Report contents
Industry landscape
Where the company competes: three adjacent voice markets, none of them concentrated
Industry structure: fragmented. SoundHound AI sells software that lets a machine hold a conversation: speech recognition, language understanding, voice synthesis, and an orchestration layer the company calls OASYS that routes each request to the right model or back-end system. Customers embed it in cars, restaurant drive-thru lanes and phone lines, call centers, and consumer devices. SoundHound gets paid through per-vehicle and per-device royalties, subscriptions, per-interaction fees and upfront license payments. The mix is moving: the CEO said on the November 5, 2026 earnings call that point-in-time license deals fell to about 10% of third-quarter 2026 revenue from more than 33% in the second quarter, leaving the rest recurring. That shift matters for the moat question because recurring revenue is the part that can compound, while a one-off license mostly proves a buyer showed up once.
Upstream, SoundHound buys cloud compute and can license third-party models, though management says its in-house models hold down cost per interaction. Neither input is scarce: several vendors sell comparable speech and language models, and open-weight models are free to download. Bargaining power sits downstream. Automotive OEMs qualify a voice stack over multiyear design cycles and routinely run more than one supplier, so they negotiate hard and control the roadmap. Restaurant chains and banks are smaller buyers individually but numerous, and they often arrive through systems integrators and telephony vendors who own the account relationship.
Geography splits the arena. Automotive voice is global with a short list of buyers, because one platform ships across markets and a design win travels with the vehicle program. Enterprise voice is national in practice: telephony, payment, menu and privacy rules differ by country, and a deployment tuned for US English drive-thru ordering needs new engineering for Germany or Japan. The boundary there is set by integration work, not by where demand exists.
The submarkets differ enough that a single share figure would mislead:
| Submarket | Who SoundHound meets there | Structure |
|---|---|---|
| Automotive embedded voice | Cerence, Google, Apple, Amazon | Few credible suppliers, long qualification |
| Restaurant and retail voice ordering | In-house builds, voice-ordering specialists, systems built on rented models | Dozens of entrants |
| Contact center and agentic service | Microsoft (Nuance), Salesforce, Google Cloud, Amazon, Sierra, Decagon | Crowded and consolidating |
| Speech components | ElevenLabs, Deepgram and similar API vendors | Price-competitive, commoditized |
Automotive carries the real entry barriers: functional-safety requirements, acoustic tuning for each language, and OEM engineering relationships measured in years. Cerence is the closest pure comparison because it sells almost nothing but automotive voice software, and Google and Apple ship their own assistants onto the same dashboards, which caps what any third party can charge per vehicle. In enterprise service, the technology barrier is close to zero; a competitor can assemble a working voice agent from purchased models in weeks. What protects revenue there is what happens after signing. The agent gets wired into the customer's phone system, CRM, menu database and payment flow, and retuning it is a project. Switching costs therefore attach account by account, not to the market.
Run the practical test and the two halves split cleanly. Enterprise voice service fields far more than a handful of credible vendors, so entry barriers at the market level are low and no supplier sets price. Automotive fields few enough to count, but incumbency is divided among several strong players, and nothing in the public record establishes how much of the design-win pool SoundHound holds.
The roster itself is partly assembled. The CEO said on the same call that six companies have been merged under SoundHound AI, and the company is folding LivePerson's messaging platform into its own with full integration targeted for early 2027. That gives SoundHound an entry into digital customer service, a market with different rivals and different buying behavior, but it also means the customer list was bought as much as won. Whether cross-selling turns those accounts into a durable position is the open question, and it will show up in retention before it shows up in share.
The competitive arena, then, is a set of accounts rather than a market with a leader. SoundHound looks strongest where qualification cycles are long and the integration is deep: automotive programs, large restaurant chains, utilities. It looks weakest where a rival can rebuild the product from rented models in a quarter: generic customer-service automation, where it now also meets Microsoft and Salesforce. Evidence of a rising, dated share of automotive programs, or enterprise gross retention above 90% sustained across several quarters, would show the account-level position compounding into market position.
Competitive advantage
Does the company have a moat? Not yet: volume is rising while gross margin falls
Overall rating: Weak. The loss is not the decisive fact. The pairing is. SoundHound reported revenue of $62.4 million in the September 2026 quarter, up 48.5% year over year, while GAAP gross margin fell to 34.1% from 42.6%, and its own non-GAAP measure dropped 9.0 percentage points to 50.3%. A protected franchise turns volume into better unit economics, because the cost of building the product spreads across more dollars and the price holds. This one is adding revenue at a lower margin per dollar. Some of that is deliberate: the CFO said the company is converting upfront license deals into hosted contracts and exiting business lines he described as dilutive, both of which move margin in the near term. Whether the hosted base eventually earns more than the licenses it replaced is exactly the question the record does not yet answer, and so far the direction is unfavorable.
