Phase Space AI

Valuation

Caris Life Sciences [CAI]

Caris Life Sciences, Inc. [CAI] — Valuation Analysis (Task 3)

Spot $15.63 (2026-07-27) · Shares 282.68m · Market cap $4,418m · Net cash $441m · EV $3,978m Model: CAI_Financial_Model.xlsx (live formulas; every figure below read back from the workbook's own formula evaluation, not from the builder script — see Research §8).

---

1. DCF

1.1 Cost of capital

A 13-month beta is not a beta. Realized volatility is 62.4% (full history) / 70.5% (90-day). A volatility-anchored discount rate would be materially higher than 11.23%, which biases the DCF upward. Sensitivity is run at 10-14% below.

1.2 Scenario architecture

Driven by two variables — clinical case volume and blended ASP — because those are the only two drivers the company itself uses.

1.3 Base-rate check (required, task-deltas.md)

Reference class first (Chan/Karceski/Lakonishok 2003): growth persistence beyond chance is close to nonexistent; sustained 20%+ growth for 5+ years at >$1bn revenue scale is a top-decile outcome.

• Base (12.4% FY26-30 CAGR) sits inside the reference class and is deliberately below both company guidance (23-26% for FY2026) and consensus. It explicitly does not extrapolate the FY2025 +97% print, because the 10-K's own bridge shows 74.6% of that increase was ASP/payer mix, which is a step-function repricing, not a growth vector.
• Bull (19.2% CAGR) is a top-decile outcome and is named as such. It requires the territory expansion to deliver and Caris Detect to scale on cash-pay.
• The Base case's volume assumption is an explicit override of the observed trend and must be named as such: disclosed clinical volume ran 50,032 → 50,763 → 52,700 → 52,800, i.e. +5.5% over twelve months, and Base assumes +18.8% for FY2026. The sole justification is the disclosed 82→146 territory expansion and the 270→~300 commercial headcount build. If Q2-2026 prints below ~58,000 cases the override is falsified and Bear becomes the reference case.

1.4 DCF sensitivity — Base case value per share

(Grid computed off the workbook's Base FCF stream; the $16.19 cell is the workbook value read back directly. Others are recomputed on the same stream and are flagged as derived, not separately audited.)

Read: the Base DCF straddles spot. At any WACC above ~11.5% the Base DCF is below $15.63. Given a 70% realized-volatility stock with a 13-month beta, that is not a comfortable margin of safety — the DCF is telling you the stock is approximately fairly priced under the Base case, not that it is cheap.

---

2. Comparable companies

2.A Trading comps (prices 2026-07-27 close, Alpaca SIP; balance-sheet items from EDGAR XBRL)

Growth-peer mean (TEM, VCYT): 6.02x. Mean ex-CAI/MYGN/ILMN: 9.57x.

Notes and honesty flags: - Exact Sciences excluded — acquired (last trade 2026-03-20). Its removal is itself a sector datapoint. - Peer FY26E revenues other than CAI's are round-number estimates, flagged as estimates, not sourced consensus. CAI's $1,001m is sourced (Alpha Vantage, 13 analysts). Peer multiples are therefore indicative. - TEM share count is WeightedAverageNumberOfDilutedSharesOutstanding for Q1-2026 (178.96m) — a proxy, since dei point-in-time shares were not tagged. - MYGN excluded from means (declining revenue; a value-trap comparator, useful only as a floor).

2.B Implied values

CAI is the cheapest name in the growth-diagnostics set on EV/FY26E revenue — despite having the best near-term financial profile of the loss-making cohort: 65-66% gross margin, positive GAAP operating income, positive FCF. TEM is loss-making at 4.81x; GH is deeply loss-making at 16.69x.

The honest question is why, and the answer is not "the market is wrong." It is that the market is discounting the quality and durability of the revenue: ~40% Medicare concentration, revenue recognised on estimated collections with $33.6m of FY2025 and ~$81m of Q4-2025 being prior-period true-ups, a disclosed history of payer recoupment, and clinical volume that has grown 5.5% in a year. A 3.94x multiple on revenue with those properties is not obviously a discount to a 4.81x multiple on cleaner revenue. Do not treat "cheapest in the peer set" as evidence of anything — the peer-group multiple is not the right anchor when the revenue-quality characteristics differ this much.

