Innoviva, Inc. INVA

Innoviva, Inc. (NASDAQ: INVA) — deep-value analysis (REFRESH)

Analyst: Claude (deep-value) · Date: September 8, 2026 · Reference price: … (tracker, 09/07/2026) · Market cap: … · Shares: 72.637 mil. (82.6 mil. fully diluted)

Regime: REFRESH over the thesis from 17.08.2026 (rapoarte\deep\2026-08-17-deep-INVA.md), triggered by the deep_delta.py triage (1 unverifiable falsifier: IST growth). Diff package: rapoarte\deep\delta-INVA-20260907.md.

What I inherit unchanged from the reference thesis, because no new filing moves it: the description of the three business engines and the moat analysis; the Entasis/La Jolla acquisition history and the capital structure; the buyback pattern; the debt structure and maturity schedule; the CEO profile and Thorndike grid; the Armata non-consolidation mechanics under ASC 810. I recap these condensed, with figures updated to 06/30/2026.

What I re-derived, and why. Three things, all from filings the August run did NOT have locally:

1. 10-Q Q1 2026 (filed 05/06/2026) and 10-Q Q2 2026 (filed 08/05/2026) — the EDGAR downloader only pulls quarters of the closed fiscal year, so in August the thesis was built on XBRL companyfacts and press releases. I pulled them manually from EDGAR (INVA_10Q_2026-05-06.htm/.txt, INVA_10Q_2026-08-05.htm/.txt) and indexed them with filing_text.py. They contain the per-product royalty breakdown, which overturns the central decline assumption, and the item-by-item Armata marks, which were estimated in August.
2. The broken falsifier — IST growth — now verifiable over two consecutive quarters. Holds, but barely.
3. The Monte Carlo simulation — the mc-INVA-20260907.json file produced by the mechanical triage was off by three orders of magnitude (--oe … in thousands with --sh 72.6 in millions → median IV … instead of 35.8; MOS …). I redid it with consistent units AND revised assumptions: mc-INVA-20260908.json.

Primary sources: 10-K FY2025 (SEC-Filings\INVA\INVA_10K_2026-02-25.txt), 10-K FY2024/2023/2022, 10-Q Q1-Q3 2025, 10-Q Q1 2026 and Q2 2026 (new in the file), XBRL companyfacts CIK 0001080014. Data pack: rapoarte\deep\data-pack-INVA-20260908.md. Brief: rapoarte\deep\research-INVA-20260908.md. Prior GBL score … (evaluated 09/06/2026).

---

Executive summary (1 page: thesis, estimated value, verdict)

The structural thesis from August remains valid and I don't rewrite it: Innoviva isn't a pharmaceutical company, it's a capital-allocation vehicle built on top of a royalty flow with a known expiration date, run by a former activist investor, who uses the royalty money to fund a hospital anti-infective platform that still loses money on its own and an illiquid stakes portfolio dominated by Armata Pharmaceuticals. What changed is the calibration of the two variables that determine value, and both moved the wrong way.

First change: the royalty isn't declining 2… a year, as I assumed in August. It's declining 8…. The Q2 2026 10-Q, which the prior run didn't have, gives the exact breakdown (l.812-817): gross RELVAR/BREO royalty … in Q2 2026 (47,655 versus 54,737 thousand) and … for the half-year (94,931 versus 105,627); ANORO … in the quarter, … for the half; total gross royalty … in the quarter and … for the half-year (118,413 versus 128,599 thousand). The company's explanation, verbatim: "lower net sales driven by pricing pressures in the United States" (l.819). The research brief independently reported, from GSK's own reporting, RELVAR/BREO … and ANORO … — figures that match the filing, so the secondary source confirms the primary one. In August I had modeled … a year through the patent's expiry in …; reality is three times faster, and the cause (US price pressure, not loss of exclusivity to a generic) has no reason to stop before the cliff.

Second change: the Armata mark isn't 45% above Armata's market cap, it's 90%. In August I'd estimated the exposure at … from the December structure; the 10-Q gives it exactly (l.429-433) and the estimate was correct to the decimal: shares 162,498 + warrants 57,044 + convertible note 105,036 + term loans 133,108 = … What I didn't know was the price embedded in the mark: the 25,076,769 shares marked at … imply …/share at 06/30/2026, not the … I'd used from August's quote. At … and a 67.5% stake, Armata's total market cap is ~… — and Innoviva marks its exposure at …, i.e. 190% of the entire company's market value that it funds. Additionally: term loans with … of principal marked at … (… over par, to a debtor with a going-concern warning) and a …-principal convertible note marked at … (3.5× par).

The flow figures, the only ones that aren't opinion. TTM CFO (Q3'25-Q2'26) …; H1 2026 …, down 5.9% versus … in H1 2025, on revenue growing 15.2% (217.585 versus 188.915 mil). Operating income for the half-year: …, … year over year. Fifteen percent of sales growth produced declining operating profit. TTM owner earnings, per the bridge in the valuation chapter: …, versus … on the FY2025 base used in August — a 9.1% decrease, exactly the magnitude of the royalty erosion.

The falsifier that triggered the revisit holds, but barely. Management's stated target is a minimum of … of US net IST sales in 2026. I now have both quarters: Q1 2026 … (… YoY), Q2 2026 … (…), half-year … (…). Hitting 150 mil requires … in H2; H2 2025 was …, so at … it would give … The target is hit only if the pace holds exactly — neither missed nor comfortable.

Estimated value. Six triangulated models give a range of … – …, with a median MOS of ** …**. The central model I support — sum-of-the-parts with the royalty declining 6% a year and expiring mid-…, the portfolio at a 43% haircut, IST at 2.5× real revenue — gives …/share, MOS …, versus … and … in August. Monte Carlo (20,000 scenarios, revised assumptions) gives a median of … MOS …, with P10 at …. My weighted estimate (65% on the pessimistic-realistic group, 35% on the optimistic one): ~… MOS ~ ….

Verdict: DON'T BUY AT … — MONITOR, with an entry threshold below … A downgrade from "SMALL BUY" in August, and the reason is explicit: the base-scenario safety margin fell from … to …, and the two things that lowered it are new facts from the filings, not changed opinions. At … MOS in the central scenario, with … of market cap sitting in cash and Level 3 marks I can't independently verify, and with the profit engine eroding three times faster than I thought, there's no ticket worth buying today. Below … the same assumptions give … MOS and the discussion becomes interesting again. The two triggers to watch at Q3 2026 (reporting 11/04/2026): quarterly gross royalty — under … confirms the 8%+ decline — and trade receivables — over … breaks the mix explanation.

---