AudioCodes: $87 Million of Recurring Revenue Wrapped Around a 4% Growth Company
VoiceAI compounds at 40-50% and management retired 6% of the shares in a quarter — but the margin line thinned, the Q&A thinned, and two of last year's loudest talking points simply vanished.
AUDC · Earnings Call · 2026-11-04
Two companies under one ticker
AudioCodes is a $250 million market cap voice-infrastructure company that is trying to become a voice-AI company, and the third quarter shows both halves at once. Revenue was “$64.1 million an increase of 4.2% over the $61.5 million reported in the third quarter of last year” — Niran Baruch, Vice President of Finance and Chief Financial Officer · 2026-11-04, with services at $37.1 million, up 19.9%, now 57.8% of the total. GAAP gross margin held at 65.7%. Then it sours: non-GAAP operating income fell to $4.8 million (7.4% of revenue) from $5.8 million (9.5%), and non-GAAP EPS slipped to $0.16 from $0.17.
That combination — revenue accelerating, profit shrinking — is the fingerprint of a company paying for its second act out of the first act's cash flow. Management wants you to look at the other number: “By the end of third quarter 2 thousand 26, they jointly represented $87 million in annual recurring revenues, a 16% increase from the prior year” — Shabtai Adlersberg, President and Chief Executive Officer · 2026-11-04, with ARR up almost 90% over three years.
The problem is arithmetic. VoiceAI is guided to $25 million in 2026 against full-year revenue guidance of $251-256 million — roughly a tenth of the company. Even hitting the $50 million 2028 target at the high end of the stated 40-50% growth corridor adds about $25 million over two years to a $250 million base. The growth engine is genuine; the denominator is the legacy business, and the legacy business is what determines the stock.
The orchestration bet, and its first marquee reference
Large enterprises prefer a best of breed approach meaning they would like to choose a different LLM that is adapted to their, specific needs, a different speech to text, and a few more. ... You would like to be able to switch to that new LLM in a matter of simple development.
That is the whole thesis: AudioCodes does not build models, it builds the layer that wires enterprise telephony — PSTN, SIP, PBX, UCaaS, CCaaS, WebRTC — into whichever model a customer picks this quarter. The validation moment is Deutsche Telekom, which selected the company to power a mobile AI-notes service for business customers, with Mistral AI providing the model. A network operator publicly naming a French model vendor and an Israeli plumbing vendor, rather than its own stack, is a decent proof point for the neutrality argument. VoiceAI Connect bookings rose roughly 60% over the first nine months.
Worth holding against the tape: the global market's editor-curated keyword list for this quarter contains no voice AI, no agent infrastructure, no orchestration — it is IRB models, Fed rate hikes, core deposits, Grok bots. Not one of the market's 360-day or 30-day price movers touches this theme. That cuts both ways. AudioCodes is not riding a wave, so there is no tide to lift it; and there is no crowding either. The nearest shared theme among the companies that reported in the last five days is personal agent from NICE — a contact-center platform vendor selling precisely the bundled, do-it-all model that AudioCodes' answer argues large enterprises reject.
What management stopped saying
Two things went quiet, and both are more informative than anything in the prepared remarks.
The first is tariffs. In February, the CFO put a number on it: “The tariff was about $2.7 million in 2025. We currently estimate it to be a bit lower, you know, probably around $2.3 million in '26.” — Shabtai Adlersberg, President and Chief Executive Officer · 2026-02-03 A guided, quantifiable cost line. This quarter it is not mentioned in the prepared remarks, not asked about in Q&A, and not visible in the keyword set at all — where "tariff rate" had once ranked near the top of the company's own themes. Either it resolved or it became uninteresting; either way, a known headwind stopped being narrated.
The second is bigger. In July 2025, on the Cisco Webex certification, management said: “in terms of revenue, we won't see much impact in '25. But I can tell you that we see a lot of interest ... we do expect to see a rise, but the majority of it will come in '26.” — Shabtai Adlersberg, President and Chief Executive Officer · 2025-07-29 Cisco Webex was a top-five keyword for this company as recently as the second quarter of 2025, and still ranked in the first quarter of 2026. We are now three quarters into the year the majority was supposed to arrive, and Webex does not appear once. Nor does it appear in the current keyword set. That is the question I would want asked next call.
There is a subtler rotation in the vocabulary, too. Last quarter's second-highest keyword was Voice Agent, third was AI Receptionist, with Voca CIC prominently featured. All three fell out of this quarter's leading terms, replaced by use-case language: Meeting Insights, real time translation, MIA Edge, Voca Connect. Meeting volume on the Cloud Edition rose 90% year over year and its recurring revenue more than doubled. The company is now leading with narrow, deployed, defensible things rather than a sweeping agent platform. That could be maturity. It could also be that the agents were easier to name than to sell.
The genuinely novel line
The most company-unique item in the entire call is an air-gapped one. MIA Edge — meeting intelligence that runs with no access to the public internet — added 11 new customers in a single quarter across defense, healthcare, security, infrastructure and government in Israel. It shipped a real-time translation application for asylum-seeker interviews, adopted by the state population and immigration authority, with two comparable European authorities evaluating. Asked how fast regulated buyers move, the CEO did not hedge: “the need is very urgent. Especially when you are talking about defense applications.” — Shabtai Adlersberg, President and Chief Executive Officer · 2026-11-04 Data sovereignty, defense use cases and closed-network AI are nowhere in the global keyword set and nowhere in last week's reporting cohort. If there is a franchise in this name that the market has not priced, it is more likely here than in another contact-center seat.
Thin Q&A, thick buyback
Only one analyst asked questions — two, both from Needham — against prior calls where the same voice probed pricing power, the 40-50% CAGR definition, and tariff exposure. On a $250 million cap, one questioner is what the coverage looks like, and it is a fair signal that the equity story has not yet been re-rated by anyone.
What management did with the silence is telling. It bought 1.5 million shares for about $15.1 million in the quarter — roughly 6% of the outstanding company, at an implied price near $10 — under a court authorization in Israel for up to $30 million through March 9, 2027, which can also fund dividends. It paid a $0.20 semiannual dividend worth about $4.8 million, generated $4.1 million of operating cash, and finished September with $47.1 million across cash, deposits and investments. The whole company trades at roughly one times the midpoint of its own revenue guidance.
The setup, then, is a cash-generative legacy franchise — SBC tracking above $135 million for a record year, gateways near $50 million, Microsoft at $43 million in the quarter and on pace for more than $170 million — funding a $25 million option on the orchestration layer of enterprise voice. Management is buying its own stock rather than telegraphing a reacceleration. That is not a growth story being told; it is a waiting story being paid for.