Cerebras Systems: What the Price Is Actually Asking You to Believe NASDAQ: CBRS · AI Inference Infrastructure
Cerebras is an AI inference infrastructure company backed by a genuinely hard piece of evidence: $25.4 billion of remaining performance obligations. But run the price backwards and it demands two things at once — roughly 3x revenue growth next year and a mature operating margin of 35–40% — while the company's own guidance calls for a −19% operating margin and sequential revenue growth of 2–3% next quarter. Meanwhile 66% of revenue comes from two customers, one of which lent the company $1 billion and walked away with warrants on 33.4 million shares struck at $0.00001. The bull case is already in the price. The 32% dilution is not.
I.Data Verification Log
Every figure in this report was checked for source, as-of date, and cross-source agreement before use.
| Item | Value | Source | As of | Grade |
|---|---|---|---|---|
| Share price | $191.93 | Yahoo ↔ StockAnalysis, two sources agree (0.00% variance) | 2026-09-11 | 🟢 |
| Previous close | $191.34 (+0.31%) | Yahoo | 2026-09-11 | 🟢 |
| Market cap / shares out | $45.60B / 237.56M | StockAnalysis | 2026-09-11 | 🟢 |
| TTM revenue | $680.67M (+111.7%) | StockAnalysis | 2026-06-30 | 🟢 |
| TTM net loss / EPS | −$487.24M / −$6.33 | StockAnalysis | 2026-06-30 | 🟢 |
| Cash + ST investments / total debt | $7.92B / $1.487B | Quarterly balance sheet | 2026-06-30 | 🟢 |
| Remaining performance obligations | $25.4B | SEC Form 10-Q | 2026-06-30 | 🟢 |
| Customer warrant asset | $1.1B | SEC Form 10-Q | 2026-06-30 | 🟢 |
| Antidilutive securities excluded | 76.01M shares | SEC Form 10-Q | 2026-06-30 | 🟢 |
| Consensus price target | $291.64 | 11 analysts (high $330 / low $209) | 2026-09-11 | 🟡 |
Gate passed. Two independent sources matched the price to the cent. The company listed on May 14, 2026 at an IPO price of $185 and closed its first day at $311.07. The current price is +3.7% above the IPO price and −38.3% below the first-day close. Four months after listing, essentially the entire debut premium has evaporated.
The market regime reads neutral with a downward tilt. August CPI at 3.4% y/y has futures pricing roughly a 90% probability of a hike at the next FOMC; the US 10-year sits at 4.97%, Brent at $104.6 (+8.7% on the week on Hormuz risk), and the Nasdaq fell 0.66% on the week. This is the least forgiving discount-rate environment possible for a long-duration, pre-profit growth name — and that is the premise this entire analysis sits on.
II.Financial Teardown: The Two-Income-Statement Problem
1. What the company shows you versus what accounting recorded
In the second quarter of 2026, this company effectively reported two different sets of results.
| Q2 2026 | GAAP | "Core" (company-adjusted) | Gap |
|---|---|---|---|
| Revenue | $180.1M | $209.9M | +$29.8M |
| Gross margin | 14.5% | 41.0% | +26.5pp |
| Operating margin | −265% | −16% | +249pp |
| Net income | −$450.5M | — | — |
The reconciliation, per the 10-Q:
- Less pass-through revenue: −$14.5M
- Add back customer warrant amortization: +$44.3M
That second line is the crux. To land anchor customers, Cerebras issued penny warrants struck at $0.00001, and under revenue recognition rules those are booked as contra-revenue. The warrant asset ballooned from $152.4M at the end of 2025 to $1.1 billion by June 30, 2026 — roughly a sevenfold increase in six months — and it will keep reducing reported revenue every quarter through October 2031.
Warrants are not a cash outlay, but they are a real transfer of shareholder ownership. "Core revenue" adds that cost back to inflate the top line while leaving the corresponding dilution out of the share count. It inflates the numerator and freezes the denominator. Feed it straight into a valuation and you are being generous twice.
