当前股价
USD 234.71
Founded in 2015, Cerebras completely rethought semiconductor architecture by refusing to slice silicon into traditional chips. Instead, they utilize an entire silicon wafer as a single massive processor, bypassing traditional memory bandwidth bottlenecks to accelerate AI training and inference. The company successfully executed its IPO on the NASDAQ in May 2026.
Cerebras Systems designs and manufactures the world's largest and fastest AI processors (Wafer-Scale Engines) and builds AI supercomputers (CS-3 systems). The company monetizes through on-premise hardware sales, cloud inference consumption models, and multi-year mega-deployments for foundation model labs and hyperscalers.
To radically accelerate AI compute and change the trajectory of artificial intelligence by building the fastest AI accelerators in the world, unconstrained by traditional semiconductor limitations.
Decision to bypass traditional die-cutting and design a massive 'wafer-scale' processor.
Established the core architectural moat that differentiates Cerebras from Nvidia, AMD, and all other ASIC competitors.
Commercialized the world's largest chip with 1.2 trillion transistors.
Proved the viability of wafer-scale yield, cooling, and packaging, earning extreme credibility in the high-performance computing (HPC) space.
Partnered with G42 to build the Condor Galaxy supercomputers.
Provided massive baseline revenue, driving FY2024 and FY2025 top-line metrics and validating the architecture for massive LLM training.
Secured a 750 MW inference capacity MRA with OpenAI and AWS term sheets.
Shifted the company's narrative from a niche hardware maker to a tier-1 multi-cloud AI inference utility.
Went public via a $5.55B offering, structured with extreme multi-class voting rights.
Generated a massive influx of capital for R&D/CapEx while fully insulating the founding management team from shareholder activism.
Hyper-Growth & Scaling
基于历史发展轨迹的推导分析
Aggressively capture the exploding AI *inference* market by providing the lowest-latency, highest-throughput platform via cloud access, circumventing the need for customers to manage extreme on-premise cooling/power.
The strategy is high-risk, high-reward. Focusing on inference is correct, as it is projected to outgrow training 2:1. However, relying on massive upfront capital expenditures to build cloud capacity for highly concentrated clients exposes them to severe counterparty and execution risks.
The market prices Cerebras flawlessly. At over 100x trailing revenue, investors are pricing in a reality where Cerebras successfully breaks Nvidia's monopoly and commands a double-digit share of the global AI inference TAM over the next 5 years.
The current $51B+ market cap does not reflect present-day fundamentals; it entirely reflects future speculative execution. It is misaligned with the current operating losses and extreme customer concentration.
Pivot from selling purely on-premise hardware to becoming an AI cloud provider via the 750 MW OpenAI MRA and AWS term sheets. This transitions them to a consumption-based recurring revenue model.
If they successfully deploy the OpenAI capacity, revenue could easily breach $1B in FY26/FY27. However, capital expenditure will drag free cash flow deeply into the red.
《One Up On Wall Street》选股视角
Lynch avoided 'hot stocks in hot industries'. Cerebras is the epitome of this. While the tech is miraculous, the valuation requires absolute perfection. Customer concentration is a glaring hazard.
"A 100x Price-to-Sales multiple is statistically impossible to generate market-beating returns from over a 10-year horizon, regardless of the underlying technological superiority."
Do not chase at IPO premiums. Wait for lock-up expirations, a macro tech pullback, or a slight miss in quarterly guidance to compress the multiple closer to 15x-20x P/S.
If held, execute tight trailing stops. Sell immediately upon any news of TSMC wafer yield issues or delays in the OpenAI deployment timeline.
Extremely high fixed costs associated with R&D, tape-outs at TSMC, and initial capital expenditures for data centers.
Hardware engineering, compiler software, next-gen WSE development.
Sales, marketing, executive compensation, public company costs.
Payments to TSMC for silicon, packaging, and testing.
Currently dominated by large-scale hardware deployments, rapidly pivoting to recurring cloud inference contracts.
Direct sales of hardware to G42 and similar entities.
Consumption-based revenue, expected to become the dominant stream via OpenAI deal.