---

2.C Transcript Mention-Frequency table (REQUIRED)

Source: SEC EDGAR 8-K Exhibit 99.1 quarterly earnings releases — one source across the whole series. History window: four quarters (2025Q2–2026Q1) — the complete public history, and a thin sample. Word counts 3,358 / 3,419 / 3,672 / 2,842. All figures per 10,000 words. These releases are 100% prepared company language; there is no Q&A section to separate, so the prepared-remarks share is 100% for every term — which makes each mention a deliberate allocation of company-authored space.

Emerging: Caris Detect, whole genome, methylation, ChromoSeq, MI Clarity, MolDX, sales force, multimodal. Decaying: ASP, volume, therapy selection, MI Profile, Caris Assure, early detection, tissue, liquid biopsy. Decayed to zero: MRD. Stable: Adjusted EBITDA, free cash flow, molecular profiling, reimbursement/payer/coverage.

Provenance (required by the reference file). The dominant hypothesis in this memo — "the ASP driver is exhausted and the reported revenue series is contaminated by prior-period true-ups" — was generated by this pass, not confirmed by it. The ASP decay was noticed first; chasing it produced the 10-K bridge ($311.4m of $417.6m from ASP) and then the J.P. Morgan preliminary release disclosing ~$81m of Q4-2025 prior-period true-ups — a figure that appears in no quarterly earnings release, in no MD&A, and in no sell-side summary read for this memo. Everything else in the analysis followed from that. The one hypothesis carried in from prior views — "a recent IPO down 60% is probably a broken growth story" — was partly falsified by this work: stripping true-ups shows clean sequential revenue fell only 2.8%, not the 26% the headline implied.

---

2.D Consensus vs. house view

Source: Alpha Vantage EARNINGS_ESTIMATES, one call, cached to data/av_estimates_CAI.json (25/day cap shared across all concurrent runs).

Consensus is guidance. $1,001.2m is the bottom of the $1.00-1.02bn guide. There is no independent Street view of FY2026 to have a variant against.

Estimate-revision direction is unambiguously negative on earnings while revenue holds: FY2026 EPS $0.1401 → $0.1329 → $0.1299 over 90 days (−7.3%); FY2027 $0.3735 → $0.3557 (−4.8%); Q2-2026 $0.0329 → $0.0128 (−61%); Q3-2026 $0.0581 → $0.0403 (−31%). Down-revisions outnumber up-revisions 6:1 on Q2. Analysts are holding the revenue line and cutting the cost line — consistent with the ~$70m incremental commercial spend of the territory build being absorbed into estimates.

The arithmetic consensus is signing up for — this is the crux of the whole memo

Consensus quarterly path: Q1 $216.2m (actual) → Q2 $238.1m (+10.1% seq) → Q3 $261.5m (+9.8%) → Q4 $285.4m (+9.1%).

Holding Q1-2026's realised blended ASP of $3,992 flat and pharma at ~$12m/quarter, that path requires:

Consensus requires year-on-year clinical volume growth to accelerate from +15% in Q1 to +30% by Q4, off a base that grew 5.5% sequentially over the preceding twelve months (50,032 → 52,800). That is the single number on which this equity turns.

House vs. Street — and the sign is the problem

The house Base case is BELOW consensus on both years while the house bias is long. That combination cannot support a long thesis and is the reason Task 5 reaches the verdict it does. A −1.3% FY2026 gap is inside the noise of a company that recognises 4-5% of quarterly revenue as prior-period true-ups; the −7.1% FY2027 gap is more material but is two years out and is produced by my scepticism about the volume ramp, not by an observed datapoint that contradicts consensus.

---

2.E Consensus bridge — the required numbers-vs-multiple decomposition

Street position (searched 2026-07-27; aggregators disagree and the spread is disclosed): consensus rating Strong Buy / Moderate Buy; average price target $26.11–$28 across sources (13 analysts per S&P Global); range $21 to $36. Evercore ISI (Vijay Kumar) maintained Outperform on 2026-07-06 while cutting the target from $30 to $25. Baird has initiated coverage. Direction: targets are falling while ratings stay bullish — sentiment is weakening at the margin even as the rating labels do not move.

Decomposition on a shared revenue base (consensus FY2026E $1,001.2m, net cash $441m, 282.7m shares):

Verdict: the gap is entirely about the multiple, not the numbers. House FY2026E revenue is within 1.3% of consensus. The Street's $26.11 average requires EV/FY26E revenue to expand from 3.97x to 6.93x — a 75% multiple re-rating — with essentially no change in the revenue forecast. That is a specific, falsifiable claim: twelve months from now either CAI trades at ~7x forward revenue (roughly Veracyte's 7.24x, above Tempus' 4.81x) or it does not.