2. The actual slope of the growth curve
The narrative the market hears is "revenue +103% year over year." The sequential picture tells a different story.
| Quarter | GAAP revenue | QoQ | Core revenue | QoQ |
|---|---|---|---|---|
| Q1 2026 | $193.4M | — | not disclosed | — |
| Q2 2026 | $180.1M | −6.9% | $209.9M | — |
| Q3 2026 guidance | — | — | $214–216M | +2.0 to +2.9% |
The company's own guidance calls for 2–3% sequential growth next quarter. Consensus, meanwhile, models FY2027 revenue of $2.95B against an FY2026 estimate of $887M — +233%.
Back out the full-year core guidance ($880–890M) against reported results plus the Q3 guide and the fourth quarter alone has to deliver roughly $260M, a 21% sequential jump inference · medium confidence. The entire hockey stick sits in quarters that have not happened yet.
3. The scale of the cash burn
| Q2 2026 cash flow | Amount |
|---|---|
| Net loss | −$450.5M |
| Stock-based compensation | +$377.0M |
| Depreciation & amortization | +$24.4M |
| Working capital change | −$100.6M |
| Operating cash flow | −$59.8M |
| Capital expenditure | −$416.9M |
| Free cash flow | −$476.7M |
| Financing inflow | +$5,815M |
Of the $377M in stock-based compensation, $273.6M was a one-time acceleration triggered by the IPO satisfying a liquidity vesting condition. Strip that out and run-rate SBC is roughly $103M per quarter — 57% of quarterly GAAP revenue. Abnormally high even after the one-off is removed.
Capex of $416.9M is 2.3x revenue. That is forward investment for volume that has not yet converted to revenue — and it creates the structural fragility of building first and collecting later.
4. Revenue quality: three customers and a circular structure
| Customer | Q2 2026 | H1 2026 | Note |
|---|---|---|---|
| Customer A = MBZUAI (UAE) | 34% | 49% | Related party |
| Customer B = OpenAI (inferred) | 32% | 20% | Ramping fast |
| Customer C | 10% | <10% | — |
| Customer D | <10% | 10% | Related party |
- The UAE cluster (MBZUAI + G42) accounted for 86% of 2025 revenue. In Q1 2026 it was still MBZUAI 63% + G42 11% = 74%.
- The claim that "the G42 concentration problem has been solved" is not accurate. The concentration simply moved from G42 to MBZUAI — the same Abu Dhabi ecosystem.
- Even in Q2 2026, the top two customers are 66% of revenue.
And OpenAI is not merely a customer
- A Master Relationship Agreement signed in December 2025 for 750MW of compute capacity
- A $1.0 billion secured promissory note (working capital loan) funded to Cerebras in January 2026 — the customer is directly financing its supplier's capex
- A warrant over up to 33.4 million Class N shares at an exercise price of $0.00001 — equal to 14.1% of shares outstanding
The customer provides the funding, the funded equipment generates revenue from that same customer, and the customer receives what is effectively free equity in return. The revenue is real. The independence of the cash and the pricing power behind it are not.
5. The backlog: the one genuinely strong positive
RPO of $25.4 billion as of June 30, 2026 — roughly 28x current annual revenue.
| Recognition window | Share | Amount |
|---|---|---|
| Through June 2028 (first 24 months) | 22% | ~$5.6B |
| July 2028 – June 2030 (months 25–48) | 43% | ~$10.9B |
| Thereafter | 35% | ~$8.9B |
That first $5.6B over two years works out to $2.8B a year — consistent with the $2.95B FY2027 consensus. So the consensus is not invented; it rests on the assumption that the RPO converts on contract terms. The AWS volume is not yet in the RPO at all, with first revenue expected mid-2027.
Every judgment about this stock collapses into a single question: does the RPO convert to revenue on schedule?