Systemic Consistency & Business Flywheel Analysis
The flywheel: Design unmatched hardware -> Secure massive anchor deals (G42/OpenAI) -> Use prepayments and scale to secure TSMC capacity -> Lower unit costs and fund WSE-4 -> Attract more cloud customers. The loop is currently stuck at the 'securing TSMC capacity' bottleneck.
Model Weaknesses & Vulnerabilities Analysis
Hardware is inherently a difficult business. Unlike SaaS, growth requires proportional CapEx. The requirement to use whole wafers means minor silicon defects can ruin entire blocks of compute, lowering effective yields. Furthermore, software ecosystem adoption is slow.
Midstream Fabless Designer & Systems Integrator.
Moderate to High. For workloads that specifically require extreme memory bandwidth and low latency, Cerebras can dictate price. However, they lack the ubiquitous software lock-in (like Nvidia's CUDA), slightly capping their pricing power.
Very Low. Cerebras is entirely reliant on TSMC for yielding entire 5nm wafers. TSMC holds absolute pricing power over them.
Low. Because the vast majority of their revenue comes from 2-3 anchor clients (OpenAI, G42), the buyers have significant leverage to dictate contract terms and margins.
竞争力威胁指数 (1-5)
TSMC is the only foundry in the world capable of reliably yielding whole wafers at the 5nm scale. If TSMC raises prices or allocates capacity elsewhere, Cerebras has zero immediate alternatives.
Extreme customer concentration. With two clients driving the vast majority of historical revenue, losing either would be catastrophic. The OpenAI deal mitigates this slightly but replaces it with another massive anchor.
The capital, R&D, and engineering talent required to design a wafer-scale engine and custom cooling/packaging is astronomical. The moat against startups is nearly insurmountable.
Nvidia is a near-monopoly with unlimited resources and the CUDA software moat. AMD is aggressive on price. Internal hyperscaler chips (TPUs) are cheaper at scale. Rivalry is existential.
Advanced chiplet architectures (putting many small chips together tightly) serve as a substitute for wafer-scale monolithic designs, achieving similar bandwidth with better theoretical yields.
Cerebras operates in a brutal competitive environment characterized by supplier monopolies and a Godzilla-sized rival (Nvidia). Their success entirely depends on their architectural superiority maintaining a distinct edge in inference speeds.
The company utilizes an extreme multi-class structure (Class A, B, and N). Class B shares carry 20 votes per share.
| 股东名称 | 持股比例 |
|---|---|
| Class B Holders (Founders & Early Insiders) | 99.2% (Voting Power) |
| Public Float (Class A) | 0.8% (Voting Power) |
Late Expansion (Broad Economy) / Early Boom (AI Sector)
“The global macro environment is characterized by heavy sovereign and corporate investment in AI sove...”
Co-founder. Former Corporate VP & GM at AMD; CEO of SeaMicro (sold to AMD). MBA Stanford.
Co-founder. Prior roles at AMD and SeaMicro. M.Eng and BS from MIT.
Former CFO at Sunrun, Flurry, Ticketfly. MBA from Harvard Business School.
Brilliant engineering leadership with a proven track record of hardware execution. However, the corporate governance is fundamentally hostile to public shareholders due to the multi-class voting structure, which warrants a severe governance discount.
Historically aggressive, focusing entirely on unconstrained R&D to prove out the wafer-scale concept. Post-IPO, capital allocation is shifting toward massive data center build-outs for cloud infrastructure.
Financially aligned through vast equity holdings, but the 99.2% voting lock means management answers to no one. They have total impunity to execute long-term visions, for better or worse.
Architectural Moat. No other company has successfully commercialized wafer-scale engines. The physical engineering required to deliver power to and cool an entire silicon wafer is incredibly difficult to replicate.
Cerebras holds the ultimate 'silver bullet' hardware for AI inference. If they can solve their supply chain scaling and software adoption friction, they will be a pillar of global AI infrastructure. If they stumble, the valuation will collapse.
支撑/压力位 • 均线 • 形态分析
Bearish / Consolidation. After opening significantly higher than its $185 IPO price (peaking near $386), the stock has pulled back ~25% since IPO to consolidate around the $234 level. It is currently seeking a support base.
机构持仓 • 成交量分布
Highly concentrated. With 99.2% of voting power locked up and early insiders holding massive, restricted positions, the true float is relatively thin, leading to the high Beta volatility (4.14).
