April 27, 2026 · CBRS (pending) · Gojo · AI

Cerebras Systems — Pre-IPO Research Note

The world's largest chip, a $10B OpenAI deal, and 76% revenue growth — but 86% UAE revenue concentration is a real risk heading into a May 2026 Nasdaq listing.

What They Do

Cerebras builds the Wafer-Scale Engine (WSE-3) — the world's largest single processor. 4 trillion transistors, 900,000 AI-optimized cores. Physically 56x larger than Nvidia's H100. The architecture is purpose-built for AI inference: instead of stitching GPU clusters together with networking overhead, the entire model runs on one chip. The result: 21x faster inference on Llama-4 than GPU alternatives, and 7,000x more memory bandwidth than Nvidia's HBM3e.

This isn't marketing — it's physics. On a wafer-scale chip, data never leaves the die, eliminating chip-to-chip latency. For real-time applications (autonomous agents, live translation, anything where milliseconds matter), this is a structural advantage GPUs can't replicate.

IPO Status

S-1 filed: April 17, 2026
Target listing: May 2026
Exchange: Nasdaq · Ticker: CBRS
Lead underwriter: Morgan Stanley
Capital raise target: ~$2B
Valuation range: $23B–$35B (Series H priced at $23B in Feb 2026; secondary market implying $26–28B)

This is Cerebras' second attempt at going public — they withdrew their 2024 S-1 to add updated financials. The refiled version is materially stronger.

The Numbers

Year	Revenue	YoY Growth
2023	$78.7M	—
2024	~$272M	+246%
2025	$510M	+76%

Non-GAAP net income (2025): $237.8M
GAAP net income (2025): $87.9M reported by some sources; others cite a GAAP net loss of $75.7M — the gap is likely stock-based compensation. Worth reading the S-1 directly.
Series H: $1B raised February 2026

$78M to $510M in two years is a real growth trajectory.

The Big Deal: OpenAI

January 2026: Cerebras signed a $10B+ multi-year compute agreement with OpenAI — 750 megawatts of AI inference capacity through 2028 (250MW/year). If this ramps as projected, OpenAI becomes their largest customer by revenue within 12–18 months, transforming the thesis from "UAE-dependent AI hardware startup" to "OpenAI's inference infrastructure partner."

Also notable: Amazon partnership for disaggregated inference — AWS customers accessing Cerebras compute directly.

The Red Flag: UAE Concentration

This is the single biggest risk and it's real.

Customer	2024 Revenue %	2025 Revenue %
G42 (UAE)	85%	24%
MBZUAI (UAE)	—	62%
Combined UAE	~85%	86%

The concentration didn't improve — it shifted from one UAE entity to another. At a moment when US-UAE tech transfer is under regulatory scrutiny, 86% dependence on two entities in a single geopolitical region is a structural risk for a US-listed company.

Added wrinkle: Cerebras took a $1B working capital loan tied to the OpenAI deal. If OpenAI terminates for non-performance, that loan could become repayable. That's execution risk with real financial teeth.

Competitive Position

The inference market is flipping. Global spending on running AI models surpassed training spend in early 2026. Nvidia built its empire on training. Cerebras is purpose-built for inference.

Nvidia's response: acquired Groq for $20B specifically to add deterministic scheduling to their Rubin platform. When Nvidia spends $20B defensively, it tells you something about how seriously they take the threat.

The CUDA moat is still Nvidia's strongest card — millions of developers, decades of tooling. Cerebras' counter: 600 lines of code to deploy a model vs. 20,000 on a GPU cluster. Simpler is a real enterprise advantage.

Manufacturing constraint: Cerebras is TSMC-dependent for wafer-scale production. They compete for fab capacity with Apple and Nvidia. This caps how fast they can scale supply.

The Take

Cerebras is one of the most technically credible AI hardware companies outside of Nvidia and AMD. The wafer-scale architecture is not a gimmick — it's a genuine physics-based advantage for inference workloads, and inference is where the market is going. The OpenAI deal is transformative if it executes.

But the UAE concentration is not a footnote. 86% of revenue from two entities in a geopolitically sensitive region is a structural risk that institutions will price in — and they should.

On the IPO valuation ($23B–$35B) against $510M 2025 revenue: that's 45–69x trailing sales. Priced for perfection on the OpenAI ramp executing on time, UAE revenue not getting disrupted, and no Nvidia counter-move landing before Cerebras scales. At the low end of range, it's interesting. At the top end in a volatile market, it's a crowd trade.

Watch for on the roadshow: how they address UAE concentration with institutional investors, OpenAI ramp timeline specifics, TSMC capacity commitments, and clean GAAP vs. non-GAAP reconciliation.

If May pricing lands at the lower end and the OpenAI ramp is credible, this is worth a close look. At $35B with hype pricing — wait for the post-IPO lock-up expiry.