2026-11-14 · 2026-11 / week-2
CBRS’s $26.1B RPO still needs delivery, margin and cash conversion
CBRS’s $26.1B RPO still needs delivery, margin and cash conversion
The Mispricing Desk | 14 November 2026, 12:42 SGT | U.S. long | Reject / no-trade screen
Summary: Cerebras reported $210.8 million of Q3 GAAP revenue and $230.9 million of issuer-defined Core Revenue, up 70%. It also reported a $33.1 million core operating loss, with a $26.1 billion remaining-performance-obligation balance tied in significant part to OpenAI and including estimates of pass-through costs. At the Nov. 13 close of $108.72, an issued-share equity-value proxy is about $26.4 billion, or 27.8 times the FY26 Core Revenue midpoint. A six-month sensitivity has a flat base and a -21.7% adverse case. Reject / no trade.
Opportunity Ranking
| Rank | Candidate | Direction | Mispricing | Evidence Freshness | Catalyst Window | Positioning | 10/5 Status | Tradeability | Main Rejection Risk |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Cerebras Systems (CBRS) | Long screen | $26.1B RPO is not profit; the $108.72 close equals 27.8× FY26 Core Revenue, while OpenAI supplied 34% of Q3 revenue | Nov. 12 SEC release and 10-Q; Nov. 13 regular close | Q4/FY26 results; Microsoft service milestones and OpenAI capacity delivery, dates unverified | Nov. 13 volume was 2.1× Nov. 12; current holder, short, options and fund-flow data unknown | Reject; reference base 0%, adverse -21.7% | Close verified; live spread/depth/exit data absent | Core operating loss, customer concentration, delivery delay, OpenAI loan and warrants |
| 2 | Liftoff Mobile (LFTO) | Long screen | Q3 revenue rose 32%, adjusted EBITDA margin was 61%, and FY26 revenue outlook rose; its Nov. 13 close was $24.44, up 30.6% | Nov. 12 SEC release and 10-Q; Nov. 13 close | Q4/FY26 reporting; Core Advertising Revenue reporting ends separately in Q1 2027 | Current positioning unknown | Not superior; 30.6% result-day rally and pre-result-close reversion risk | Liquid close reference; execution not audited | About $1.4B debt principal and no 5% adverse bound established |
| 3 | Jones Soda (JSDA) | Long screen | Q3 revenue rose 205% to $13.7M and net income turned positive; the Nov. 13 OTC close was $0.2373, but nine-month operating cash flow remained negative and shares rose | Nov. 10 SEC 10-Q and issuer release; Nov. 13 OTC close | Next quarterly report; date unverified | Ownership and current positioning unknown | Reject as a low-liquidity, underwritten-equity case; no bounded map | OTC close reference, 20.7K shares reported; exit quality unknown | $2.3M nine-month operating cash use, new shares, options and warrants |
| 4 | SPDR S&P 500 ETF (SPY) | Long control | Broad-market reference only; no company-specific disagreement | Nov. 13 regular close: $754.37, down 0.36% | No single-issuer catalyst | Index-level positioning | No idiosyncratic 10/5 setup | Liquid broad-market instrument; live execution not audited | Market exposure is not an issuer thesis |
Selected screen candidate: CBRS, for information value. It is not the best executable opportunity.
Why this one now: Cerebras’s Nov. 12 filing adds a current bridge from contract backlog to revenue, financing, customer warrants and capacity delivery. The Nov. 13 close then shows how the market responded to that evidence.
What should surprise the reader: The $26.1 billion RPO is numerically close to the share-value proxy, but it is not a backlog-to-market-cap valuation. The company expects about one-third of it to be recognized over the first 24 months, and the filing says it includes estimated variable consideration and pass-through costs.
Why This Is the Best Opportunity Right Now
This is the strongest current research screen because the same customer relationship links several parts of the underwriting: OpenAI is a major source of RPO, supplied a working-capital loan, and received a warrant. Capacity delivery determines service revenue and may also determine loan repayment and warrant vesting. A backlog headline alone hides that structure.
