Funding Landscape

How a product that records a meeting and hands back a transcript, a summary and action items actually gets paid, rail by rail, and which of those rails anyone in this category has ever been paid through. There is no insurer, no employer benefit and no reimbursement code anywhere in this field, so every dollar in it comes from a user, from a company buying seats, from a platform that bundles the feature into something already sold, or from an investor.

Abbreviations

Funding Mechanism Primer

How each distinct type of payment mechanism in this landscape works, in the abstract, before any specific rail uses it. None of these paragraphs is about any particular product.

App-store in-app purchase. The application sells a subscription inside itself, and the store, not the developer, is the merchant: it collects the money, owns the billing relationship and the renewal, handles refunds and tax, and remits the balance after a commission. The developer never sees a card number. The commission is set unilaterally by the store operator in a developer agreement, varies by the seller's annual earnings and by how long a subscription has been running, and can be changed by the operator or by a court. Access is a one-time developer enrolment plus per-release review; the rate is not negotiated.

Direct off-store subscription. The developer bills the customer itself, on the web, through a payment processor or a merchant of record that charges a percentage plus a fixed fee per transaction. There is no store commission on the transaction itself, though a store may claim a commission on sales that follow a link out of its application, within a stated window. The developer now owns the renewal, the dunning and the tax filing. Access is immediate and requires no permission from anyone, but acquisition no longer arrives through store search, and the buyer has to be brought to a web page.

One-time purchase. The customer pays once and keeps the product. No renewal exists, so no churn exists, and the seller's revenue in any month depends entirely on new buyers. This mechanism only survives where the seller's cost of serving an existing customer is near zero, because a perpetual price cannot fund a bill that recurs. It is common where the compute runs on the buyer's own hardware, and it is also how a hardware premium can carry a service that would otherwise be metered.

Seat-based invoicing to an organisation. A company, not a person, is the customer. The vendor signs a contract, issues an invoice, and is paid by accounts payable against a purchase order, with no store in the path and no commission. The price is per user per month, and revenue grows by adding seats inside accounts already won. Getting in means passing a procurement and security review before any money moves: an audited security report, a data processing agreement, stated data residency, and corporate login. The buyer is an information technology, security or finance function, not the person who will use the product.

Usage-metered platform licensing. Instead of selling a finished application to an end user, the seller meters its pipeline and charges other developers per unit consumed, typically per minute or per hour of audio, with per-feature upcharges. The buyer is an engineering team, the contract is a self-serve developer agreement with an enterprise tier above it, and the product competes on price, latency and accuracy rather than on interface. Revenue is proportional to volume with no flat cap, which inverts the risk of a flat subscription: heavy usage is good for the seller here and bad for the seller in a flat-rate model.

Hardware sale with an attached subscription. The customer buys a physical object at a retail margin, and the object is the entry point to a recurring software plan sold separately. Cash arrives in full at the point of sale, before any service is delivered, which finances the operation without outside capital. The object also acts as a sunk cost that the buyer has already paid, which vendors treat as a retention mechanism. The channel is retail and marketplace distribution, and the rail requires manufacturing, inventory, returns handling and a supply chain.

Reward crowdfunding pre-orders. Backers pay in advance on a campaign platform for a product that does not exist yet, and the platform takes a fee and releases the funds when the campaign closes above its target. It is not investment: backers receive the product, not equity, and have no claim if it never ships. For the seller it is simultaneously a capital source, a market test with a public number attached, and a pre-committed first cohort. Platforms structure this around physical goods, and the public totals are the only audited-looking demand figures a private hardware category ever produces.

Bundled inclusion. The capability ships inside something the customer already pays for: a productivity suite, a communications platform, a mobile network plan, or the operating system itself. No purchase decision is made about the capability at all, and no line item appears for it. The economics belong to whoever owns the container: the feature is funded as a retention and lock-in expense, its acquisition cost is zero, and it never appears on a comparison page. A third party can only reach this rail by being distributed by the container's owner.

