Morgan Stanley Initiates Coverage on Cerebras with Overweight Rating and $250 Target Price

Overview

Morgan Stanley initiates coverage on Cerebras Systems with an Overweight rating and a $250 target price. The report highlights that Cerebras possesses the only commercially deployed wafer-scale processor technology in the industry, capable of addressing memory bandwidth and communication bottlenecks in AI inference. As AI workloads increasingly shift toward inference-intensive tasks, demand for low-latency inference is surging from late 2025. Cerebras has signed a 750MW capacity commitment agreement with OpenAI, supporting the forecast of reaching $6 billion in revenue by 2028, corresponding to approximately 1% of the AI processor market share.

Core views

Market Opportunity for Low-Latency Inference: The report notes that demand for low-latency inference is surging from late 2025 and could account for more than 10% of inference hardware sales. As inference models generate more tokens and execute more complex workflows, performance is becoming limited by memory bandwidth and communication rather than raw compute power. Cerebras' wafer-scale engines are specifically designed to address these bottlenecks, positioning them as one of the few architectures optimized for the emerging ultra-fast inference market. Differentiated Technological Advantage: Cerebras' wafer-scale engine (WSE) is the industry's only commercially deployed wafer-scale processor. The WSE-3 contains 4 trillion transistors, 900,000 AI-optimized compute cores, 44GB of on-chip SRAM memory, and 21PB/s memory bandwidth. Compared to NVIDIA Blackwell GPUs, the WSE-3 is approximately 58 times larger and offers over 2,500 times the memory bandwidth. This architecture keeps compute, memory, and communication on a single silicon die, reducing inter-chip communication needs. Growth Driven by OpenAI Agreement: Cerebras has signed a take-or-pay agreement with OpenAI for 750MW of inference capacity, to be deployed in three phases of 250MW each over the next three years. The first 250MW will be delivered via Cerebras Cloud with a base term of 3 years extendable to 5 years. OpenAI also holds an option to purchase an additional 1.25GW of capacity through 2029. This agreement is expected to drive most of the company's growth, with OpenAI-related revenue comprising 49%, 85%, and 76% of revenues from 2026 to 2028 respectively. Financial Forecasts and Valuation: The report forecasts a 2026-2028 revenue CAGR of 178%, with Non-GAAP revenue reaching $6.4 billion in 2028. Valuation employs a 12x EV/Sales multiple on adjusted 2028 revenue of $6.0 billion (excluding warrant hedge revenue), resulting in a $250 target price. A 12x multiple aligns with high-growth small-to-mid-cap AI companies. Cerebras should command a premium due to its faster growth profile and exposure to the fastest-growing segment of AI infrastructure. Cloud vs. Hardware Business Models: The report analyzes differences between cloud and hardware business models. The cloud model increases capital expenditure requirements and data center resale revenue, while the hardware model recognizes revenue upfront and reduces capital needs. The base case assumes all OpenAI agreements are delivered via cloud, making cloud revenue the primary contributor by 2028.

Analysis framework

The report employs a bottom-up capacity deployment framework for modeling, based on implied lifetime contract value per GW. Total agreement value is estimated at approximately $30-40 billion/GW; taking the lower end and assuming a 4-year blended contract term, this implies annual revenue of approximately $7.5 billion/GW. Revenue is driven primarily by deployed capacity rather than contracted capacity, with quarterly revenue increasing as additional MW comes online. Valuation Methodology: The report compares Cerebras with AI semiconductor peers rather than traditional cloud providers, given its differentiated hardware IP and economic structure closer to a semiconductor company. The median EV/Sales multiple for AI semiconductor peers in 2028 is approximately 9x, with small-to-mid-cap AI companies trading at higher multiples. The report considers a 12x 2028 EV/Sales multiple reasonable, reflecting Cerebras' exceptional revenue visibility gained through committed capacity purchases. The report also analyzes differences between cloud and hardware business models: the cloud model increases capital expenditure requirements and data center resale revenue, while the hardware model recognizes revenue upfront and reduces capital needs. The base case assumes all OpenAI agreements are delivered via cloud, making cloud revenue the primary contributor by 2028.

Methodology notes

• Valuation MethodPrice-to-Sales (P/S) Valuation

  EV/Sales Multiple Valuation Approach

  The report values high-growth AI companies using an Enterprise Value to Sales (EV/Sales) multiple, selecting a 12x multiple for 2028 aligned with high-growth small-to-mid-cap AI companies. This approach is suitable for companies not yet profitable but experiencing rapid revenue growth.

• Industry/Industrial Analysis FrameworkSupply-demand framework

  AI Inference Market Supply and Demand Analysis

  The report analyzes the trend of shifting AI infrastructure spending from training to inference, with low-latency inference potentially accounting for 10-20% of the market. Using the framework estimated by NVIDIA's CEO (approximately 25% of processors in data centers used for low-latency inference), the report calculates a market opportunity of approximately $80 billion.