III.Lens A — The Buffett Filter
| Filter | Result | Reasoning |
|---|---|---|
| Circle of competence | Fail | Whether wafer-scale architecture is still the right technical path in ten years is not knowable. Buffett bought Apple as a consumer brand, not as a semiconductor company |
| Moat / pricing power | Fail | A company that gives away $1.1 billion of equity to secure customers and borrows working capital from one of them does not have pricing power. Gross margin went 42.3% (FY24) → 39.0% (FY25) → 41–43% (FY26 guide) — flat, not expanding |
| Earnings predictability | Fail | Three consecutive years of operating losses. FY25 net income of +$87.9M arrived alongside a −$145.9M operating loss, driven by non-operating items — accounting profit, not business profit |
| Owner earnings | Fail | Operating cash flow of −$59.8M and no way to separate maintenance from growth capex. There are no owner earnings to capitalize |
| ROE / ROIC | Fail | ROIC of −24.0%. Buffett's threshold is 15%+ |
| Balance sheet | Pass | $6.43B net cash. But $1B of the debt is owed to a customer (OpenAI) — not an independent source of funding |
| Management alignment | Fail | $377M of SBC in one quarter and share count up 61.2% year over year. A Class A/B/N structure leaves ordinary shareholders without control |
| Price | Fail | 67x sales, 57.5x EV/Sales |
Seven of eight filters fail — straight into the "too hard" pile. It is the same pattern he learned from IBM (technological change outrunning the moat), with the added wrinkle of handing shareholder equity to customers.
In fairness, this should be recorded too: Buffett's filter is designed to exclude early-stage, high-growth infrastructure companies systematically. He missed Amazon. A Buffett failure is not a statement that the stock will fall — it is a statement that the stock does not offer the degree of certainty Buffett requires.
IV.Lens B — The Damodaran Approach
Unlike Buffett, Damodaran computes a value. Following his method, the story comes first and is then translated into numbers.
1. Three stories
- Bear — the OpenAI ramp slips, NVIDIA's inference roadmap and hyperscaler custom silicon compress pricing, specialized silicon commoditizes, and UAE demand proves politically contingent. → terminal operating margin 10%
- Base — OpenAI volume arrives roughly a year later than consensus assumes, AWS contributes from 2028, and Cerebras captures real benefit from the inference shift but settles at margins well below the NVIDIA franchise. → terminal operating margin 19%
- Bull — the RPO converts on contract terms with no slippage, and latency advantage becomes a structural standard in inference workloads, compounding share. → terminal operating margin 27%
2. Results
| Scenario | FY27 rev | FY31 rev | FY36 rev | Enterprise value | Value per share | vs price |
|---|---|---|---|---|---|---|
| Bear (30%) | $1.50B | $3.23B | $4.40B | $0.1B | $20.75 | −89.2% |
| Base (50%) | $2.20B | $7.64B | $12.03B | $10.4B | $53.80 | −72.0% |
| Bull (20%) | $2.95B | $12.68B | $20.32B | $36.0B | $135.26 | −29.5% |
| Probability-weighted | — | — | — | — | $60.18 | −68.6% |
On basic shares (237.56M) the base case is $71.01 and the bull case is $178.53.
Ignore dilution entirely and believe the bull case outright, and you get $178.53 — essentially today's price of $191.93. In other words, the current price already reflects the bull case in full, and does not reflect the 32% dilution at all.
3. Reverse DCF: what the price requires
Damodaran's favorite move. "What must be true for $191.93 to be correct?"
| FY2027 revenue | Required terminal operating margin | Implied price |
|---|---|---|
| $2.48B (+180%) | 40% | $173 |
| $2.95B (consensus, +233%) | 35% | $176 |
| $2.95B (consensus, +233%) | 40% | $202 |
| $3.36B (+280%) | 35% | $198 |
| $3.81B (+330%) | 30% | $188 |
The price demands two things simultaneously:
- FY2027 revenue at or above consensus ($2.95B) — a jump from 2–3% sequential growth to roughly 4x annually
- A mature operating margin of 35–40% — the very top of the semiconductor industry inference
The company's own FY2026 core operating margin guidance is −19% to −17%, and the Q3 guide of −25% to −23% is worse, not better. On a GAAP basis, Q2's operating margin was −265%.
4. Attacking my own conclusion
The four strongest arguments that this analysis is wrong:
- The $25.4B RPO is real. It is a contract, not a story. The base case here (FY27 at $2.20B) is 21% more conservative than the RPO's own recognition schedule of ~$2.8B per year. If the contracts perform as written, the base case is wrong.
- DCF structurally undervalues high-growth companies. In the base case most of enterprise value sits in the terminal value while early free cash flow is deeply negative, so discounting punishes hard. Apply this model to NVIDIA in 2016 and it would likewise have printed "85% overvalued."