The boundary is new. A May Desk article mentioned CBRS only as a rejected IPO-float comparator. This screen uses the filed Q3 operating results, the OpenAI loan and warrant balances, the RPO schedule, newly disclosed Microsoft milestones and the first full regular-session response. It does not reuse the earlier IPO-float thesis.
Why This Can Move More Than 5% Soon
The Nov. 13 regular session itself moved 11.2% from the prior close. The next meaningful test is whether Q4 Core Revenue reaches the $310–$330 million guide and whether Core operating margin approaches the guided -1% to +1%. The company has not published a confirmed next-results date in the sources reviewed.
The Microsoft agreement includes milestones that must be met before cloud-inference services begin. Its price and revenue start date were not disclosed in the 10-Q. The OpenAI MRA is already in delivery, but Cerebras said September capacity delivered was below the scheduled quantity before accounting for potentially excused delays; it expected delays to continue through part of 2027. These facts create a path to moves above or below 5%, but they do not bound the downside for a long.
10/5 Asymmetry Gate
Reject / no trade. The highest-probability base case is flat from the $108.72 regular close, below the required +10%. The bottom-case sensitivity is -21.7%; the favorable-base/adverse ratio is 0:1. The weighted reference value is -4.6% before costs. A more optimistic top case cannot repair a flat base or the unbounded downside. No current spread, depth, venue quality, volume quality or exit capacity was verified. I do not have sufficient reliable data to quantify slippage and exit costs accurately.
The Setup
Cerebras reported $210.8 million in Q3 GAAP revenue, up 55% year over year. Hardware revenue was $68.9 million, down from $96.8 million a year earlier; cloud and services revenue rose to $141.9 million from $38.9 million. Management’s non-GAAP Core Revenue was $230.9 million, up 70%. The company defines this metric to exclude non-cash customer-warrant amortization, stock compensation, data-center pass-through revenue and costs, and other items. Core operating loss was still $33.1 million, or about 14% of Core Revenue. GAAP operating loss was $224.1 million.
The company guided Q4 Core Revenue to $310–$330 million and Core operating margin to -1% to +1%. That is a sharp projected sequential improvement, not a reported outcome. Full-year Core Revenue guidance is $940–$960 million. The release did not provide a forward GAAP reconciliation for adjusted EBITDA or margins because stock compensation and other reconciling items may vary.
The Market Price
StockAnalysis reports a Nov. 13 regular close of $108.72, up 11.22%, with a $101.18–$120.44 range and 36.87 million shares traded. The previous regular close was $97.75. Its page also displayed a $106.00 after-hours snapshot at 7:59 p.m. EST. That mark is reference context only; the scenario map uses the regular close. No executable quote or market-structure data were verified.
The Q3 10-Q reports 243.003 million combined Class A, B and N shares outstanding as of Nov. 5. Multiplying that dated count by $108.72 gives a $26.42 billion equity-value proxy. It is not a fully diluted market capitalization: the OpenAI warrant permits up to 33.445 million Class N shares; 10.034 million had been exercised by Sept. 30, and further vesting depends on MRA milestones. The calculation is an analyst estimate from a dated share count and a later close, not the issuer’s market-cap figure.
At the $950 million Core Revenue midpoint, the proxy is 27.8× equity value to Core Revenue. This is not EV/sales. It does not subtract cash and investments or add debt, leases and other claims. The ratio also uses an issuer-defined non-GAAP revenue measure, not GAAP sales. The observable price therefore puts a high multiple on next-year adjusted revenue, but it does not reveal the market’s precise margin or RPO-conversion assumptions. I did not verify a reliable consensus bridge for either; those expectations remain unknown.
