Caris Life Sciences, Inc. [CAI] — Company Research

Phase Space Research · investment-memo v1.4.2 · Task 1 · 2026-07-28

0. Resolved identity (Step 0, done before any analysis)

Ticker-collision check. CAI previously belonged to CAI International (container leasing, acquired by Mitsubishi HC Capital in 2021). It was reassigned to Caris at the June-2025 IPO. Nothing in this memo relates to the container-leasing company. Partial matches (CAII, HCAI, PCAI) are unrelated registrants.

1. Mention-frequency FIRST — run before any view was formed

Per references/mention-frequency.md, this was executed as the first analytical step, generatively, before the model, the comps or the sell-side were touched. What follows is the raw output and the open questions it generated; the interpretation is separated deliberately.

Corpus — pinned, single-source, and named

Source: SEC EDGAR 8-K Exhibit 99.1 quarterly earnings releases, narrative section only (everything before the "About Caris Life Sciences" boilerplate). Alpha Vantage's transcript endpoint was unavailable — the shared 25-request daily cap was exhausted at ~14:10 PDT on 2026-07-28 — so the first-party EDGAR corpus was used instead, which references/mention-frequency.md explicitly endorses ("8-K Ex-99.1 earnings releases from EDGAR (complete and near-constant length) both worked well on the MU and NET re-runs").

One source across the whole series. No mixing. Counts are normalised per 10,000 words — the ISRG failure mode.

WINDOW HONESTY — read this before any "first-ever" claim. Caris IPO'd on 2025-06-18. The window examined is four quarters, which is the company's entire life as a public reporting company. That is simultaneously "all public history" and a very short series. Four points cannot establish a trend with confidence. Every claim below is stated as "first in the four-quarter window examined." No claim in this memo asserts a first-ever mention across the company's full history, because no such history exists in the public record.

Raw counts, per 10,000 words

Open questions generated (written down before investigation, no interpretation attached)

1. Why did ASP mentions fall 76% between Q3-2025 and Q1-2026 — a term that had been the single most-emphasised metric in the two prior releases?
2. Why did volume decay 66% over the same span?
3. What is out-of-year revenue from over collections — a phrase appearing exactly once, in Q4-2025, never defined and never quantified in the release?
4. What are Caris Detect, Achieve and whole genome, which appear from nothing in the two most recent quarters?
5. Is the decay in ASP/volume an artefact of the shorter Q1-2026 release (895 words vs 1,290)? — the ISRG Ion trap.

Investigation results

Q5 first, because it is the falsification test. The decay is not a length artefact: the figures above are already normalised per 10k words, and the raw counts fall too (ASP 5 → 5 → 2 → 1). The decay survives normalisation. This is the check that killed the first ISRG finding; it does not kill this one.

Q1/Q2 — and this is the real finding. The mention decay tracks an actual retirement of disclosure. The Q2-2025 and Q3-2025 releases each contain a "Summary Financial Results" table disclosing:

• Total clinical case volume
• Total clinical ASP
• MI Profile for therapy selection ASP and volume
• Caris Assure for therapy selection ASP and volume
• Molecular profiling vs pharma revenue split

That table is ABSENT from the Q4-2025 and Q1-2026 releases. Verified by direct string search: the phrase "Summary Financial Results" returns 1 hit in each of Q2-25 and Q3-25 and 0 hits in each of Q4-25 and Q1-26; "MI Profile for therapy selection" returns 2 and 2, then 0 and 0.

Caris disclosed per-product ASP for two quarters, then stopped — and stopped in the quarter in which, by our own reconstruction, ASP peaked. This is the same shape as the TWST finding (orders disclosure retired for six quarters) and the CIEN finding (backlog section deleted from the FY2025 10-K), and it was surfaced here by the same generative method rather than by looking for it.

Q3 — out-of-year revenue from over collections. This became the spine of the memo and is treated in §2.

Q4 — the emerging cluster. Caris Detect (multi-cancer early detection), Achieve (the ACHIEVE 1 validation study) and whole genome are a coherent, corroborated pivot, not noise. Independent corroboration from outside the transcripts: - Commercial launch of Caris Detect on 2026-07-01 (company press release, BioSpace, GenomeWeb) — three weeks before this memo. ACHIEVE 1 reported 61.3% sensitivity in stage I/II cancers. - ClinicalTrials.gov (API v2, query.term=Caris, sponsor filtered to Caris entities): NCT07680868 "Caris Chromoseq Data Collection", RECRUITING, started 2026-07 — a genuinely new registration in AML/MDS/MPN, matching the ChromoSeq mention that appears from zero in Q1-2026. NCT01499394 "Caris Biorepository Research Protocol" is registered under conditions including "Early Detection of Cancer" and "Minimal Residual Disease" — the underlying registry for the Detect/MRD programmes. - Partnership with Everlywell to distribute Detect (consumer channel, cash-pay); collaboration with Genentech on target discovery.