- Front-loaded capex may be evidence, not cost. $417M of quarterly investment is the physical preparation for 750MW of delivery — arguably proof the ramp is real.
- Warrant issuance may not be a weakness. It is a standard playbook for an early platform company securing anchor customers, and if lock-in follows it becomes a cheap customer acquisition cost in hindsight.
Arguments 1 and 3 hold. But they defend the revenue path, not the margin path. What the reverse DCF demands is not revenue alone but a 35–40% operating margin — and under 41–43% gross margin guidance that requires compressing operating expenses to 3–8% of revenue. For a company whose run-rate stock compensation alone consumes 57% of revenue, that is close to arithmetically impossible. Argument 2 is methodologically sound, so the bull weighting stays at 20% while the bear is capped at 30%.
Adjusted probability: 62% that the stock trades below today's price in six months.
V.Technical and Flow Diagnostics
| Indicator | Value | Reading |
|---|---|---|
| Drawdown from 52-week high | −38.3% | Downtrend in progress |
| Position in 52-week range | 16.4% | Lower band |
| 50-day moving average | $201.79 | Broken — short-term trend negative |
| 200-day moving average | not computable | Only 83 sessions since listing — trend cannot be judged |
| RSI (14) | 47.9 | Neutral, not oversold |
| Annualized volatility (3M) | 122.7% | Extreme |
| ATR (14) | 6.27% / day | A 6% daily move is normal here |
| Distribution days (last 25) | 4 | Institutional selling pressure present |
| Float | 110.11M / 237.56M | 46% — more than half still locked |
| Short interest | 15.84M shares (6.67%) | Meaningful but below squeeze threshold |
| Share count change YoY | +61.2% | Persistent dilution |
Trend filter: unavailable. A stock four months post-listing has neither 12-month momentum nor a 10-month moving average. There is no basis for a trend-following entry at all.
This is a staged release, not a single unlock date. Directors, officers and pre-IPO investors became eligible to sell 15% after the Q1 report; more unlocked after Q2 earnings; additional fixed release dates fall in August, September and October, before the 180-day expiry in mid-November. The window from now through November is a supply window. The CEO's stake alone is worth roughly $1.9 billion at the offer price, the CTO's about $1 billion.
VI.Price Target Framework
1. Intrinsic value
| Scenario | Fully diluted | Basic shares |
|---|---|---|
| Bear | $20.75 | $27.40 |
| Base | $53.80 | $71.01 |
| Bull | $135.26 | $178.53 |
| Probability-weighted | $60.18 | $79.40 |
2. Twelve-month trading target (multiple-based)
Assumes FY2027 revenue of $2.20B (base case) and $4.5B of net cash after burn. This reflects the reality that markets do not converge to intrinsic value on command.
| EV/Sales applied | Implied price | Premise |
|---|---|---|
| 8x | $84.60 | Growth narrative damaged, treated as an ordinary semiconductor name |
| 12x | $98.50 | High growth intact but premium compresses |
| 18x | $164.00 | Current premium sustained |
12-month target band $85 – $165 · midpoint $120
3. Entry ladder — at what price does this become interesting?
no basis to enter
small starter only
serious consideration
margin of safety
4. Why this differs from consensus
The consensus target across 11 analysts is $291.64 with a Strong Buy rating, and even the lowest target of $209 sits above the current price. That is roughly 5x the $60.18 probability-weighted value derived here.
Decomposing the gap honestly:
- Consensus is likely modeled on core revenue and core margins — the figures with warrant costs added back
- The consensus FY2027 EPS of +$1.27 is difficult to reconcile on a GAAP basis (Q2 2026 GAAP EPS was −$2.98)
- Much of the coverage is plausibly from IPO underwriters; initiation four months after listing carries a structural optimism bias inference
- Whether the 76.01M antidilutive shares are reflected could not be verified
A gap this wide is itself a risk signal. This analysis may be wrong, or consensus may be wrong. Either way, the valuation uncertainty around this name sits far outside a normal range.
VII.Strategy by Time Horizon
Three-month expected band from the volatility model (directionless dispersion, 117% annualized vol)
- 1σ: $137 – $269
- 2σ: $98 – $378
- Probability of a +5% move 44.3%, of a −5% move 44.0% — no directional edge
A ±6% daily move is normal for this stock. A ±40% move within three months is statistically unremarkable.
