CRNC — Cerence Inc. · 2026-09-07 · Verdict: WATCH · Conviction 3

Price $8.53 (screen row, universe_under2b.csv dated 2026-09-07; no live prices available) · Mkt cap $385.8M · EV ~$437M · EV/EBIT ~21x FY26 guided GAAP operating income · FCF yield 20.5% headline, ~14% ex-Samsung · Net debt ~$52M · ADV ~$5.9M Sources read: 10-K 2025-11-20 (Items 1, 1A, 7), 10-Q 2026-08-06, DEF 14A 2026-01-02, 8-Ks 2026-05-07, 2026-07-02, 2026-08-06, 2026-08-26, Form 4s (12m), Q3 FY26 release and slide deck 2026-08-06.

1. What the business actually does

Cerence sells white-label conversational and audio AI to carmakers: edge software licensed per vehicle head unit, cloud-connected services prepaid by the OEM for one to five years, and professional services for integration. Customers are nearly all major OEMs and tier-one suppliers, split roughly 50/50 by revenue, and about 52% of cars shipped in FY25 carried Cerence technology (10-K 2025, Item 1). Spun out of Nuance in 2019.

2. Why it is mispriced — the edge case

The cheapness is mostly an accounting artifact, and the screen errs in both directions. It carries cash at $0.00, publishing an EV of $559M; real cash was $127.6M at 30 June and principal debt is $170M after the August note repurchase, so true EV is about $437M.

The 12.1% FCF yield errs the other way. FY26 is guided to $76M-$82M of free cash flow (8-K 2026-08-06), a 20.5% yield. But the company's own footnotes say Q1 FY26 revenue included $49.5M of IP licence revenue from resolving patent litigation with Samsung against $20.8M of related expenses, and the 10-Q attributes most of the year's cash flow gain to "the net impact of the lump sum payment, net of taxes, received from Samsung and related legal expenses." Strip it out and FY26 revenue is about $262M with roughly $39M of adjusted EBITDA, not $66M-$70M. No forced seller, just a headline that will not repeat.

3. Unit economics and growth

Variable licence revenue, the true per-car royalty, ran $34.2M, $31.6M, $30.5M, $31.8M, $29.1M over five quarters; pro forma royalties (total value of licences shipped) ran $43.2M, $40.1M, $39.3M, $40.3M, $37.8M (8-K 2026-08-06, slides 5 and 6). Both decline. Units shipped fell 8.1% in Q3 while IHS global production fell 1.8%, and Cerence's share of world production is 50% against 52% in FY25. It is losing share into a flat market.

Connected services is genuinely good: $15.5M in Q3, up 20.4%, attach rate 34.1% versus 31.0%. Five-year backlog grew to $1,169.2M from $952.7M (10-K 2025, Item 1) and adjusted total billings were $240.0M TTM, up 6.3%. Professional services fell 17.7%.

Costs are where the case breaks. Non-GAAP operating expenses were $39.6M in Q3 FY25 and are guided to $46.6M-$47.6M in Q4 FY26. Nine-month R&D rose 18.1% to $84.2M and G&A rose 63.9% to $59.5M on $23.9M more in legal and professional fees tied to the Samsung agreement and IP enforcement (10-Q 2026-08-06). The cleanest comparison: Q3 FY25 had zero fixed licence revenue and delivered $62.2M of revenue and $9.0M of adjusted EBITDA; Q4 FY26 is guided to zero fixed licence revenue, $61M-$65M of revenue and $1M-$5M of adjusted EBITDA. Same revenue, less than half the profit, one year later.

4. Balance sheet and capital allocation

$127.6M cash and no covenants since the credit agreement was terminated in December 2024. Debt is $105M of 1.50% converts due July 2028 (conversion price $40.72) plus $75M of 2025 Modified Notes, both retired at discounts: $30M of principal for $28M of cash in the nine months, then $10M at 92.75% in August (8-K 2026-08-26). Value-adding.

The $30M buyback authorised 5 August is 7.8% of the cap, but FY26 stock compensation is guided at $27.2M and shares rose from 43,374k to 45,188k in nine months, so it offsets one year of dilution rather than shrinking the count, and none had been repurchased as of the 10-Q filing date. Directors and officers hold 1.39% (DEF 14A 2026-01-02). In twelve months there were zero open-market insider purchases and one sale, 25,998 shares by the Chief Administrative Officer at $8.83 four days after the buyback was announced (Form 4, 2026-08-12).