The Mispricing
The possible disagreement is between a $26.1 billion RPO balance and a $26.4 billion equity-value proxy. The comparison is tempting and incomplete. RPO includes deferred revenue, amounts expected to be invoiced, and estimates of variable consideration, including pass-through costs. The 10-Q says OpenAI represented 34% of Q3 revenue and 25% of nine-month revenue; MBZUAI represented 38% and 45%, and G42 7% and 9%. MBZUAI and G42 are related parties. This confirms customer concentration without treating OpenAI as the whole customer base. Cerebras expects about 33% of RPO to be recognized over the initial 24 months ending Sept. 30, 2028, 44% over months 25–48, and the remainder later. Those dollars are not current revenue, gross profit, free cash flow or common-equity value.
The core measure also needs scrutiny. Cerebras reported a $1.1 billion customer-warrant asset at Sept. 30 and recorded $36.4 million in Q3 and $82.7 million year to date as reductions in GAAP revenue. The 2026 Core Revenue measure excludes that amortization. The OpenAI warrant is equity consideration: it can add shares as contractual milestones are met. This accounting treatment is not evidence of wrongdoing, but it is an economic cost that a sales multiple cannot ignore.
The financing side is linked to delivery. OpenAI funded $1.005 billion in January; $175.489 million had been credited through non-cash services and other permitted credits, leaving $829.082 million outstanding at Sept. 30. The loan bears 6% interest and can be repaid in cash or through services under the MRA. On specified termination or trigger events, the filing says the balance may become immediately repayable. That is customer financing, not earned revenue; a delivery shortfall can affect both service economics and the path for settling the loan.
The strongest positive evidence is tangible: Core Revenue grew 70%, services grew rapidly, the Q4 guide implies further scale, and the company reports $8.21 billion of total liquidity under its definition, including an $850 million undrawn revolver. Those facts can support a premium. The unresolved gap is whether RPO can be delivered and converted into cash after data-center costs, service credits and equity consideration. I cannot show that the resulting cash conversion supports the 10/5 hurdle.
The Positioning
CBRS traded 36.87 million shares on Nov. 13, about 2.1 times its Nov. 12 volume. That is event activity, not proof of who bought or whether shorts covered. Current short interest, holders, borrow terms, options positioning, fund flows, spread, depth and exit capacity were not verified. The data do not support a squeeze or crowding claim.
The Catalyst
The next operating test is Q4/FY26 reporting. The date was not verified. Record:
- Core Revenue against $310–$330 million, alongside GAAP revenue and the reconciliation items.
- Core gross margin of 47%–49% and Core operating margin of -1% to +1%, against GAAP cash flow and stock compensation.
- OpenAI capacity delivered against the MRA schedule, including any excused delays, service credits and non-cash loan offsets.
- Any commencement milestones, contract value or revenue timing disclosed for Microsoft inference services.
- Issued shares and any further OpenAI or AWS warrant vesting or exercise.
The cheapest falsification test is the next 10-Q: compare contracted capacity with delivery, Core Revenue with GAAP sales, operating cash with capital spending, and new shares with warrant milestones. A headline that repeats “$26 billion backlog” without this bridge is not new evidence.
The Payoff
This six-month sensitivity runs through May 14, 2027. It uses the Nov. 13 close, 243.003 million shares reported Nov. 5, and the FY26 Core Revenue guide. The current equity/Core Revenue multiple is 27.81×. The top case expands that multiple to 31× on the $960 million guide high; the base holds 27.81× on the $950 million midpoint; the bottom compresses it to 22× on the $940 million low. The 31× and 22× assumptions are not calibrated to a peer set or historical CBRS band. They are transparent valuation sensitivities, not fair value or price limits. The per-share model does not include future warrant exercises.