I opened each registry record and read the condition fields before citing them — the ISRG misattribution failure mode. NCT07680868 and NCT01499394 do say what is claimed here.

Provenance of hypotheses (required by the skill). The ASP-decay finding, the disclosure-retirement finding and the out-of-year thread were all generated by this process, not confirmed by it. No view existed when the counts were run. The Caris Detect thread was also generatively surfaced. Nothing in this memo originated from a prior view about Caris.

2. The central finding: 74% of the first-ever operating profit is a prior-period true-up

The trail

Q4-2025 release, in the FY2026 guidance section, one sentence, never repeated, never quantified:

"Molecular profiling revenue is expected to grow approximately 21% to 22% year over year in 2026. Excluding out-of-year revenue from over collections recorded in 2025, this range implies growth of approximately 26% to 28%."

That sentence is the only acknowledgement in any company communication. The phrase "out-of-year" appears zero times in the FY2025 10-K. So the disclosure was solved for two independent ways.

Method 1 — solve the guidance arithmetic. FY2025 molecular profiling revenue was $766.7m. If FY2026 MP revenue grows 21–22% on the reported base but 26–28% on the clean base, the implied out-of-year amount is:

→ $30–36m.

Method 2 — the audited 10-K. Buried in the revenue note (not MD&A, not the release):

"for the years ended December 31, 2025, 2024, and 2023, the Company recorded $33.6 million, $3.9 million, and $(1.9) million, respectively, of adjustments to revenue related to services delivered in prior periods, which is based on variability that was subsequently resolved."

$33.6m — inside the independently derived $30–36m range. Two methods, one answer.

What it does to the earnings

Caris recognises molecular-profiling revenue at an estimated collectable amount (ASC 606 variable consideration, portfolio method), then trues it up as cash actually arrives. A prior-period true-up therefore carries ~100% incremental margin — the cases were run, sequenced and costed in an earlier period, so no incremental cost of services attaches.

74.5% of FY2025's first-ever positive GAAP operating income is revenue for services delivered in prior years. And Q1-2026 was operating-loss-making on the same basis: reported +$5.3m less a $10.3m true-up = −$5.0m.

The true-up is also accelerating, not normalising: $(1.9)m → $3.9m → $33.6m → $10.3m in a single quarter (vs $3.9m in Q1-2025, a 2.6x increase year on year).

Independent corroboration of the clean margin

FY2025 reported gross margin was 66.4%. Ex the $33.6m: 64.9%. Q1-2026 reported gross margin: 65.0%. The two agree to 10bps from completely different directions. The run-rate gross margin of this business is ~65%, and the reported FY2025 66.4% is flattered.

What this finding is NOT

Intellectual honesty requires naming the limits:

1. The growth is real. Ex true-up in both years, FY2025 revenue grew +90.6% (vs +97.0% reported). This is not a case where the growth is manufactured.
2. The cash is real. These are over-collections — actual cash received above the accrued estimate. CFO of +$83.2m in FY2025 is genuine. This is a timing and quality issue, not a fictitious-revenue issue.
3. It is disclosed. The $33.6m sits in the audited 10-K. Any analyst who reads the revenue note can find it. This materially weakens any claim that "the market doesn't know" — and it is why Gate 2B is not claimed.
4. Conservative estimation is not misconduct. A company that under-accrues and later collects more is being conservative, which is arguably the right behaviour. The criticism is narrower: the market is capitalising a profitability inflection whose majority component is non-recurring, and the company chose to retire the ASP disclosure that would have made that visible.

3. The business

Caris is a precision-oncology molecular-profiling laboratory that has bolted an AI/data business onto a reference-lab core. A physician sends a tumour (or blood) sample; Caris runs whole exome + whole transcriptome sequencing (and increasingly whole genome), returns a therapy-selection report, and retains the clinico-genomic record in what it claims is the largest such database — 1,070,000+ total profiles and 790,000+ matched profiles through 2026-03-31, including 677,000 whole-exome and 728,000 whole-transcriptome profiles.

Revenue architecture

Two things stand out:

• Molecular profiling is 94.4% of revenue. This is a reimbursement business, not a data business, whatever the "AI TechBio" positioning says.
• Pharma revenue FELL 28.2% in FY2025 ($63.1m → $45.3m) — and FY2026 guidance of $75–85m implies +65% to +88% growth off that depressed base. That is the single most aggressive component of the guide and it is attached to the segment that just shrank.

Geography — and a fact the "global platform" language obscures

International revenue is lower today than in FY2023. Caris is a US reimbursement business with a rounding error attached. 99.0% of total assets are US-located (10-K).

Unit economics — the ASP series

Molecular-profiling revenue = clinical case volume × blended clinical ASP (verified exactly: Q3-2025 50,763 cases × $4,089 = $207.57m against $207.587m reported).