Market structure reinforces the skepticism. Automotive embedded voice is concentrated among a handful of suppliers, but SoundHound is not demonstrably the largest of them; contact-center AI is crowded and consolidating; speech components are sold on price. Concentration that does not favor the company provides no shelter. Excess returns are the second test, and here the record is unambiguous: across FY2021 through FY2025 and the first nine months of FY2026, there is no year of positive operating income to protect.
Market position: qualified at large buyers, but no measured lead
Market position and ranking: Moderate. No dated share or ranking figure establishes leadership in any of SoundHound's three main arenas. What the record does show is a supplier that clears technical review at large buyers. In the third quarter of 2026 it signed a seven-figure minimum-guarantee agreement with a tier-one Chinese automotive supplier for voice and on-device work, took a commerce channel across roughly 20 million US smart TVs with LG, and added SoundHound ordering at nearly 100 more locations for a pizza chain, the company said in its November 5, 2026 release. Minimum guarantees and multi-year programs are what a supplier wins after passing qualification. They are not evidence that it sets price.
Stability and trend: Moderate. Share stability cannot be confirmed: no comparable, dated series covers three or more years for these submarkets, so the revenue mix has moved faster than any share measure could track. The direction of travel in account count is up. The direction in margin is down, partly by the company's own design. Expansion in the number of customers, without a price or margin gain, is growth, not proof of position.
Profitability: no positive operating income in any covered year
Returns on capital: Weak. Reported operating income is negative in every period from FY2021 through the first nine months of FY2026, which makes estimated after-tax ROIC negative throughout, since NOPAT is operating income times one minus the tax rate and a negative numerator cannot clear a positive hurdle. Invested capital at September 30, 2026 was about $0.61 billion (equity $0.81 billion plus debt $0.01 billion less cash $0.21 billion), all of it earning a negative return. ROE and ROA are not usable substitutes here: equity was $0.03 billion at the end of FY2023, which is why reported ROE reads 2,118% that year and −333% in FY2024. Those figures describe a thin balance sheet, not earning power.
The single quarter of positive GAAP operating income in the series, Q4 2025, came with an operating cash outflow of $0.02 billion, so it was not a cash profit. In Q3 2026 the reported GAAP operating loss of $46.5 million was struck after a $23.4 million fair-value gain on contingent acquisition liabilities, the company disclosed; the underlying operating result was worse than the headline. Adjusted EBITDA was a loss of $20.7 million, and management guides to sustained adjusted EBITDA profitability only within the second half of 2027.
Margins and scale: Weak. Gross margin went from 75.4% in FY2023 to 42.4% in FY2025 and 34.1% in the September 2026 quarter; net margin was −74.3% in that quarter. Revenue over the same span rose from $0.17 billion in FY2025, itself roughly eight times the $0.02 billion booked in FY2021, and the first nine months of FY2026 already total about $0.16 billion. Scale arrived. Margins fell anyway. The year-over-year gross margin drop came with less than four weeks of LivePerson in the numbers, so it reflects the existing mix, not the newly acquired business.
Overall rating and core rationale
Overall rating: Weak. Profitability carries the most weight, and it fails on every measure that matters: negative ROIC in each covered period, negative operating cash flow in every year from FY2021 through FY2025 and again in the first nine months of FY2026, and a gross margin that has roughly halved in three years while revenue compounded. The company holds a real but narrow asset: long qualification cycles and deep deployment work at specific accounts. That shows up as renewals and expansions, not as pricing power, and it is too small to carry the rest of the business. The rating is not lower than Weak because the automotive and enterprise deployments do create genuine friction for a customer considering a switch, and repeat business at named accounts is visible in the record.
Where the advantage comes from and how long it can last
Three sources could plausibly protect SoundHound: proprietary models that lower cost per interaction, customer captivity inside each deployment, and scale across a shared platform. Only the middle one has support in the record, and it is partial. The single number that tests all three is gross margin on recurring revenue, because it captures whether in-house models and a common platform actually lower the cost of serving each incremental interaction. If the advantage is forming, that line recovers as hosted volume grows and integration costs from the acquisitions roll off. If it stalls near 50% or keeps sliding, there is no cost edge to defend.