Price Target and Probability Map
| Scenario | Probability | Target / Level | Return / Payoff | Horizon | Conditions | Evidence Quality |
|---|---|---|---|---|---|---|
| Top Case | 15% | $122.47 | +12.6% | 6 months | 31× Core Revenue multiple on the $960M guide high; Q4 guide is met and Microsoft/OpenAI capacity is delivered | Low |
| Base Case | 55% | $108.72 | 0.0% | 6 months | 27.81× on the $950M midpoint; the guide holds, but no additional multiple expansion is supported | Low |
| Bottom Case | 30% | $85.10 | -21.7% | 6 months | 22× on the $940M guide low after delivery delays, lower gross economics or customer/financing concerns | Low |
| Invalidation | n/a | Insufficient data | Not a price stop | 6 months | Filed capacity delivery, positive GAAP operating cash generation after capex, and reduced customer concentration would require a new model | Low |
The probabilities are subjective weights, not observed frequencies. The 55% base gives the most weight to the filed guide holding without multiple expansion; the guide is specific, but fulfillment and cash conversion are unproved. The 30% bottom reflects disclosed delivery delays, customer concentration, financing dependence and a high starting multiple. The 15% top recognizes rapid Core Revenue growth and new hyperscaler relationships, but has low evidence quality because the 31× multiple is uncalibrated. At 30× on the guide high, the top sensitivity is $118.52, or +9.0%, short of the +10% base hurdle. A +10% price needs about 30.27× even at the $960 million guide high. At 24× on the low guide, the bottom sensitivity is $92.84, still -14.6%. The weighted reference value is $103.70, or -4.6% gross before costs. The base is flat; the adverse case is not bounded within -5%; base/adverse reward-risk is 0:1.
Current market level and timestamp: $108.72, Nov. 13, 2026, 4:00 p.m. EST; regular close, reference only.
Primary instrument: CBRS Class A common stock, USD. No entry is set.
10/5 favorable base move: 0.0%.
10/5 credible adverse move: 21.7% in the valuation sensitivity; downside beyond it is possible.
10/5 measurement basis: reference-only; share count dated Nov. 5, no live entry.
10/5 status: Reject.
Confidence: Low. Revenue, RPO, debt, warrant, share and control facts are filed; future delivery, cash conversion and a defensible valuation multiple are uncertain.
The Kill Shot
The strongest bull case is not a slogan. Q3 Core Revenue grew 70%, services revenue climbed, the Q4 outlook calls for near-breakeven core operations, and the RPO schedule provides multi-year demand visibility. Cerebras has cash and marketable investments, plus substantial unused revolving capacity. Microsoft and Oracle contracts could broaden the customer base.
The fragile assumption is that reported RPO will convert into recurring, profitable, collectible service revenue on schedule. The filing says current OpenAI capacity delivery lagged the MRA schedule before excused-delay adjustments, with delays possibly continuing into part of 2027. Service credits began accruing Oct. 1 on capacity shortfalls that are not excused and remain uncured. The OpenAI loan is secured and linked to this same delivery relationship. If OpenAI slows, or capacity and margin ramp late, the backlog, loan repayment and customer warrant all weaken together.
What Could Go Wrong
The company could meet the revenue target while missing cash conversion. Data-center leases and equipment spending precede recognition; nine-month PP&E purchases were $1.0 billion. Pass-through costs can raise reported RPO and revenue without equivalent gross profit. The $829 million OpenAI loan is financing, not revenue, and $444 million of cash is restricted. Customer warrants and employee equity can expand the denominator. The filing also says disclosure controls were not effective as of Sept. 30, including weaknesses around revenue recognition, inventory, data-center assets and equity administration; it reported no material misstatement, but remediation is incomplete.
What Would Prove This Wrong
Reopen the long screen if a later filing shows the OpenAI delivery schedule is back on track, service credits are resolved, Microsoft milestones are met, GAAP operating cash exceeds ongoing capital needs, and equity dilution is reconciled. The model should then use actual delivered volumes, recognized core and GAAP revenue, cash interest, leases and current shares. Do not treat meeting the revenue guide alone as proof of a 10% base or 5% downside bound.
Risk Audit
- Strongest counterparty argument: core growth is rapid, Q4 Core Revenue guidance is $310–$330 million, the OpenAI agreement is committed capacity, and new Oracle/Microsoft deals may diversify demand.
- Load-bearing assumption: the equity value requires the RPO to convert into margins and cash despite the large customer-linked financing and warrant terms.
- What the market may already know: Cerebras has disclosed the RPO, $1 billion loan, customer-warrant accounting, milestone delays and concentration in the same 10-Q; the price reaction is public.