ASP fell 25.4% sequentially on essentially flat volume (52,700 → 52,800 cases). Q4-2025 and Q1-2026 figures are derived, because that is exactly the quarter the company retired the ASP table.

Honest counterweight: Q1 is genuinely seasonally weak in US diagnostics (deductible resets shift payer mix), and Caris ran a sales-force re-alignment in January 2026 that it says cost it volume, exiting at a ~56,000-case quarterly run-rate for February/March. Both are real mitigants. But note the prior-period true-up was $10.2m in Q4-2025 and $10.3m in Q1-2026 — essentially identical — so the true-up does not explain the Q4→Q1 ASP collapse. Something else did.

Volume growth is decelerating independently of price

+22.0% (Q2-25) → +18.2% (Q3-25) → ~+20% (Q4-25) → +15% (Q1-26), against FY2026 guidance of +20%. Q1 already ran 5pp below the full-year guide.

Receivables

DSO tripled from 11 to 38 days across the two quarters in which the ASP disclosure disappeared. Some of this is simply a larger revenue base and normalising collection cycles from an unusually low Q3 — but a 4.2x AR build in Q4-2025 alongside a 30.9% sequential ASP jump is the classic shape of revenue recognised ahead of cash.

Payer concentration — deteriorating

Payer 1 is almost certainly Medicare/MolDX. Seven in ten revenue dollars now come from three payers, up 12pp in one year. Payer 2 is 22.6% of accounts receivable against 21.1% of revenue. A single MolDX rate action or coverage-policy change is a material, non-diversifiable event for this business — and this is exactly the mechanism by which the ASP series can reverse.

Competition

Caris's differentiation is genuinely defensible on the science: whole exome + whole transcriptome as standard (most competitors run targeted panels), and MI Cancer Seek has FDA approval as a companion diagnostic. The transcriptome is the real moat — it is what powers the AI Insights products (GPSai, MI Clarity) and it is expensive to replicate retrospectively.

Balance sheet and capital

Debt is expensive: FY2025 interest expense $56.9m on ~$378m ($44.0m excluding discount amortisation ≈ 11.6% cash cost). The $400m facility was refinanced in Q1-2026 at a lower cost, with access to additional capital from Blue Owl and Blackstone. The $2.5bn accumulated deficit means no cash taxes for the entire forecast horizon — worth real money in the DCF.

$100m share repurchase authorised 2026-06-08 — twelve months after IPO, by a company that has never been full-year GAAP profitable. Read charitably: management thinks the stock is cheap. Read sceptically: buying back stock at 3.9x revenue while carrying 11%+ debt is an odd capital allocation.

4. Ownership and insider activity

Insider transactions — real, clustered, open-market BUYING

Applying the empirical asymmetry required by references/task-deltas.md (Cohen, Malloy & Pomorski 2012 — buys are signal, sales are mostly noise), all 39 Form 4s since IPO were parsed:

Discretionary selling: essentially none. The only open-market sale in the entire record is Luke Power's December-2025 disposal under a 10b5-1 plan (60,850 + 1,400 shares) — precisely the routine, low-information category. Everything else coded F (May-2026, five insiders including CEO David Halbert) is tax withholding on vesting, not a sale decision, and A codes are grants.

This is the textbook positive configuration: multiple insiders, non-routine (no 10b5-1), open-market purchases, into a falling price, with the most recent and largest purchase two weeks before this memo at $14.56 — 6.6% below today's price. Under the Cohen/Malloy/Pomorski framing this carries genuine evidentiary weight and it is the strongest single argument against the cautious view.

Institutional ownership

• ~192 institutional holders, ~140.5m shares (13F/13D/G filers).
• Institutional ownership ~38.9%, reported down ~30% quarter-over-quarter — consistent with post-lockup distribution (lockup expired 2025-12-15).
• Float ~145.6m of 282.7m shares → ~48% is insider/strategic and not freely floating. Founder/CEO David Halbert converted 104.76m shares at IPO (Form 4, 2025-06-23), plus J.H. Whitney VI (12.52m), Peter Castleman (10.25m), TSSP/Sixth Street entities (~20m).
• Single share class, one vote per share — no dual-class structure (verified in the 10-K governance section). Unusual and creditable for a founder-controlled recent IPO; Tempus by contrast is dual-class.
• Short interest: 6.84m shares = 4.7% of float. Low. Not a crowded short, and no squeeze dynamic.

5. Unpublished-scoping pass — what is NOT in the filings, and what I did about it

Per references/unpublished-scoping.md, "not disclosed" is a research task, not a limitation to report. Each item states which alternative corpora were attempted and what they returned.