Venture equity. Investors buy newly issued shares at a negotiated valuation, in named rounds, and the proceeds fund spending ahead of revenue. The money is not repaid; the return comes from a later sale of the shares. Access is a negotiation, not an application, and the price is set by comparable transactions rather than by the business's current cash flows. Rounds and valuations become public through press announcements and private-market trackers rather than through filings, which is why almost every number of this kind is an estimate carrying a tracker's name.

Retail equity crowdfunding. A company offers shares to the general public through a regulated online investment portal, taking small amounts from a large number of individual investors, with the portal handling the offering mechanics and the disclosure. It is usually run alongside a conventional round at the same valuation, and its second function is to convert users into shareholders. The mechanism carries reporting obligations to a securities regulator and a cap on how much may be raised, both set by the exemption relied on.

Trade sale to a strategic acquirer. An operating company buys the whole business, its assets or its team, usually to acquire a capability faster than it could build one. Terms may be disclosed or not, and undisclosed terms carry no implication either way. This is the mechanism through which venture capital in a category is actually returned when no public listing happens, and the identity of the buyers in a category is itself information: it says who thinks the capability belongs inside a larger product.

Vendor credit programmes. Infrastructure suppliers give qualifying young companies a fixed quantity of service credit, valid for a stated period, which offsets a cost rather than producing revenue. It is not dilutive and nothing is repaid, but it expires, it applies to one supplier's services only, and consuming it builds a dependency on that supplier's prices after it runs out. Access is an application against published eligibility criteria, often routed through an accelerator, an investor or a partner programme.

Public innovation grants. A government body pays part of a company's research and development cost against an approved work plan, on a published call with fixed deadlines, competitive evaluation and reporting obligations. The money funds the work described in the application and nothing else, is paid against evidence of spending, and typically requires the recipient to be established in the awarding jurisdiction. Some programmes fund a share of an approved budget; others award a fixed sum in phases. The administrative burden is real and is paid before any award is known.

Field Map

There is no payer layer in this field, and that absence is the first thing to understand about it. No insurer reimburses a meeting summary. No employer benefit plan covers one. No billing code exists for it, no regulator classifies the product, and no clinical or safety pathway gates it. A reader arriving from a healthcare or regulated field will look for the reimbursement section and there is none to find, because the product is consumer software rather than a regulated medical device: this project's own contract records its regulatory class as not applicable, and the binding legal surface is recording-consent law, which constrains the product without classifying it. Every dollar in this category therefore travels a commercial rail, and every entrant pays its own acquisition cost with no subsidy in front of it.

Who actually pays today, in order of how much money visibly moves. A person pays a subscription, at 8 to 19 US dollars a month for an individual tier across the shelf, a spread that separates the two billing cadences rather than two products: its top is a month-to-month list price and its floor is a comparable tier billed annually, and one vendor's single individual tier lands near each end, at 16.99 and 8.33. A company pays for seats, at 20 to 98 US dollars per user per month at the top of the same vendors' price lists, invoiced directly and off the stores, where the 20 is again an annually-billed price set against month-to-month tiers above it. A device buyer pays 149 to 299 US dollars once and then a subscription anyway. A suite owner, a mobile carrier or an operating-system vendor pays nothing at all, because it bundles the feature into something the user already has. And an investor pays, which is not revenue: 192 million US dollars into one meeting-notes application at a 1.5 billion valuation, against roughly 5 to 6 million US dollars of disclosed funding at the hardware incumbent that shipped over two million devices.

The categories this document walks through. First the direct consumer rails, in-app purchase, direct off-store billing and the one-time purchase. Then the hardware-financed rails, the device with a subscription behind it and reward crowdfunding, which are the largest revenue in the category and require a physical object. Then the organisational rails, team and enterprise seats and usage-metered platform licensing, which is where every scaled incumbent's growth actually comes from. Then the bundled rails, where no purchase decision is made at all and where the free pre-installed feature sets the floor under every price above it. Then investor capital and the exit, venture equity, revenue financing, the trade sale and retail equity crowdfunding. Last the non-dilutive rails, cloud credits and public innovation grants, which offset cost rather than producing revenue and which no company in this category is recorded as having used.