Why the Street might be right, stated as a live possibility rather than a straw man. Three arguments support a re-rating that has nothing to do with the numbers: (i) Caris is the only name in the growth cohort that is simultaneously FCF-positive and growing 20%+, and the market has not yet priced that combination because it has only existed for three reported quarters; (ii) EXAS was taken out and the sector is consolidating, so a 3.9x asset with a genuine WES+WTS moat is strategically cheap; (iii) the Q1-2026 sequential "collapse" that caused the derating was ~90% a true-up artifact — a fact that is knowable but is not in any earnings release, and mean-reverts as the comparison base cleans up over 2026. Any of these could produce a 6-7x multiple without a single estimate moving.

The real risk to the house view, stated plainly: if the market's 3.94x is a temporary post-IPO, post-derating dislocation rather than a considered discount for revenue quality, then anchoring on the current multiple systematically understates fair value, and the Street's numbers-neutral re-rating call is simply correct. The house does not have evidence against that — it has scepticism about a volume ramp, which is not the same thing.

---

2.F Options market read

Alpaca options snapshots, 2026-07-27. The relevant finding is that this options market is barely functional.

• Total open interest across the entire Aug + Sep chain is ~3,500 contracts, of which 2,326 sit in a single out-of-the-money $20 call.
• Implied vs realized: IV 81-101% against 90-day realized volatility of 70.5% and full-history realized of 62.4%. Implied exceeds realized by 10-30 points — the variance risk premium is wide open here, which makes buying premium negative-EV before spreads.
• Implied move into the catalyst: the 21-Aug $15 straddle at last close (call $1.77 + put $1.23) = $3.00 ≈ 19.2% of spot, for an expiry 16 days after the 5-Aug print.
• Skew: 90-day realized 70.5% vs full-history 62.4% — realized volatility is rising, so IV is not obviously mispriced, merely expensive and untradeable.

Conclusion carried to Task 5 Gate 5: any options expression pays a bid/ask spread of 87-166% of mid plus a 10-30 point volatility premium. The equity is the only sane vehicle.

---

3. Factor & Anomaly Scorecard (required)

Direction assessed for a contemplated LONG.

Synthesis — the factors split cleanly, and along an informative fault line

Everything measuring the business is a tailwind: gross profitability is top-decile (47.9%), accruals are strongly negative (−13.4%, i.e. cash flow exceeds accounting earnings by a wide margin), and the F-score is a respectable 6/9. Everything measuring the market's opinion is a headwind: bottom-decile 12-1 momentum (−37.4%), 60% off the high, below both moving averages, and EPS estimates being cut hard.

Two consequences flow directly into Task 5:

1. This is an unusually poor short candidate. The reference file's Gate 1 standard says shorts work empirically in high-accrual, low-F-score, weak-gross-profitability names. CAI is the opposite on all three. A short here would need extraordinary narrative evidence to overcome the quantitative disagreement, and does not have it — it would also have to overcome $1.49m of clustered director buying at the lows.
2. It is also a factor-conflicted long. Buying bottom-decile momentum into negative revision momentum is the "falling knife" the Gate 6 literature specifically penalises. The profitability and accruals evidence is real, but momentum and revisions are the two factors with the strongest out-of-sample record for timing, and both say not yet.

Would this name have survived the idea-screener funnel? Almost certainly not. The screen's momentum/trend stage rejects a name at 0.398 of its 52-week high and below its 200-day MA, and the estimate-revision overlay rejects negative 30-day revisions. It would have cleared liquidity and the EDGAR quality overlay comfortably. This is a user-named ticker, not a screened one, and that is base-rate information that belongs in the decision — the empirical prior for names arriving this way is worse than for names that cleared the funnel.

---

4. Valuation summary

Blending rationale: over a 12-month horizon a price is set by a multiple far more than by a DCF, so the Bear leans toward the comps floor (a DCF-only $5.82 implies 1.30x EV/sales for a 65%-gross-margin, FCF-positive lab, which is below anything in the peer set except a shrinking Myriad) and the Bull is held below its DCF (a $35.56 DCF implies 9.27x, above Veracyte and near Natera, which the Bull's own assumptions do not earn).

Probability-weighted value (30% / 50% / 20%): $16.80 — 7.5% above spot.

House rating: HOLD / fairly valued. Not "cheap." The 3.94x headline multiple looks like a discount only if the revenue is comparable to the peer set's, and Task 1 §2.1 documents specifically why it is not.

→ Continue to CAI_Trade_Construction.md (Task 5). A HOLD-with-upside valuation is not automatically a position, and this one is not.