6. Management

David Dean Halbert — Founder, Chairman & CEO. Founded Caris in 2008. Previously founded and led Advance Paradigm / AdvancePCS, a pharmacy benefit manager sold to Caremark in 2004 for ~$5.6bn. He is a genuine operator with one large realised outcome, and he controls ~104.8m shares. He has not sold in the open market post-IPO; his only dispositions are tax withholding on vesting.

David Spetzler — President & Chief Scientific Officer. Molecular biologist; the scientific architect of the WES/WTS platform. Bought stock (small) in November 2025.

Concentration risk in the CODM disclosure. The 10-K states the CODM is the Chairman/CEO and the company operates as a single operating segment managed on a consolidated basis. Combined with founder control of ~37% of shares and a board that includes his relative (Jon Halbert), governance is founder-dominated — mitigated, materially, by the single-class share structure.

7. TAM — bottom-up, with the required implied-penetration statement

Per references/tam-sizing.md, built bottom-up rather than cited top-down.

Therapy selection (the current business): - New US cancer diagnoses per year: ~2.04m (ACS 2025, excluding non-melanoma skin) - Share advanced/metastatic or otherwise CGP-eligible: ~35% → ~715,000 patients - Realistic profiling rate at scale: ~65% → ~465,000 profiled patients/yr - Blended reimbursed ASP: $3,992 (Q1-2026 actual, used rather than an aspirational figure) - US therapy-selection TAM ≈ $1.86bn/yr

Required implied-penetration statement (mandatory, and the point of the exercise): Caris ran ~211,000 clinical cases in the twelve months to Q1-2026. Against ~465,000 eligible profiled patients, that is ~45% penetration of the realistically addressable US therapy-selection market already. At the FY2031 Base-case revenue of $2.08bn, Caris would need either (a) more than 100% of the US therapy-selection TAM as defined, or (b) meaningful revenue from MRD/early detection, or (c) materially higher ASP.

This is the most important number in the memo. The Base case is not a penetration story — the core business is already close to saturating its realistic addressable market at current ASP. The Base case is implicitly a bet on Caris Detect and MRD, i.e. on products that launched three weeks ago with no reimbursement.

MCED (Caris Detect) — deliberately NOT sized into the Base case. Screening-eligible US adults 50–79 ≈ 111m; at 10% penetration and a $949 cash price that is a ~$10.5bn gross opportunity — a number I regard as close to meaningless without reimbursement. Galleri (GRAIL) has been commercially available since 2021 and remains unreimbursed by Medicare. Caris Detect is treated as unpriced optionality in the Bull case only, and the Bull case is given 20%.

8. Risks

1. Payer concentration (70.1% top three, +12pp in a year). The dominant risk. One MolDX rate action reprices the whole ASP series.
2. ASP is the growth engine and it just fell 25.4% sequentially. Volume alone gives ~15%.
3. Earnings quality. 74% of FY2025 operating income is a prior-period true-up; Q1-2026 was clean-basis loss-making.
4. Disclosure retirement. The ASP table disappeared exactly when ASP peaked.
5. Pharma segment. Fell 28% in FY2025; guided +65–88%.
6. Saturation. ~45% penetration of the realistic US therapy-selection TAM already.
7. Caris Detect is unreimbursed and the MCED field's publication interest has plateaued.
8. Expensive debt, $381m at ~10–11.6%, against an accumulated deficit of $2.5bn.
9. Tempus, Roche/Foundation — better-capitalised competitors in the same lane.
10. Post-IPO distribution. Institutional ownership down ~30% q/q; ~48% of shares are insider/strategic.

Counterweights (stated with equal force)

1. Growth is real — +90.6% FY2025 ex-true-up.
2. Cash generation is real — CFO +$83.2m FY2025, +$32.9m Q1-2026.
3. Net cash +$441m, 10% of market cap.
4. Clustered, non-routine insider buying, most recently 2026-07-13 at $14.56.
5. $100m buyback authorised June 2026.
6. Cheapest multiple in the peer set — 3.9x FY2026E revenue vs a 7.8x peer median.
7. PubMed citation volume accelerating — independent evidence of real clinical adoption.
8. Single share class.
9. Every published Street target is above spot ($23–$32, all Buy, all July 2026).

Sources: SEC EDGAR (10-K 2026-03-03 acc. 0002019410-26-000016; 10-Q 2026-05-08 acc. 0002019410-26-000044; 10-Q 2025-11-05 acc. 0002019410-25-000028; 8-K EX-99.1 releases 2025-08-12, 2025-11-05, 2026-01-12, 2026-02-26, 2026-05-07; 39 Form 4s; 424B4 2025-06-20; 8-A12B 2026-06-16), Alpaca Markets SIP (equity bars, options snapshots), ClinicalTrials.gov API v2, PubMed E-utilities, stockanalysis.com (Street targets). Alpha Vantage was unavailable — daily quota exhausted 2026-07-28.