What governs this window is supply, not fundamentals. Fixed lock-up releases are running through September and October, and the 180-day expiry in mid-November takes the float well above 46%. The Q3 report expected in November already carries guidance for deteriorating margins (−25% to −23%), with revenue of $214–216M representing only 2–3% sequential growth.
Upside catalysts (formalized OpenAI delivery milestones, an early AWS contract) and downside catalysts (lock-up supply, margin deterioration) are queued simultaneously, and there is no basis for predicting which lands first. A short-term position here is a coin flip dressed as analysis.
Four indicators — and only four — are worth tracking
| # | Indicator | Bull confirmation | Bear confirmation |
|---|---|---|---|
| 1 | Core revenue QoQ | Two consecutive quarters above +15% | Below +5% persistently |
| 2 | RPO balance | Rising from $25.4B | Falling = cancellations or renegotiation |
| 3 | Core operating margin | Inside −10% by Q4 | Stuck at −25% |
| 4 | Customer concentration | Top two below 50% combined | Holding or rising from 66% |
Indicator 1 is the single most important variable in this investment. A 15% sequential rate annualizes to 75% — the minimum required to hit the FY2027 consensus. Current guidance is 2–3%.
Medium-term target: $85–165, midpoint $120. If indicators 1 and 3 clear their bull thresholds, re-rate toward the top of the band or above; if they hit the bear thresholds, below the bottom.
The long-term case reduces to one question: is inference-specific silicon a durable category, or a transitional one that NVIDIA's roadmap and hyperscaler in-house chips eventually absorb?
No data available today can settle that question. And that is precisely the situation Buffett's circle-of-competence principle describes.
| Long-term scenario | Value today | Value in 5 years |
|---|---|---|
| Category survives, share secured (bull) | $135 | $200–250 |
| Niche survival (base) | $53.80 | $85–95 |
| Absorbed or commoditized (bear) | $20.75 | $25–35, or acquired |
Justifying a long-term hold requires more than 60% confidence in the bull scenario — and no evidence available today supports that level of confidence.
VIII.Risk Register
Critical Thesis-breaking if triggered
| # | Risk | Transmission path | Impact |
|---|---|---|---|
| R1 | Circular financing reverses | OpenAI hits a funding constraint and the $1B loan terms change or the MRA is renegotiated. A "substantial portion" of the $25.4B RPO is the OpenAI contract | 60%+ of revenue at immediate risk; −50% or worse in the stock |
| R2 | Concentration plus geopolitics | MBZUAI 34% + G42 = a single UAE ecosystem. Changes to US AI chip export rules for the Middle East, renewed CFIUS scrutiny, or Gulf instability | ~40% of revenue at immediate risk |
| R3 | Stranded forward capex | $417M per quarter becomes idle capacity if 750MW delivery slips. Depreciation continues; revenue does not arrive | Widening losses plus impairment |
High
| # | Risk | Detail |
|---|---|---|
| R4 | Dilution cascade | 76.01M antidilutive shares (+32%), OpenAI penny warrants on 33.4M shares (14.1%), share count up 61.2% YoY. The structure means SBC cost rises as the stock rises |
| R5 | Lock-up supply | Float expands from 46% through November. Founder holdings alone are worth roughly $2.9 billion. Fixed release dates are already running |
| R6 | Margin convergence failure | Reverse DCF requires 35–40%; guidance says −19%. A gap of more than 50 percentage points |
| R7 | Reporting complexity | Dual core/GAAP metrics, a $1.1B warrant asset amortizing over five years, two related-party customers. High interpretation difficulty and ample room for surprises |
Moderate
| # | Risk | Detail |
|---|---|---|
| R8 | Macro headwind | US 10-year at 4.97%, ~90% odds of a hike. A direct hit to the discount rate on long-duration, pre-profit equity |
| R9 | Competitive pressure | NVIDIA's inference roadmap, peer specialized silicon (Groq, SambaNova), hyperscaler in-house chips |
| R10 | Extreme volatility | 122.7% annualized, 6.27% daily ATR. Setting a sane stop loss is physically difficult |
| R11 | Governance | Class A/B/N dual-class structure leaves ordinary shareholders no mechanism to check management |
| R12 | No trend signal | 83 sessions of history. Without a 200-day average or 12-month momentum, technical safeguards cannot function |
Upside risk What would flip the call
- Early recognition of, or an increase in, the $25.4B RPO
- The AWS contract entering the RPO — currently excluded, so pure upside
- Inference demand outrunning capacity, restoring pricing power
- Coding and security segments scaling enough to dissolve concentration organically
IX.Conclusion
WATCH · 0% new position
Probability of trading below today's price within six months: 62%
Reasoning
- At $191.93 the price fully reflects the bull case but does not reflect 32% dilution (bull case on basic shares is $178.53 ≈ today's price)
- The reverse DCF requires consensus FY27 revenue and a mature operating margin of 35–40% together. Company guidance is −19% to −17%
- The top two customers are 66% of revenue, and one of them is simultaneously a lender and a penny-warrant holder
- Probability-weighted value of $60.18 against a price of $191.93 — −68.6%
- But the $25.4B RPO is real, and if it converts the base case is wrong — which is why this is a watch, not a sell or a short
Kill conditions
Any one of these invalidates the verdict above.
