INSP — Inspire Medical Systems, Inc.

Deep-value report: INSP — Inspire Medical Systems, Inc.

Date: 09/03/2026 · Price: … (NYSE) · Market cap: … · EV (own calculation): … Sector: Healthcare / Medical Devices (implantable neurostimulation for obstructive sleep apnea) · Currency: reporting USD, quoted USD — no conversion, no ADR, no ADS/ordinary ratio Regime: REFRESH — resuming the 07/26/2026 analysis (verdict then: INTERESTING-SPECULATIVE), triggered by the mechanical deep_delta.py triage: the roe falsifier was flagged BROKEN (ROE ttm 16.4% > the 10% threshold).

What I carry over from the 07/26/2026 report, unchanged and not re-verified: the business description and moat structure (patents, clinical data, reimbursement infrastructure, proctoring network), the Q1 2026 guidance-cut chronology, the mechanics of the WISeR program and the CPT coding transition, the GLP-1 argument, and the FY2023-FY2024 figures from filings already read. What I fully re-derive: the broken falsifier (ROE), the owner-earnings base, the FCF bridge, capital allocation (now with real average buyback prices, not estimates), O’Glove-style earnings quality across 5 quarters and 3 years, the CEO profile, legal exposure, and the entire valuation triangulation.

New filings read for this refresh (didn’t exist in the local file at 07/26/2026): Q2 2026 10-Q (filed 08/03/2026, accession 0001609550-26-000047) and Q1 2026 10-Q (filed 05/04/2026, accession 0001609550-26-000023) — downloaded and converted to text in this session; the 2026 DEF 14A (filed 03/20/2026) — the proxy that wasn’t available in the previous analysis and which resolves three gaps explicitly flagged then (CFO identity, CEO compensation, named insider ownership).

Executive summary

The thesis in brief: the falsifier broke for accounting reasons, not economic ones — the underlying thesis remains intact and has hardened. The triage flagged a TTM ROE of 16.4% (today’s data pack gives 18.0%), above the 10% threshold that would have invalidated the claim “capital allocation is weak, normalized ROE ~7.6%, below the cost of capital.” The filing check shows the entire jump comes from a single non-cash event: on 12/31/2025 Inspire released its deferred-tax-asset valuation allowance and booked an …lion deferred tax benefit, out of a total FY2025 tax benefit of …lion (FY2025 10-K, Item 7 and Note 7 — “recognized a non-recurring tax benefit of …lion”). FY2025 GAAP net income of …lion thus contains ~…lion that will never recur. Normalized ROE = (145.4 − 88.8) / average equity 735.4 = 7.7% — practically identical to the 7.6% in the original thesis. The falsifier wasn’t refuted; it was tricked by an XBRL taxonomy artifact.

What actually changed since 07/26/2026 — in order of importance: (1) Project Horizon, a restructuring announced 08/03/2026, …-25 million pre-tax cost, ~…lion of annualized capacity that gets reinvested into growth, not left in margin (Q2 2026 10-Q, “Recent Developments”); (2) the class action expanded materially — on 07/30/2026 plaintiffs filed an amended complaint adding fraud allegations about Medicare/CPT coding and extending the class period to 11/04/2024 – 05/04/2026; alongside it are 3 derivative suits, a DOJ civil investigative demand (CID) under the False Claims Act dated 01/17/2025 concerning marketing and reimbursement practices, and a qui tam action where the government declined to intervene but the plaintiff continues (Note 9, Q2 2026 10-Q); (3) the effective tax rate turned punitive — 299.2% in H1 2026, from stock-compensation tax “shortfalls” and 162(m) limitation, so the half closed with a net loss of …lion on pre-tax income of …lion; (4) zero buybacks in H1 2026, with …lion authorized and available, after …lion spent at an average price of …/share; (5) on 04/14/2026 the PTAB denied all three of Nyxoah’s IPR petitions against Inspire’s patents — the one clearly good piece of news in the period, and one that reinforces the IP moat; (6) a permanent CFO was appointed (Matthew J. Osberg, effective 02/17/2026), closing a governance signal flagged “to monitor” in July.

Estimated value. The FCF bridge, redone with XBRL tags, shows why this company’s valuation splits in two: CFO − capex gives a “simple” FY2025 FCF of …lion (yield 4.3% on market cap, 5.5% on EV), but that number exists only because of a …lion stock-based compensation add-back. Subtracting SBC — a real economic cost that transfers ownership — the business produced −…lion in 2025, −…lion in 2024 and −…lion in 2023. The correct owner-earnings base isn’t FCF, it’s normalized after-tax EBIT (EBIT already expenses SBC): …lion, stated in DEEPFCF. Triangulating across 5 models gives a value range of … – …/share, with a median of … (median MOS …), and a Monte Carlo with 20,000 scenarios gives a probability of undervaluation of 0.1%. Only the models that don’t rest on realized earnings — operating margin recovering to 14% by 2030, or a 2.0× EV/Sales multiple — clear the price, and both are assumption beds, not observations.

Verdict: AVOID (a downgrade from INTERESTING-SPECULATIVE, 07/26/2026). Not because the price rose — the move from … to … is … in 38 days and is context, not the reason for the refresh. But because three things became visible in the numbers: the owner-earnings base is much thinner than the FCF yield published by screeners suggests; finished-goods inventory reached …lion (77.6% of inventory, DIO 534 days) while revenue falls 7.6% year over year; and the board paid a MIP bonus at 111.4% of target for 2025, the year the stock lost ~65%, while completely halting buybacks exactly when the stock got twice as cheap. The moat is real and was just confirmed by the PTAB; the price doesn’t reflect it with a margin of safety — it reflects it with a premium.