Three walls close the naive assumptions. The first is the absent payer layer above. The second is platform policy on cellular call capture: no public interface exists on iOS for the audio of an active call, and Google Play has banned use of the accessibility service interface for call recording since May 2022, which closes to software the single largest segment of stated reasons hardware buyers give for their purchase, 28.1 percent of 64 collected statements. The third is that the store commission itself is in motion in both markets that matter: after the May 2025 ruling in the Epic Games litigation a United States developer can currently link out to a web checkout with no Apple commission at all, a position an appeals court has already qualified and a district court has not yet priced, while in the European Union a unified Digital Markets Act (DMA) schedule takes effect on 1 October 2026. A commission rate is not background here; it is a live variable that moves the net on every consumer rail.

Landscape at a Glance

Direct Consumer Payment Rails

The rails a single person's own money travels. They are first because they admit any application developer with no permission beyond store enrolment, and because they are what this field's published prices actually describe. Ordered by how readily an application of this shape reaches them.

Consumer subscription via app-store in-app purchase | Live, the category default

• Statutory / regulatory basis: the Apple Developer Program License Agreement and the Google Play Developer Distribution Agreement, both set unilaterally by the store operator; in the European Union, Apple's unified fee schedule published to comply with the Digital Markets Act, effective 1 October 2026.
• Dollar size: Apple's standard commission is 30 percent, falling to 15 percent under the App Store Small Business Program for developers earning under 1 million US dollars a year, and to 15 percent for auto-renewing subscriptions in their second year. On a 15 US dollar monthly subscription at the small-business rate the developer nets 12.75 US dollars. Under the European Union schedule from 1 October 2026, standard in-app purchase is 26 percent and alternative in-app payment 20 percent, each falling to 15 and 10 percent respectively for Small Business Program participants or second-year subscriptions, and a 5 percent Core Technology Commission replaces the 0.50 euro per-install Core Technology Fee (https://ppc.land/apple-kills-eu-per-install-fee-for-5-commission-on-non-app-store-sales/). Google Play quotes its schedule as a service fee plus a separately stated billing fee: from 30 June 2026 in the European Economic Area, the United Kingdom and the United States, an auto-renewing subscription is 10 percent plus a 5 percent billing fee on the first 1 million US dollars of annual earnings, and in the markets that schedule has not reached subscriptions are 15 percent regardless of revenue; non-subscription sales run to 25 percent plus the billing fee above the threshold, or 20 percent through an external web link (https://support.google.com/googleplay/android-developer/answer/112622). A subscription of this shape therefore meets an all-in 15 percent on Google Play either way, against 15 or 30 on Apple's.
• Who qualifies / the buyer: any enrolled developer. The buyer is the individual user, and the store is the merchant of record, so the store holds the billing relationship and the renewal.
• Precedent: the entire consumer shelf. Individual paid tiers cluster at 8 to 19 US dollars a month, a spread that runs across both billing cadences rather than across tiers: Granola at 14 per user, Otter Pro at 16.99 month-to-month or 8.33 billed annually for 1,200 transcription minutes (https://otter.ai/pricing), Fathom at 15 to 19, Fireflies at 10 to 18, MeetGeek at 9.99 to 19, Notta at 8.17 to 14.99, Wave at 11.67 annualised, and Voicenotes at 9 per user per month for unlimited transcription, which undercuts every month-to-month price on the shelf (https://voicenotes.com/pricing). Otter's single individual tier supplies a figure near each end of the spread, so the 8 and the 19 are not two products.
• Gatekeeper / enrollment path: store enrolment, then per-release review. The review is a live risk for this product shape specifically: declaring the background audio mode is required to keep recording with the screen locked, and reviewers are reported to reject applications that declare it with no visible user-facing feature needing it. That report rests on developer accounts rather than on published policy text and is unconfirmed.
• Years to first dollar and blocker: revenue begins at the first paid signup. The blocker is arithmetic rather than access: the commission sits on top of a cost of goods that is metered by the hour, so a flat subscription is most exposed to its most attached users.

