Cerebras Systems (CBRS) - Fundamental Analysis Report 2026 (Updated)
Dear Readers, Welcome to Deep Research Global.
Let’s analyze the topic in detail.
Executive TL;DR
Cerebras Systems (CBRS) debuted on Nasdaq on May 14, 2026 under ticker CBRS at $185 per share, rocketing to a close of $311.07 on day one, in what became the biggest tech IPO of 2026 at the time.
The company reported $510 million in FY2025 revenue, up 76% year over year, alongside $237.8 million in non-GAAP net income, but 86% of revenue was concentrated in two UAE-affiliated customers (MBZUAI and G42).
A landmark $10+ billion multi-year deal with OpenAI, signed in January 2026, deploys 750 megawatts of wafer-scale systems and meaningfully diversifies the customer base going forward.
The bull case rests on inference dominance, where the WSE-3 has set repeated world records for tokens per second; the bear case rests on customer concentration, single-foundry dependence on TSMC, and a valuation that trades at roughly 50x to 75x trailing revenue.
Recommended - Read Full Reports
Read All Reports
Table of Contents
Executive TL;DR
Introduction
Cerebras Systems Company Profile: Key Facts Snapshot
The Cerebras Investment Thesis
Why investors are paying attention
The three pillars I weigh most heavily
What the thesis requires to remain intact
Cerebras Business Model Overview
Two product lines, one architectural philosophy
How a CS-3 system actually generates revenue
How the inference cloud business compounds
The vertical integration question
Cerebras Revenue Analysis
A four-year revenue trajectory worth studying carefully
The customer concentration reality
The backlog story
Latest Quarterly Earnings, Guidance, Margins and Earnings Quality
What the most recent disclosures show
Gross margin trajectory
Earnings quality considerations
EPS Trajectory and Cash Flow Mechanics
From non-GAAP profitability to public-market scrutiny
Cash flow mechanics tell a different story
Balance Sheet Health
A fortress balance sheet, for now
What balance sheet risks remain
Cerebras Segment-by-Segment Teardown
Segment 1: The Wafer Scale Engine 3 (WSE-3)
Segment 2: The CS-3 system
Segment 3: Cerebras Cloud and inference services
Segment 4: Software and developer tools
Major Cerebras Competitors
List of major competitors
Cerebras vs. Nvidia
Cerebras vs. Groq
Cerebras vs. SambaNova
Cerebras vs. AMD
Cerebras vs. hyperscaler captive silicon
Cerebras Strategic Context
The OpenAI relationship reshapes the competitive map
The G42 anchor partnership
The hyperscaler distribution layer
Geopolitical positioning
Cerebras Valuation Framework
What the public market is currently paying for
How to think about a 75x to 100x sales multiple
The framework I use
Bull, Base and Bear Case Scenario Analysis
Bull case
Base case
Bear case
Key Risks for Cerebras
Risk 1
Risk 2
Risk 3
Risk 4
Risk 5
Risk 6
Risk 7
Catalysts to Watch
Near-term catalysts (next 6 months)
Medium-term catalysts (6 to 18 months)
Longer-term catalysts (18 to 36 months)
My Final Thoughts
Latest Analyst Price Targets
Official Sources & Data
Disclaimer: This analysis is for informational & educational purposes only and should not be construed as investment advice. Investors should conduct their own due diligence and consult with their personal financial advisors before making investment decisions. Past performance does not guarantee future results.
Introduction
When Cerebras Systems opened for trading at $350 on its IPO debut, doubling its $185 offer price within minutes, public markets effectively re-priced the entire alternative AI silicon category overnight.
This is no ordinary chip company. Cerebras builds a single AI processor the size of a dinner plate, packed with four trillion transistors and 900,000 cores, and it has spent the last few quarters convincing the world’s most demanding AI labs that wafer-scale physics simply beats GPU clusters at the workloads that matter most for the next phase of generative AI.
For investors, the question is no longer whether Cerebras has interesting technology.
The question is whether the company can convert a stunning revenue trajectory (from $24.6 million in 2022 to $510 million in 2025) into a durable, diversified business that justifies a market capitalization that has hovered above $50 billion since its first week of trading.