多空分歧度 • 预期一致性 • 买卖建议
Market consensus is deeply polarized. Sell-side analysts are bullish on the tech narrative, while value/quant investors are heavily shorting due to the >100x P/S ratio. Advice: Avoid entering standard long positions until the first post-IPO earnings report confirms the OpenAI deployment trajectory.
机构一致评级
基于 10 位分析师的评级
价格指标排序 (低到高) (USD)
Wall Street is overwhelmingly bullish out of the IPO gate. Firms like Wedbush, Rosenblatt, and Craig-Hallum initiated coverage with aggressive Outperform ratings, citing the undeniable speed advantage of WSE-3 for AI inference and the massive TAM.
当前股价
内在价值 (Intrinsic)
安全边际
-174.5%
Standard DCF models penalize Cerebras heavily due to current negative operating margins and high WACC for an unprofitable hardware firm. The intrinsic value of $85 indicates the stock is trading on pure growth momentum, not discounted cash flows.
市盈率 (PE RATIO)
Distorted by one-time non-operating income. Functionally meaningless as the core operating base is negative.
市净率 (PB RATIO)
Extremely high, typical of asset-light software or hyper-growth tech, but unusual for heavy hardware infrastructure.
PEG 比率
N/A (Operating unprofitability invalidates traditional PEG).
EV/EBITDA
Negative EBITDA makes EV/EBITDA unusable for valuation purposes.
股息率 (DIVIDEND YIELD)
Payout: 0%。N/A
N/A
Because Cerebras operates with massive structural R&D deficits while in hyper-growth, standard earnings models fail. A Forward P/S model reflects how the venture and growth equity markets are currently pricing its potential against incumbent Nvidia. To justify its current price, investors must assign an unprecedented 60x forward multiple to optimistic $850M revenue projections.
Estimated revenue for the next 12 months based on G42 and OpenAI commitments.
Target Price-to-Sales valuation multiple (Nvidia trades at ~35x).
Total diluted shares outstanding post-IPO.
*注：以上内容为 AI 模仿知名投资人风格生成的虚拟点评，不代表本人实际观点，仅供参考。
寻找弱点、拷问逻辑，避免盲目的乐观
ROE was artificially distorted in FY25 due to a massive non-operating net income spike ($237M) despite an operating loss. Core ROE remains inherently negative due to operating unprofitability.
The $51B market cap is entirely disconnected from current trailing financials. It represents a 'scarcity premium'—Cerebras is viewed as the only credible, public, pure-play alternative to Nvidia for extreme AI inference workloads.
At -28.6% for FY25, margins improved from prior years but reflect the brutal reality of hardware design. Gross margins of 39% are solid but lag Nvidia's 70%+, highlighting a lack of equivalent pricing power and software lock-in.
Poor quality. The positive FY25 net income was driven by non-recurring financial or tax adjustments, not operational cash generation. True operational equity return is negative.
Inventory turnover is improving slightly, but prepayments to TSMC for 5nm wafers tie up significant working capital. Receivables turnover is concentrated entirely in 2-3 mega-clients.
The May 2026 IPO diluted existing private backers marginally but flooded the company with cash. The critical factor is voting capital: retail and standard institutional investors hold virtually no say in corporate governance.
Operating Cash Flow flipped positive recently due to massive prepayments/deferred revenues from G42 and OpenAI. However, massive CapEx (Investing Cash Flow) is required to deploy their CS-3 supercomputers, making sustained Free Cash Flow elusive without continuous scaling.
Post-IPO in May 2026, the asset structure radically transformed. The $5.5B cash injection ballooned total assets. The balance sheet is heavily liquid, ensuring solvency for years, though fixed assets will rapidly expand as they build out the OpenAI 750MW capacity.
Leverage is functionally zero. The trend is static as the company eschews traditional debt markets in favor of equity and strategic customer prepayments.
Cerebras is growing rapidly but still lags Nvidia's hyper-growth top-line. The most glaring difference is profitability: Nvidia generates extreme free cash flow via massive gross margins, while Cerebras operates at a loss with half the gross margin, yet trades at a significantly higher P/S multiple.