-vendor space. The customer list reinforces this: OpenAI (in a circular equity-revenue relationship), UAE’s G42, Saudi Arabia’s MBZUAI, and AWS are all on board.

Two implications for founders. First, inference is overtaking training as the dominant AI cost line. The fact that OpenAI itself depends on Cerebras while simultaneously building its own datacenters tells you they’re hedging against single-vendor lock-in at the highest levels. If your unit economics assume Nvidia as the only compute option, you’re pricing for a world that’s quietly disappearing.

Second, the “8 years burning $8M a month, then profitable in 12 months” arc is a reminder that infrastructure plays look like failures until they don’t. The market punishes — until structural demand catches up.

What You Can Do Now

1. Audit your inference line items: Break down your monthly LLM spend by training, inference, and embedding. Inference-heavy services are the first to benefit from alternative chip providers like Cerebras, Groq, and SambaNova.
2. Study circular deal structures: The OpenAI-Cerebras-G42 model — where capital, supply, and revenue are interwoven — is becoming standard in AI infrastructure. Ask whether your supply chain can plausibly evolve in this direction.
3. Build vendor leverage: Multi-cloud, multi-chip architectures are no longer just a risk hedge — they’re pricing leverage. Start designing for it before you need it.