Supply-side advantage: no proprietary model, and the margin line shows it — Weak
SoundHound's pitch rests on in-house speech and language models plus OASYS, the layer that routes each request to the right model or back-end system. That is a cost claim, not a barrier. Comparable models are widely sold, and open-weight versions are free to download, so a rival can assemble a working stack without licensing anything from SoundHound. The financial signature of a real cost advantage would be a gross margin that holds or improves as hosted volume grows. Instead the margin has compressed for three straight years as revenue scaled, and hosted delivery makes that mechanical: every hosted dollar carries a cloud compute bill that a license did not, and OASYS routes some traffic to third-party models whose owners set their own price.
The size of the gap is worth putting in dollars. At September 2026 revenue of $62.4 million a quarter, one percentage point of gross margin is about $0.6 million; the 8.5-point year-over-year decline equals roughly $5.3 million of quarterly gross profit that the current mix does not deliver at last year's rate, an estimate from reported figures. Spread that across four quarters and it is more than a fifth of the adjusted EBITDA loss management is trying to close. Cost per interaction may well fall with better models. What would break this advantage entirely is a model vendor or cloud provider cutting speech pricing or bundling it into committed-spend discounts, which would erase the cost gap without SoundHound losing a single customer.
Demand-side advantage: switching costs sit in each deployment, not in the market — Moderate
Customers stay for operational reasons rather than affection for the product. A voice agent inside a drive-thru lane, a hospital phone line or an appliance is wired into the buyer's menu database, CRM, payment flow and telephony, and retuning it is a project with real risk: an order-taking system that mishears customers loses sales, so nobody swaps a working deployment to save a little money. The proof shows up as scope creep within accounts rather than new logos. Resorts World Las Vegas moved from an on-property digital concierge to an external booking agent, a British multinational telecom added services, and a European broadcaster and a UK water utility renewed, all named in the November 5, 2026 release. Each of those is a buyer deepening a deployment it already had.
The protection is narrower than it looks. The switching cost attaches to the deployment, not to the technology, and the same integrator who installed SoundHound can install a competitor at the next site. Large buyers routinely run more than one voice supplier, particularly automotive OEMs that ship several assistants on one dashboard. Captivity only becomes a moat when it produces pricing power, meaning a rival's discount fails to take the account. The margin record does not show that, though the shift to hosted contracts confounds the comparison. The clear adverse outcome to watch is a buyer's existing software vendor bundling voice agents into seats the customer already pays for. That flips the switching cost against SoundHound, because ripping out the incumbent platform costs more than adding its voice feature.
Economies of scale: a shared platform on too small a base — Weak
The fixed costs are real: several model families, language coverage for each market, and integration work per vertical. Spreading them is the stated logic behind consolidating six businesses onto one platform, and it is the main reason to think margin could improve. The obstacle is relative size. SoundHound booked $0.17 billion of revenue in FY2025 against competitors for whom voice is a feature attached to a product the customer already owns: Google, Apple and Amazon place assistants on devices at no separate charge, and Microsoft and Salesforce sell service software by the seat, with voice folded in. Those rivals do not need voice to earn a standalone return, which is the asymmetry that matters. Scale protects a supplier only when customers are also captive; without that, an entrant buys the same models and reaches similar unit costs at far lower volume.
Durability: the substitute is the buyer's own vendor list
The threat that ends this business is not a better voice startup. It is voice becoming a standard line item in contracts the customer has already signed, which turns a specialist purchase into a feature upgrade. Rented models have improved fast enough that a competent team can now assemble a usable agent in weeks, so the technical hurdle that justified hiring SoundHound keeps falling. What survives that shift is the integration work inside each account, and it survives only as long as the buyer's core software vendor does not do the integration itself. That is the mechanism to follow, along with the gross margin line: if non-GAAP gross margin on recurring revenue does not move up as hosted volume scales, the cost advantage is not there, and the account-level friction will not be enough to earn a return above the cost of capital.
Generated by DeepVal AI. Data and models may contain errors or be outdated. For research only; not personalized investment advice.
Continue reading the full report
- Comparison and summaryHow SoundHound AI Inc compares with its closest rivals, and our moat rating
Free to sign up · 20 research points · unlocking this report uses 10
20 points cover 2 full reports. No card needed.
Research reflects its generation date. Preview the first two modules.
More on SoundHound AI Inc
SoundHound AI Inc
Business, moat and key questions
Open company page