| Direction | Condition |
|---|---|
| Upgrade | ① Core revenue above +15% QoQ for two consecutive quarters ② OpenAI 750MW delivery milestones reconfirmed in specific terms in a quarterly filing ③ A close back above $250 ④ The AWS contract disclosed as entering the RPO |
| Downgrade | ① RPO balance declines ② Core operating margin stuck at −25% ③ Top-two customer concentration above 70% ④ A close below $150 |
X.What Could Not Be Verified
Recorded honestly. The following items were either unobtainable or carry low confidence.
- Q1 2026 core revenue — not disclosed, which limits the core-basis sequential trend to two quarters
- Peer multiples — current EV/Sales and operating margins for NVIDIA, Broadcom and AMD were not verified in this session. The claim that "35–40% is top-tier for semiconductors" is an inference
- Identity and underwriter affiliation of the 11 analysts — the basis for the consensus-bias argument was not confirmed
- Confirmed identity of Customer B — inferred as OpenAI (20% in H1 rising to 32% in Q2, matching the MRA timing) but not named in the 10-Q
- 52-week range discrepancy — $168.52–$311.07 on a closing basis versus $160.81–$386.34 intraday. Interpreted as a methodology difference but not confirmed
- RPO cancellation provisions — whether contracts are cancellable and on what terms is the central variable in this analysis, and it could not be confirmed
- Some macro inputs — credit spreads and real yields were carried over from a prior record rather than independently re-verified
XI.Sources
- Cerebras Systems Inc., Form 10-Q (period ending 2026-06-30) — SEC EDGAR
- Cerebras Systems Inc., Form 10-Q (period ending 2026-03-31) — SEC EDGAR
- Cerebras Q2 2026 results — company investor relations
- CBRS Form 8-K, non-GAAP reconciliation — StockTitan
- CBRS financials, statistics and consensus — StockAnalysis
- Cerebras (CBRS) starts trading on Nasdaq after IPO — CNBC
- Cerebras leans on OpenAI and AWS as anchor customers — DigiTimes
- Cerebras: IPO lockup comes fast — Seeking Alpha
- Cerebras Q2 2026 earnings: stock falls after guidance raise — Quartz
- Price data: Yahoo Finance chart API (2026-09-11 close, cross-verified)
Disclaimer
This document is research based on public sources and is not investment advice. The author is not a licensed investment adviser and has no knowledge of any reader's assets, income, time horizon or risk tolerance. The price targets and probabilities above are the output of calculations under stated assumptions; change the assumptions and the output changes.
In particular, the roughly 5x gap between the $60 probability-weighted value derived here and the $292 market consensus indicates a fundamental disagreement among informed observers about what this company is worth. Which side is correct is only knowable in hindsight.
All investment decisions and their consequences rest with the individual investor. Price movements and disclosures after the analysis date are not reflected.
'For Kimchi Crusaders~!' 카테고리의 다른 글
| SPCX : Investment Analysis Report (20260613) (0) | 2026.06.14 |
|---|