Phase Space AI

Financial Model Notes

Exelixis [EXEL]

Exelixis, Inc. [EXEL] — Financial Model Notes

2026-07-29 · every figure below traces to a named primary filing. No estimate is presented as a fact.

---

1. Quarterly revenue and profitability, built from XBRL

Q4 figures are derived (fiscal year minus the nine-month cumulative) because Exelixis, like most filers, does not tag a standalone Q4. All $000.

Trailing-twelve-month series:

Read this table with §5 of the Research doc. The +30pp of margin in nine quarters is a completed cost reset (two restructurings, 2024 and August 2025, plus an abnormal FY2023 R&D load), not a repeatable curve. FY2026 guidance implies 38.1% — below the current run-rate.

---

2. Revenue mix and the milestone adjustment

From the Note 3 / Note 2 disaggregation of the FY2023 and FY2025 10-Ks, and MD&A milestone disclosure. $m.

Collaboration services turns negative in FY2025 and Q1 FY2026 because the 3% Royalty Pharma royalty on Ipsen's and Takeda's sales is netted against it ($22.8m FY2025, $21.3m FY2024) and development-cost reimbursements collapsed ($3.7m FY2025 vs $25.8m FY2024). This is disclosure working correctly, not an anomaly.

Growth, headline versus clean:

The FY2024 distortion is one line item: a $150.0m commercial milestone from Ipsen on reaching $600.0m of cumulative cabozantinib net sales over four consecutive quarters in its territory, plus $2.2m (Canada) and $11.4m (EMA regulatory). Visible independently in the geography table — Europe was $318.6m in FY2024 versus $145.0m and $154.2m either side — and in Ipsen's share of total revenue: 15% FY2024, 8% FY2023, 7% FY2025.

---

3. Cabozantinib economics — who gets what

Where the royalty sits in the P&L — the reason it is easy to miss: - Exelixis' own U.S. sales → inside cost of goods sold. FY2025 COGS $83.697m on $2,122.8m of product revenue (96% gross margin), of which ~$63.7m is the 3% royalty and ~$20m is genuine product cost. - Ipsen's and Takeda's sales → netted against collaboration services revenue ($22.8m FY2025).

There is no profit share on cabozantinib anywhere. The KNSA structure (a profit-share arrangement capping the terminal margin by construction) is absent here. The only sharing is a 3-point gross royalty already embedded in the reported 37.6% / 39.4% operating margin.

---

4. Balance sheet — Q1 FY2026 (period ended 2026-04-03), $000

Net cash = 226,152 + 551,055 + 649,144 = $1,426,351k. There is no debt of any kind on this balance sheet. The screen's $1,029,692k is understated by $396,659k (27.8%) and could not be reconstructed from any combination of filed lines at any period end from Q1 FY2025 forward.

Accumulated deficit went from $(76,830)k to $(215,794)k in one quarter despite $210.5m of net income — because $430.4m of buybacks were charged against it. That is the buyback showing up on the balance sheet, not a loss.

---

5. Share count and buyback

−9.95% outstanding in five quarters. $2.16bn repurchased since March 2023, 76.7m shares retired at an average $28.14. October 2025 program of $750m completing May 2026; a further $750m authorised May 2026 to end-2027. At $56.51 that is ~13.3m more shares, another ~5%.

Consequence for the model: EPS grows faster than net income, and will keep doing so. Any forward EPS built off a static share count is wrong.

EPS cross-check (the brief's required scale test):

---

6. FY2026 guidance — the forward model, from the company

Issued 2026-01-11, reaffirmed unchanged 2026-05-05.

Guidance excludes all zanzalintinib revenue, stated explicitly in both releases.

Guidance-derived FY2026 operating income at the midpoints: $2,575m − $95m − $900m − $600m = $980m, a 38.1% margin — below the 39.4% TTM run-rate. Both opex lines are guided up in absolute dollars. The screen's op_margin_delta_pp: +9.7 records a completed event; the company is not guiding a repeat.

The bridge that must hold: product revenue guidance of $2,325–2,425m against FY2025's $2,122.8m is +9.5% to +14.2%, midpoint +11.9%. Q1 delivered +8.1%. Q2–Q4 must therefore do +10.0% (low end) to +13.1% (midpoint) against $1,609.5m a year earlier. Tested 5 August 2026.

---

7. Own-multiple history — method and full result

Daily trailing multiples: closing price × verified shares outstanding, less as-known net cash, over as-known TTM earnings stepped in at each 10-K/10-Q filing date (no look-ahead). Basic shares used throughout so the series is internally consistent.

