Rankings Semiconductors Cerebras
Cerebras
22
15070100
A view of the same findings, not a new score. How it works
The model found several serious problems. That does not mean the stock will fall.
Biggest deductions
- Multiples vs peers (short history)−30
- Margin−9
- Customer concentration−6
Strengths
- Net cash+3
- Raised guidance and order book+3
How the score is built
From 100 to 22, step by step: points off for the business, the price and the cracks, back for documented strengths, then three data adjustments.
Start
Score
Hover or select a step to see what drove it. Steps with findings open them below.
Key numbers
As of 2 Nov 2026. Analyst targets tend to be optimistic; read upside as a hint, not a forecast.
- Price
- 96.36 USD
- Median analyst target
- 290.00 USD
- Upside to median
- +201.0%
- Analysts
- 11
- Target range
- 175.00 – 325.00 USD
- 52-week change
- –
- P/E (trailing)
- n/a
- Forward P/E
- n/a
- PEG
- n/a
Findings
Every deduction and every plus, with its source. Open a step to read them.
Business
Is the company getting stronger, not just bigger? · 4 findings−18 of −30
- −9MarginGAAP operating margin is −81% TTM (operating loss −$552m on revenue of $681m) and −265% in Q2. GAAP gross margin fell to 14.2% in Q2 from 44.6% in Q1; even core gross margin is only 40.6%, and full-year guidance for core operating margin is −17 to −19%. FCF is −$680m TTM because capex has risen to $747m. The loss counts here and not again under ROIC (−24%).
- −4Quality of growthGAAP revenue in Q2 was $180.1m (+74%), below the expectation of $194m and lower than Q1 ($193.4m). Hardware sales halved from $110.6m to $54.1m in a quarter; growth now comes from its own cloud operations ($126m), which require Cerebras to build and lease data centres itself (600 MW under contract). The business thus becomes more capital-heavy, not stronger.
- −3Earnings quality and share-based payShare-based pay was $386.6m in the first half of 2026 vs $22.4m the year before (of which $273.6m RSU expense triggered by the IPO). Management's 'core' figures exclude both the pay and the amortization of customer warrants; the difference between core net result (−$6.9m) and GAAP (−$450.5m) in Q2 is $444m.
- −2Moat and pricing powerThe wafer-scale chip gives a real speed advantage in inference, but the core gross margin of approx. 40% (guidance 38–40% in Q3) vs Nvidia's approx. 75% and AMD's 56% shows weak pricing power. The software ecosystem is far smaller than CUDA, and Nvidia has licensed Groq's inference technology since December 2025.
Price
Today's multiples against the company's own history and its peers. · 1 finding−30 of −35
- −30Multiples vs peers (short history)Market cap $41.5bn and EV $35.1bn for a loss-making company: P/S 60 and EV/Sales 51 TTM, forward P/S 28. For comparison: Nvidia P/S 18.9 (forward 10.7), AMD 25.4 (15.5), Broadcom 20.1 (11.4) and CoreWeave EV/Sales 12.6 - all profitable or at far larger scale. At the quarter-end in June P/S was 87. Even on analysts' 2027 estimate ($2.95bn, almost a tripling) EV/Sales is about 12, and that estimate rests on the OpenAI ramp.
Cracks
Cash flow, customers, debt, insiders, guidance and what sits behind the headline numbers. · 8 findings−30 of −35
- −6Customer concentrationCustomer A accounted for 34% of revenue in Q2 and 49% in the first half, customer B for 32% (Q2) and customer C for 10%; three customers = 76% of Q2. A and D are related parties (UAE: MBZUAI and G42, which accounted for 86% in 2025), and OpenAI revenue was $56.8m in Q2 (about 32%). Two customers make up 76% of receivables.
- −5OpenAI counterparty riskOpenAI is at once the largest growth customer (deal of over $20bn for 750 MW through 2028, most of RPO of $25.4bn), lender (loan of $1,004.6m, $918m outstanding, due immediately and where OpenAI can take over the account if the agreement ends) and future shareholder (warrants for up to 33.4m shares at $0.00001, approx. 14% of the shares; 10m vested). Revenue is in practice partly bought with shares.
- −4CompetitionAccording to SemiAnalysis, OpenAI's new GPT-6.1 Sol Ultrafast (DevDay 29/9) runs on Nvidia GPUs and not on Cerebras, which powered the previous Ultrafast version; the stock fell about 20% in the week. Nvidia (with the Groq licence), AMD and the hyperscalers' own chips attack precisely fast inference, which is Cerebras' core.
- −5Lock-up, dilution and insider salesThe share count has risen 61% over a year. The lock-up is released in tranches: up to 14.6m shares every other week since mid-August, 19.4m on 30/9 and the main expiry around November. The same day COO Dhiraj Mallick filed Form 144 for the sale of 396,000 shares ($77.9m) and CFO Robert Komin 32,500 shares ($6.4m). The stock trades below the IPO price of $185.
- −3Governance (A/B shares)B shares have 20 votes and A shares 1 (N shares 0). As of 5/8 2026 there were 112.2m A and 111.6m B shares, so the B shareholders (the founders Feldman, Lie, Fricker, James and early investors) have about 95% of the votes; ordinary shareholders effectively have no influence.
- −4Execution and supplier dependenceAll wafers are produced by TSMC, and according to the prospectus a change of foundry would require considerable development and delay products. At the same time 600 MW of data centre capacity must be built out before the end of 2027; in Q2, leasing of third-party facilities to OpenAI pushed the gross margin down, and capex was $549m in half a year.
- −2Geopolitics and CFIUSThe IPO was delayed for over a year by CFIUS' review of G42's ownership stake (former Huawei ties); the solution was non-voting shares and a security agreement. The largest customer is still UAE-based and depends on US export licenses for advanced AI chips.
- −1Price and sentimentThe price ($174.67) is about 55% below the peak of $386 and 50% below the opening price; 11.6% of the free float is sold short.
Plus
Only documented strengths. · 2 findings+6 of +15
- +3Net cashNet cash $6.43bn ($7.92bn in cash and investments vs $1.49bn debt incl. the OpenAI loan), approx. 15% of the market cap; current ratio 5.8. The IPO raised $5.55bn, so the buildout is funded for several years.
- +3Raised guidance and order bookFull-year guidance for core revenue raised to $880–890m (+73–75%) and all core targets beaten in Q2; core cloud revenue +287%. Remaining performance obligations (RPO) are $25.4bn, about 37 times TTM revenue.
Caveats
What the score does not capture, or where the data is weak.
Listed 14 May 2026 (IPO price $185, opened at $350), so there is no own multiple history; T2 is assessed against peers and the absolute level. Loss-making: P/E and PEG are meaningless. The company steers by 'core' figures that exclude share-based pay ($377m in Q2, of which $273.6m is a one-off IPO expense) and amortisation of customer warrants ($44m in Q2, booked as a deduction from revenue). The 10-Q does not name customers A–D; that A and D are G42/MBZUAI is based on the prospectus's related-party disclosures. SemiAnalysis' claim that OpenAI runs GPT-6.1 Sol Ultrafast on Nvidia is unconfirmed, and Sam Altman called Cerebras a 'close partner' on 3/10. The exact lock-up schedule (main expiry about 9 to mid-November 2026), unrealised share-based pay, actually executed insider sales (Form 4) and the terms of G42's security agreement with CFIUS have not been examined. Price $174.67 on 7/10 2026.
Also in Semiconductors
The highest scores in the same industry.