Direct off-store subscription | Live, and provisional in the United States

• Statutory / regulatory basis: in the United States, the May 2025 injunction issued by Judge Yvonne Gonzalez Rogers in the Epic Games litigation, which barred Apple from enforcing its external-purchase-link entitlement and its associated reporting framework on the United States storefront. In the European Union, the Digital Markets Act schedule above.
• Dollar size: on the United States storefront a developer can currently place buttons and links to an external web checkout with no entitlement, no warning screen and no Apple commission, so the take is the payment processor's fee alone, approximately 2.9 percent plus 0.30 US dollars. On a 15 US dollar subscription that nets about 14.26 US dollars, against 12.75 through in-app purchase at the small-business rate. In the European Union from 1 October 2026 an external link-out carries a 15 percent commission, 10 percent for Small Business Program participants, applied to sales completed within seven days of the tap; a small-business European seller of the same subscription nets about 12.77 US dollars, an effective take of roughly 14.8 percent (https://tiun.io/blog/ios-external-payments-us-cost-2026).
• Who qualifies / the buyer: any developer, and the same individual user. The developer becomes the merchant and takes on renewals, failed-payment recovery and tax.
• Precedent: the mechanism is in general use across consumer software. Within this category the corpus names no vendor's specific split between store billing and web billing, which is a gap rather than an absence.
• Gatekeeper / enrollment path: a payment processor or merchant of record, plus compliance with whichever store rule currently applies in each storefront. There is no application and no approval.
• Years to first dollar and blocker: immediate. The blocker is that the United States position is unsettled: an appeals court has ruled Apple is entitled to charge something for external links and the district court has not set the fee, so the zero-commission net above is a dated snapshot rather than a rate.

One-time lifetime purchase | Live only where marginal cost is near zero

• Statutory / regulatory basis: none beyond ordinary sale of goods and the store's own rules for non-consumable purchases.
• Dollar size: in this field, roughly 69 US dollars for a desktop transcription wrapper (MacWhisper Pro) and 140 to 299 US dollars for a recorder whose price includes lifetime on-device transcription with no subscription at all (iFlytek Smart Recorders, models SR302 and SR502) (https://www.umevo.ai/blogs/ume-all-posts/iflytek-smart-recorder-vs-plaud-note-which-ai-recorder-is-better-in-2026). The iFlytek figures come from a competing vendor's own blog and are the vendor-adjacent tier of evidence.
• Who qualifies / the buyer: a seller whose cost of serving an existing customer is near zero. Every instance in this field either runs the model on the buyer's own hardware or sells a device with a processing chip inside it.
• Precedent: MacWhisper Pro at a one-time fee; iFlytek's recorders, which invert the subscription model outright and are described as holding enterprise and government buyers who need offline processing. There is no lifetime tier at Voicenotes: its own pricing page carries Basic at zero dollars with 100 weekly transcription minutes and a 30-day history cap, Pro at 9 US dollars per user per month with unlimited transcription, and Enterprise at 24, and no perpetual option (https://voicenotes.com/pricing).
• Gatekeeper / enrollment path: none. The constraint is architectural, not administrative.
• Years to first dollar and blocker: immediate. The blocker is that a perpetual price cannot fund a recurring bill, so the rail is available only after the recurring bill has been engineered away.

Hardware-Financed Rails

The largest revenue in this category travels these two rails, and both require a physical object. They are profiled because that is where the demand evidence in this field actually is, and because the first of them is the rail the money the seed points at is currently flowing through.