In this report, we’ll walk through every meaningful line of the S-1 prospectus, the architecture of the WSE-3, the economics of the CS-3 system, the precise contours of the OpenAI and G42 relationships, segment-by-segment competitive positioning against Nvidia, Groq, SambaNova and AMD, and a clean bull / base / bear scenario framework.
We’ll also lay out the specific catalysts (and the specific risks) we are watching across the rest of fiscal 2026.
Cerebras Systems Company Profile: Key Facts Snapshot
Cerebras Systems Inc. was founded in 2015 in Sunnyvale, California by a team led by Andrew Feldman, who previously co-founded SeaMicro (acquired by AMD for $355 million in 2012).
The founding hypothesis was simple and contrarian.
Rather than tiling thousands of small GPUs and paying the bandwidth, latency and power tax of interconnects, the company would build a single processor that occupied the entire silicon wafer.
Company: Cerebras Systems Inc.
Ticker: CBRS (Nasdaq Global Select)
IPO Date: May 14, 2026
IPO Price: $185.00 per share
Headquarters: Sunnyvale, California
Founded: 2015
CEO and Co-Founder: Andrew Feldman
Chief System Architect: Jean-Philippe Fricker
Chief Architect: Michael James
Flagship Product: CS-3 system powered by WSE-3
FY2025 Revenue: $510.0 million
FY2025 Non-GAAP Net Income: $237.8 million
Largest Customers (FY2025): MBZUAI, G42, OpenAI
The executive bench includes co-founders Jean-Philippe Fricker (Chief System Architect) and Michael James (Chief Architect), both of whom have been with the company since its founding. Andrew Feldman has spent more than two decades shipping silicon and systems, and he has now taken three companies through significant liquidity events.
The product portfolio centers on the CS-3 system, a rack-scale machine that integrates a single Wafer Scale Engine, custom packaging, advanced cooling and an internal fabric.
Around that hardware, Cerebras operates a growing cloud inference service that serves enterprise developers via API.
Together these comprise a vertically integrated stack that very few rivals can match.
The Cerebras Investment Thesis
Why investors are paying attention
The core thesis around Cerebras is structural, not narrative.
AI inference is shifting from a “nice to have” workload into the single largest line item in hyperscaler capital expenditure, and the workload itself is changing shape in ways that favor wafer-scale architectures.
Long context windows, agentic tool use, multi-step reasoning and real-time multimodal interaction all reward memory bandwidth more than peak FLOPs.
That is precisely where the WSE-3 has a defensible architectural moat.
The chip carries 44 GB of on-chip SRAM with 21 PB/s of memory bandwidth, which is reported to be roughly 7,000 times the on-die bandwidth of an Nvidia H100.
For workloads where memory traffic, not arithmetic throughput, is the binding constraint, this is a generational gap rather than an incremental one.
The three pillars I weigh most heavily
First, the inference performance gap is empirically documented.
Cerebras has publicly recorded over 2,500 tokens per second on the 400-billion-parameter Llama 4 Maverick model and has published direct benchmarks showing the CS-3 running 21x faster than Nvidia’s DGX B200 Blackwell on certain models.
Independent third-party evaluators have corroborated that Cerebras has outperformed Blackwell on certain inference workloads in 2026.
Second, the customer book has matured well beyond the original G42 anchor.
The September 2025 Series G raise of $1.1 billion at an $8.1 billion valuation referenced AWS, Meta, IBM, Mistral, Cognition, AlphaSense and Notion as production customers, plus hundreds of enterprises and government customers behind them.
The OpenAI agreement signed in January 2026 added the most important AI customer in the world.
Third, the financial profile has crossed an important threshold.
Cerebras reported $237.8 million in non-GAAP net income in the trailing twelve months ending December 2025, which means the company has demonstrated unit economics at scale, not merely top-line growth.
What the thesis requires to remain intact
The thesis requires three things to remain true.
The inference performance advantage must persist through Nvidia’s Rubin generation and beyond.
Revenue concentration must decline meaningfully from the 86% combined MBZUAI plus G42 share of 2025.
And the OpenAI deployment must execute on schedule, ramping over 2026 to 2028.
The three-leg thesis stool:
1. Architectural moat in memory-bound inference
2. Customer diversification away from UAE concentration
3. OpenAI rollout converts headline contract value into recognized revenue
If any one of those legs breaks, the multiple compresses sharply.