- Why a directionally right long can lose: the stock can re-rate ahead of the next report, but a delivery delay, financing reset or lower margin can compress a still-high multiple.
- Execution and tail risks: the $106 after-hours mark is not used; no current executable quote, spread, depth, venue, volume quality or exit capacity was verified. Customer loss, lease obligations, warrant dilution and reporting-control failures add discontinuous risk.
- Thesis invalidation: verified delivery, margin and cash improvement would require a fresh valuation; the current Reject is not a short recommendation.
- Execution cancellation: absent a current regular-session quote and exit-liquidity checks, keep entry null and execution disabled.
Best Trade Strategy
No trade. Use CBRS common stock only as the research instrument. entry.price remains null and execution.can_execute is false. Do not use options, leverage, margin, market orders or price-floor rules. Revisit after the next filed results and documented capacity delivery. This signal is research context, not a pending order.
Sources
- Cerebras, Q3 2026 earnings release, SEC Exhibit 99.1, Nov. 12, 2026.
- Cerebras, Q3 2026 Form 10-Q, filed Nov. 12, 2026.
- CBRS regular close and after-hours snapshot, StockAnalysis, Nov. 13, 2026.
- Liftoff Mobile, Q3 2026 issuer release, Nov. 12, 2026.
- Liftoff Mobile, Q3 2026 Form 10-Q, filed Nov. 13, 2026.
- LFTO regular close and history, StockAnalysis, Nov. 13, 2026.
- Jones Soda, Q3 2026 issuer release, Nov. 10, 2026.
- Jones Soda, Q3 2026 Form 10-Q, filed Nov. 10, 2026.
- JSDA Nov. 13 historical price, Investing.com, Nov. 13, 2026.
- SPY regular close and history, StockAnalysis, Nov. 13, 2026.
- May 29 Desk screen that mentioned CBRS as a rejected IPO comparator.
Research Quality Scorecard
| Criterion | Score | Evidence-based reason |
|---|---|---|
| Market disagreement | 4/5 | A $26.1B multi-year RPO sits beside a $26.4B equity-value proxy, but timing, pass-through and margin prevent a direct valuation comparison |
| Evidence base | 4/5 | Current SEC results and 10-Q provide revenue, RPO, financing and warrant bridges; contract identity and future delivery remain uncertain |
| Positioning and flows | 2/5 | Event volume is observed; current ownership, short, derivatives and investor flows are not |
| Catalyst path | 3/5 | Q4 guidance, Microsoft milestones and OpenAI delivery can be tested; dates and amounts are not fully disclosed |
| Payoff architecture | 2/5 | Transparent valuation sensitivity, but multiple assumptions are uncalibrated and downside is not bounded |
| Invalidation discipline | 4/5 | Delivery, service credits, margins, cash, and shares are monitorable |
| Differentiated insight | 4/5 | The same customer is a large buyer, lender and warrant holder; the relationship couples revenue, funding and dilution |
| Client value | 4/5 | Separates RPO from profit and restricted customer finance from free cash |
| Total | 27/40 | Reject / no-trade; the failed 10/5 economics take precedence over score |
Bottom Line
Cerebras has real demand and a large RPO. But the same customer relationship supplies a $1 billion working-capital loan, a share warrant and a large part of contracted capacity. As of Sept. 30, revenue delivery lagged the MRA schedule before excused delays, core operating margin remained negative, and new capacity requires heavy investment. At 27.8× FY26 Core Revenue, a flat base and -21.7% stress fail 10/5. Reject / no trade.
AI Illustration Prompt
Create a restrained financial-editorial illustration of an AI data center buildout: a long contractual ledger marked “$26.1B RPO” flows toward a row of partly energized server halls, while a separate restricted cash account and a customer warrant sit beside a capacity-delivery gauge. Show a small customer badge linked to both the contract and loan, and leave some server racks unpowered to signal timing risk. Avoid company logos, dramatic rockets, or fabricated price charts. Use graphite, pale blue and muted copper. Add a small, readable “The Mispricing Desk” watermark at the lower right.