Cerebras Systems (CBRS) IPO: Reading the Day 1 print

Cerebras opened at $350 and closed Day 1 at $311.07 (+68.15%), valuing the company at approximately $95 billion. The $5.55B raise on 30M shares marks the largest US tech IPO since Uber's 2019 debut. The reaction validates the market's appetite for pure-play AI infrastructure exposure — but the underlying setup carries asymmetric tail risks that deserve careful framing.

The bull case in three parts:

1. Architectural differentiation. The Wafer-Scale Engine 3 is not a GPU variant. Building compute on a single silicon wafer eliminates inter-chip latency penalties — a meaningful advantage for AI inference workloads where throughput economics dominate. This positions Cerebras as the most differentiated NVDA alternative currently scalable.
2. Customer roster validates the product. AWS, Meta, Mistral, Notion, Perplexity, Cognition, and AlphaSense represent a credible cross-section of inference-heavy workloads. The reported multi-year OpenAI agreement valued at >$20B anchors the future revenue pipeline.
3. TAM tailwind. Inference compute is widely expected to overtake training compute by 2027–2028. Cerebras is one of very few pure-play exposures to that mix shift outside the NVDA ecosystem.

The bear case sits in the footnotes:

• Customer concentration is severe. MBZUAI accounted for 62% of 2025 sales, with G42 contributing another 24% — combined ~86% UAE-linked. That's a single-region revenue base on which to underwrite a ~$95B valuation.
• Lock-up overhang. Day 90 / Day 180 lock-up expiries can introduce insider supply pressure that's not yet priced.
• Competitive intensity. NVDA's Blackwell + Rubin roadmap, AMD's MI400 series with HBM4, and hyperscaler in-house silicon (Trainium 3, TPU v7, MAIA) all target the same inference workloads.

What I'm watching post-debut:

• 4Q26: Customer concentration trajectory in the next 10-Q
• 1H27: WSE-4 roadmap and shipped-unit data
• Lock-up calendar inflection points
• Sequential growth in non-UAE deployments

Framing for portfolios: This is a high-variance, high-conviction-needed name. Position sizing matters far more than entry timing.

What's your single biggest watch item over the next 3 quarters?