If all three hold, the company has the potential to become a structurally important second pillar of AI compute alongside Nvidia.
Cerebras Business Model Overview
Two product lines, one architectural philosophy
Cerebras operates two reinforcing revenue streams.
The first is system sales, where customers buy CS-3 systems outright (or in clusters of dozens to thousands of units) and operate them in their own data centers or in colocation.
The second is cloud inference, where Cerebras hosts the systems and sells access via API at a pay-per-token price.
Both lines share the same silicon, the same software stack and the same operating economics, which is rare for a hardware company of this scale.
How a CS-3 system actually generates revenue
A single CS-3 system is priced at roughly $2 million to $3 million per unit at list, and large multi-system deployments stretch into hundreds of millions of dollars per customer.
The Condor Galaxy supercomputer network with G42 was an early proof point, delivering 8 exaFLOPs across CG-1 and CG-2 before scaling to CG-3.
CS-3 system revenue mechanics:
List price per CS-3 system: ~$2M to $3M
Typical cluster size (enterprise): 4 to 64 systems
Hyperscale cluster size: 256+ systems
Approximate Condor Galaxy 1 spend: $100M+ over multi-year deployment
Recurring revenue layer: Software, support, expansion
Systems revenue is lumpy by nature.
Cerebras typically signs multi-year purchase commitments with delivery and acceptance milestones spread across quarters, which is why the company’s $24.6 billion backlog is dominated by a single customer.
How the inference cloud business compounds
The cloud business is the more strategically valuable engine, even if it is currently smaller.
Cerebras Inference is priced at $6 per million input tokens and $12 per million output tokens for Llama 3.1 405B, with comparable tiers for other models. The cloud workflow brings high-value developer relationships, recurring usage data and incremental gross margin on already-deployed silicon.
Hugging Face integrated Cerebras Cloud into the Hugging Face Hub in 2025, exposing the platform to millions of developers.
AWS has also made Cerebras-powered inference available through its Bedrock catalog, which is one of the few examples of Nvidia’s largest customer simultaneously distributing Nvidia’s most credible rival.
The vertical integration question
A persistent debate in AI hardware is whether the long-term winners will be horizontal accelerator vendors selling chips into someone else’s data center, or vertically integrated AI cloud operators owning chips, systems and services.
Cerebras is doing both, and the IPO proceeds give it the capital to keep both lanes open.
Cerebras Revenue Analysis
A four-year revenue trajectory worth studying carefully
Cerebras’ top line has gone through three distinct phases.
Between 2022 and 2023, revenue moved from $24.6 million to $78.7 million, reflecting the first generation of meaningful system sales and the early Condor Galaxy phases.
Between 2023 and 2024, revenue jumped to $290.3 million, a tripling driven primarily by the G42 ramp.
Between 2024 and 2025, revenue grew to $510 million, with the customer mix beginning to broaden as MBZUAI emerged as a new anchor and OpenAI signed on at year-end.
Annual revenue snapshot:
FY2022: $24.6 million
FY2023: $78.7 million (+220% YoY)
FY2024: $290.3 million (+269% YoY)
FY2025: $510.0 million (+76% YoY)
The deceleration from 269% to 76% growth is mathematically natural at this scale, but it also reflects a transition from one large customer ramping aggressively to a more diversified portfolio building gradually.
Forward growth expectations are anchored to the OpenAI deployment schedule, which begins to contribute materially in 2026.
The customer concentration reality
The S-1 prospectus is unusually candid about concentration.
In 2024, G42 represented 85% of revenue. In 2025, the composition shifted: MBZUAI became 62% of revenue and G42 dropped to 24%, while OpenAI began contributing at the end of the year. Combined, the two UAE-affiliated entities still accounted for 86% of FY2025 revenue.
The S-1 also discloses that as of December 31, 2025, MBZUAI alone accounted for 77.9% of accounts receivable.
This is the kind of disclosure that long-term investors should track quarterly, because the trajectory of that single number tells you more about the durability of the business than most other line items.
The backlog story
Cerebras carries a remarkable $24.6 billion backlog at the time of the IPO. The composition of that backlog matters as much as the headline figure.
A meaningful share is driven by