• Competition and Strategy FrameworkMoat / competitive advantage

  First-Mover Advantage Driven by Technological Differentiation

  The report emphasizes that Cerebras' wafer-scale technology is the only commercially deployed architecture in the industry, offering unique value propositions in memory bandwidth and communication efficiency. This technological differentiation constitutes a competitive moat, despite competition from NVIDIA, AMD, and others.

• Company Fundamentals and Financial FrameworkThree-Statement Reconciliation

  Analysis of Differences Between GAAP and Non-GAAP Financial Metrics

  The report provides a detailed explanation of the differences between Cerebras' GAAP and Non-GAAP financial metrics, including the impact of warrant hedge revenue and data center resale revenue. As the cloud business scales, these adjustments will lead to significant disparities in reported financial performance.

• Event Gamble and Behavioral FinanceExpectation Gap / Expectation Management

  Capacity Deployment Pace as a Key Expectation Variable

  The report identifies capacity deployment pace as the most critical driver of financial performance. Any acceleration or delay in deployment will have significant impacts on revenue recognition, margins, and cash flow, representing a key expectation variable for investors to monitor.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Cerebras Systems (CBRS.US)

  Primary target of initial coverage, benefiting from growing low-latency inference demand and OpenAI capacity agreement

  Strengths

  Only commercially deployed wafer-scale processor technology in the industry; more than 2,500x memory bandwidth advantage; 750MW committed capacity provides revenue visibility

  Weaknesses

  High customer concentration (OpenAI accounts for 76-85%); gross margin under temporary pressure in 2026 due to rented capacity; execution risk in cloud deployment

  Comparison

  Compared to AI accelerator vendors like NVIDIA and AMD, Cerebras focuses on the low-latency inference niche rather than general-purpose GPUs; technologically differentiated but faces intensifying competition

  Risks

  Capacity deployment delays; customer concentration risk; competitive risk (e.g., NVIDIA acquiring Groq assets)

• NVIDIA Corp. (NVDA.US)

  Major competitor; report mentions NVIDIA's acquisition of Groq assets would enhance low-latency inference competitiveness

  Strengths

  Dominant GPU business; world-class Mellanox networking capabilities; mature software ecosystem

  Weaknesses

  Lacks wafer-scale technology capabilities

  Comparison

  NVIDIA maintains leadership in general AI workloads, but Cerebras holds a performance advantage in communication-constrained inference workloads

  Risks

  Not elaborated in the report

Key data

• Target Price$250Based on 12x EV/Sales on 2028 adjusted revenue of $6 billion
• Current Stock Price$201.01Closing price on June 5, 2026
• 2026-2028 Revenue CAGR178%Core revenue forecast compound annual growth rate
• 2028 Forecasted Revenue$6.4 billionNon-GAAP core revenue; valuation uses $6 billion (excluding warrant impact)
• OpenAI Agreement Capacity750MWDeployed in three phases of 250MW each, with an additional 1.25GW option
• OpenAI Revenue Share49% in 2026 / 85% in 2027 / 76% in 2028High customer concentration
• Long-Term Gross Margin TargetApproximately 60%Pressure in 2026 due to temporary reliance on G42 rented capacity
• WSE-3 Memory Bandwidth21PB/sMore than 2,500 times higher than NVIDIA Blackwell GPU

Impact & implications

The report views Cerebras as presenting a unique investment opportunity in AI processor companies with a first-mover advantage against NVIDIA. As inference models, AI agents, and real-time AI applications continue to proliferate, the company is well-positioned to participate in one of the fastest-growing and highest-value segments of AI infrastructure spending. The OpenAI agreement serves both as a technical commercial validation and a catalyst for transforming the company from a niche hardware supplier into a scalable AI infrastructure platform. Long-term upside potential is significant; the base case forecast assumes only the 750MW base commitment and does not include potential upside from option exercises or additional customer acquisition.

Risks

• Capacity Deployment / Infrastructure Risk: Acquiring, building, and deploying sufficient cloud capacity to meet contract ramp-up represents the most significant near-term risk. Delays in manufacturing, power supply, or data center construction could postpone revenue recognition.
• Customer Concentration Risk: Business is highly concentrated over the next three years. While the take-or-pay structure provides visibility, any changes to the customer's deployment plans, timing, or future capacity choices could have a material impact on financial performance.
• Competitive Risk: Despite technological differentiation, long-term success depends on converting advantages into a broader customer base. If customers adopt competing solutions, revenue growth and utilization rates may fall short of expectations.

What to watch

• Contract Extension: The additional 1.25GW option is far larger than the existing 750MW commitment; monitoring staggered selection timing, extension decisions, and discussions beyond the base commitment capacity is crucial.
• Second Hyperscale Customer: Meaningful deployments with another hyperscaler would validate the architecture and reduce concerns regarding customer concentration.
• Capacity Deployment / Margin Inflection Point: Faster-than-expected capacity上线 would lead to accelerated revenue recognition and higher gross margins due to reduced dependence on G42 rentals.