2026-05-15 — views

Cerebras drops 10% on Day 2 after 68% IPO pop — OpenAI-concentration risk lands

Read this because First public verdict on whether the market believes Nvidia has a credible challenger — or whether Cerebras' OpenAI/AWS revenue concentration was priced for a fall.

CBRS closed Friday at $279.72, -10% from Thursday's $311.07 debut. Still +51% over the $185 IPO price; market cap ~$60B on 215M shares outstanding.

Cerebras Systems (CBRS) closed its second day of public trading at $279.72 on Friday, May 15 — down roughly 10% from Thursday’s $311.07 close. Thursday’s debut had soared 68% from the $185 IPO price, hitting an intraday high near $385 and briefly pushing the company toward a $100B market cap. After Friday’s pull-back, the implied market cap settled around $60B on roughly 215M shares outstanding. The stock remains +51% above the IPO price, but the reset was sharp enough that it now sets the public-market reference for AI-chip alternatives.

What the reset means

This was the first public verdict on Cerebras’ core thesis: that the Wafer Scale Engine 3 — an entire silicon wafer treated as one chip — can credibly displace Nvidia GPUs for high-throughput inference. Three concerns dominated the Friday tape:

Customer concentration. OpenAI, Amazon Web Services, and Meta together drive a large share of Cerebras’ revenue. A revenue base built on three counterparties is fragile in a way the headline numbers don’t show.
TAM already priced in. OpenAI’s inference workload is enormous, but bears argued much of that future ramp was already in the $311 close. Friday’s 10% give-back is the market re-pricing optionality, not capitulating.
Wafer-scale architecture history. Wafer-scale designs have shipped before and historically struggled outside hero workloads. The bull case requires Cerebras to be the exception, not the rule.

The growth case — what bulls bought on Friday

The financial trajectory remains striking: $24.6M revenue in 2022 → $510M in 2025 (a 76% YoY jump for the most recent year), against a $145.9M operating loss reflecting heavy R&D and capacity buildout. The 30M-share offering raised $5.55B, the largest US tech IPO since Uber, and the book was reportedly 20× oversubscribed.

Bulls argue the customer concentration is a function of being early — that the same OpenAI/AWS imprimatur is exactly what convinces enterprises 2-3 quarters out to commit serious inference spend to a non-Nvidia path. The next two earnings reports will resolve whether the customer list broadens or narrows.

Practitioner note

For builders running inference workloads at scale:

Actually benchmark before you switch. Cerebras Cloud is the cheapest way to run a real workload on the WSE-3. Pick a representative model and decode pattern, measure tokens-per-second-per-dollar end-to-end (including network egress), and compare to your current H200/B200 setup. Headline FLOPs are a sales sheet, not a serving cost.
Watch “available capacity” language in Cerebras quarterly updates. If utilization is gated, the bottleneck is still Nvidia — Cerebras isn’t shipping volume yet. If capacity is open with healthy queue depth, the supply story is real.
Don’t rewrite your serving stack for a single vendor on Day 2. A 10% public-market swing is noise relative to the rewrite cost. The right move for most teams is dual-source preparation — make your serving abstraction vendor-agnostic — not a hard switch.

The under-considered angle: the IPO proceeds matter more than the price. Cerebras now has $5.55B of fresh capital to fund capacity expansion. If they execute on capacity in 2026 H2 and earn a fourth and fifth major customer outside the OpenAI/AWS/Meta triangle, the concentration risk dissolves and the bear thesis goes with it. That’s the trigger worth watching, not the stock price tick by tick.

Cerebras drops 10% on Day 2 after 68% IPO pop — OpenAI-concentration risk lands

What the reset means

The growth case — what bulls bought on Friday

Practitioner note

Sources

Tags