The full window is discarded and the post-2024 window is used. Exelixis' operating margin was 9.3% in FY2023 and is 39.4% today; multiples computed on a near-zero denominator describe a different company. This is a declared regime change, per references/regime-change-test.md — the median of 28.8x is an artifact of the pre-inflection era and applying it would manufacture a target.

---

8. Transcript mention-frequency — full 17-quarter series

Corpus: EDGAR 8-K Exhibit 99.1 quarterly results press releases, 2023-02-07 → 2026-05-05, n = 17, single source, never mixed. Alpha Vantage EARNINGS_CALL_TRANSCRIPT was unavailable (shared free-tier 25/day quota exhausted; the API returned its rate-limit Information payload). Per references/mention-frequency.md the EDGAR first-party corpus is the documented fallback: complete, near-constant length (3,265–5,053 words, mean 3,963), and prepared company language only — no Q&A, which removes the analyst-question artifact entirely.

All values are per 10,000 words. Raw counts are never read; that is the ISRG failure mode.

Emerging / decaying / stable, and open questions

EMERGING — zanzalintinib (trough 8.3 → 98.4, 11.9x), STELLAR (5.2 → 86.1, 16.6x), CRC (zero through 24-08, first material 24-10, now 17.2), meningioma (zero through 25-07, first appears 25-11, now 9.8), atezolizumab (4.0 → 19.7), Merck/belzutifan (zero before 24-10).

DECAYING — cabozantinib (80.2 peak 24-10 → 44.3, −45%), CABOMETYX (45.4 peak 25-05 → 14.8, the lowest reading since the first release in the series), patent (zero in each of the last four releases), MSN (zero in each of the last six).

STABLE — repurchase (high and constant, 14–39, the one thing management never stops saying), generic (2–9 throughout, no trend), royalt, Takeda.

SPIKE, EXPLAINED — Ipsen at 29.2 in the 24-08-06 release. That is the quarter of the $150.0m Ipsen commercial milestone. The mention series and the milestone-revenue table (§2) corroborate each other, which is the standard the method requires.

ARTIFACT, DISCARDED — NET was in the initial term set at 24–118 per 10k words and was removed: it collides with "net product revenues" and "net income" and its counts are almost entirely the homonym. This is exactly the word-boundary failure the reference warns about. neuroendocrine is the clean proxy and it is too noisy (0.0 → 15.1 → 4.9) to call a trend.

Open questions this generated, before a thesis existed

1. Why has patent been absent from four consecutive results releases, and MSN from six — the exact window in which the appeal deciding the company's terminal value was briefed, consolidated, settled with one defendant, and argued? Investigated: the litigation is disclosed in full in 10-K Note 12 and 10-Q Note 11, and the CAFC docket (No. 2025-1236) confirms argument on 2026-06-04. So it is not concealment, and this memo does not claim it is. The narrower read: management's prepared language has moved entirely to the second franchise while the first franchise's terminal date remains unresolved.
2. Why is CABOMETYX at a series low in the same release where zanzalintinib hits a series high? The 2026-05-05 release mentions the marketed product that generates 90.5% of revenue 14.8 times per 10k words and the unapproved one 98.4 times — a 6.6x ratio. In 2023-05 the ratio was 0.83x.
3. meningioma from nothing to 9.8 — corroborated: STELLAR-201 was initiated, per the 2026-05-05 release, in an indication with no approved systemic therapy.
4. repurchase never decays. Corroborated by the balance sheet: $430.4m in Q1 FY2026 alone, $2.16bn since March 2023, $1.5bn more authorised.

---

9. Reconciliation to the Tier-1 screen

Three findings for the scanner, in order of value:

1. Net cash is wrong again — the tenth consecutive name. Here the error is $396.7m on a name where marketable securities are split across current ($551.1m) and non-current ($649.1m) lines and there is no debt at all. The verified figure comes straight off one balance sheet.
2. op_margin_pct is computed on the fiscal year while revenue_ttm is computed on the trailing twelve months. Mixing bases understates the margin by 1.8pp here and will do so systematically on any name whose margin is moving.
3. The exit multiple is never sensitised, and it decides every verdict. Both screen runs varied the terminal margin (37.6% vs a 22.9% industry cap) and produced opposite verdicts, while leaving the exit multiple fixed at 20.6x — the parameter that moves the required CAGR from 30.9% to −0.8%. On this name the 20.6x anchor is not merely unsensitised, it is unidentified: a 542-name set matched on +12.9% trailing growth cannot bracket the exit-year growth of an asset whose patent estate expires eighteen months before the exit year.