5. Management: what they said versus what they did

Free cash flow guidance was raised twice and Q3 adjusted EBITDA beat, but gross margin guidance was cut from 79%-80% to 78%-79% and the Q3 beat came partly from signing $12.5M of fixed licence contracts against "approximately $10 million" guided in May (8-K 2026-05-07). Those contracts book full value at signing and "typically provide the customer with a price discount and can include the conversion of a variable contract that is already in our variable backlog" (10-Q 2026-08-06), yet the FY25 10-K said management had decided "to limit the level of such contracts on a go-forward basis" (Item 7). The lever they said they would limit made Q3 look like growth.

6. Valuation

Normalised free cash flow sits well below the headline: $39M of ex-Samsung adjusted EBITDA less roughly $3M cash interest, $6M-$8M of normal cash taxes and $4M capex gives about $25M, since FY26 cash flow also absorbed a $10.4M deferred revenue build and a $12.0M release of prepaid assets (10-Q 2026-08-06). Reverse DCF: at $437M EV, a 10% discount rate and 3% terminal growth, the price implies about $30.6M of perpetual free cash flow, which at Q4's 74% gross margin and a held $188M opex base needs revenue near $295M, some 13% above the ex-Samsung run rate, held forever.

Base (50%): FY28 revenue $280M, connected services compounding mid-teens, variable licence flat, opex discipline returns, $42M adjusted EBITDA at 10x, net cash: about $9.75. Bear (30%): OEMs in-house their assistants on general-purpose LLMs (Item 1A), revenue $240M, adjusted EBITDA $14M, 1.0x revenue: about $4.00. Bull (20%): a second Samsung-scale IP settlement, revenue $310M, $60M adjusted EBITDA at 12x: about $17.70. Probability-weighted about $9.60, roughly 13% above $8.53.

7. Catalysts and timeline

Q4 FY26 results with first FY27 guidance and the long-term roadmap Krzanich promised for "next quarter" (8-K 2026-08-06), expected November 2026, settles the run-rate question. Then buyback execution, further convertible repurchases at discounts, and any second IP enforcement outcome.

8. Risks and pre-registered kill criteria

Main risks: OEM in-housing on general-purpose LLMs, share loss in a flat market, opex outrunning revenue. Kill criteria, all in company disclosure: 1. Variable licence revenue below $29.0M in two consecutive quarters (it was $29.1M in Q3 FY26, the low of the five disclosed). 2. Non-GAAP operating expenses above $46M in any FY27 quarter, making the Q4 FY26 step-up permanent rather than seasonal. 3. Percent of world auto production with Cerence technology (TTM KPI) below 48% in any quarter (52% FY25, 50% Q3 FY26). 4. Zero shares repurchased under the $30M authorisation as of the FY26 10-K.

9. Verdict and one-paragraph summary

WATCH, conviction 3. Cerence's 20% free cash flow yield is an artifact of resolving patent litigation with Samsung: $49.5M of one-time revenue against $20.8M of costs, roughly $29M of the $66M-$70M adjusted EBITDA guide. Take it out and you have a $262M revenue business earning about $39M of adjusted EBITDA and maybe $25M of sustainable free cash flow against a $437M enterprise value, which is fair rather than cheap. The tell is management's own Q4 guide of $61M-$65M of revenue and $1M-$5M of adjusted EBITDA against $62.2M and $9.0M in the comparable zero-fixed-licence quarter a year earlier, because non-GAAP opex went from $39.6M to a guided $46.6M-$47.6M while variable licence royalties fell from $34.2M to $29.1M and shipments fell 8.1% into a market down 1.8%. Connected services growing 20% and backlog up to $1.17B are real, but no insider has bought in twelve months, the only trade was a sale four days after the buyback was announced, and that $30M authorisation barely covers $27.2M of annual stock compensation. Probability-weighted value is about $9.60 against $8.53, so wait for the November print, where management must guide FY27 without a Samsung in it.

Research for discussion, not investment advice. Positions and sizing are the reader's decision.