Device sold up front with a subscription behind it | Live, and the largest single revenue in the category

• Statutory / regulatory basis: ordinary sale of goods, plus a separate software subscription agreement. The device vendor's terms push the consent burden onto the buyer: Plaud requires the user to warrant a lawful basis "under the law of every jurisdiction in which the participants in the conversation are located".
• Dollar size: the device sells at 149 to 299 US dollars across the segment, and the subscription behind it is charged anyway. Plaud lists the Note at 1,149 RMB (160.90 US dollars, converted 1 September 2026), the NotePin S at 1,249 RMB (174.90) and the Note Pro at 1,299 RMB (181.90); its plans are a free tier at 300 transcription minutes a month, Pro at 17.99 US dollars month-to-month or 8.33 per month billed annually for 1,200 minutes, Unlimited at 29.99 monthly or 19.99 billed annually, and Team at 35 per user per month or 20 billed annually through 30 November 2026 and 25 thereafter (https://www.plaud.ai/pages/plaud-ai-plan-pricing). That Pro price is the same recurring price, at the same cadence and for the same 1,200 minutes, that the leading application charges: Otter Pro is 16.99 month-to-month or 8.33 billed annually (https://otter.ai/pricing). The device therefore costs its hardware price on top of a subscription that is charged either way. Competing devices: Umevo Note Plus at 149 US dollars with unlimited transcription in year one falling to 400 minutes a month in year two; Notta Memo at 149; Bee at 49.99 with a 24 US dollar monthly subscription; the Humane AI Pin at 699 falling to 499 with a mandatory 24 US dollar monthly cellular plan.
• Who qualifies / the buyer: a company that can manufacture, stock and return hardware. The buyer is an individual, frequently expensing the purchase afterwards, in a bottom-up pattern the sources describe as shadow information technology.
• Precedent: Plaud is the case. It shipped over a million and a half devices by January 2026 on its own count and over two million globally by mid-2026 on the trackers', on 5 to 6 million US dollars of disclosed funding. Its revenue is reported two ways by the same tracker and the two do not reconcile: 100 million US dollars of software-only annual recurring revenue as of June 2026, and a 250 million US dollar total annualised run-rate calculated in September 2025 (https://sacra.com/c/plaud/). The 60 percent hardware and 40 percent software split that circulates is computed by dividing the first of those figures by the second, that is, a June 2026 number by a September 2025 number, in the same document that calls them conflicting. The split is reported here as an artifact of that division and not as a revenue mix. Per the project's own correction register, no completed investment round or valuation for Plaud is established: one report records only that the company "has reportedly engaged in discussions" at a 1 to 2 billion US dollar range, and the completed Tencent-and-Meituan round at a 1 billion valuation belongs to a different company in a different form factor.
• Gatekeeper / enrollment path: a supply chain, then retail and marketplace distribution. Amazon and Best Buy listings are where this category's buyer evidence is publicly visible.
• Years to first dollar and blocker: at the first device sale, and the cash arrives before the service does. The blocker for anything shaped like an application is the object itself; the blocker for the rail generally is that the attach rate, the share of device buyers who pay for software, is disclosed nowhere, so whether the 100 million US dollar software figure means high attach or high volume cannot be determined.

Reward crowdfunding pre-orders | Live for a physical product, and this field's only public demand series

• Statutory / regulatory basis: a campaign platform's terms. Backers receive a product, not equity, and have no claim if it does not ship.
• Dollar size: three campaigns, all hardware. Plaud Note, under the iZYREC creator account on Kickstarter, 27 June to 16 August 2023: 1,108,181 US dollars from 7,564 backers (https://www.kickstarter.com/projects/izyrec/plaud-note-chatgpt-empowered-ai-voice-recorder). HiDock H1, Kickstarter and Indiegogo, 10 October to 9 December 2023: 554,444 US dollars from 2,431 backers. Senstone, Kickstarter, February 2017: approximately 100,000 US dollars from approximately 1,000 backers.
• Who qualifies / the buyer: a team with a manufacturable physical product. The buyers are individual backers paying in advance.
• Precedent: the three above and no others in the sources. No campaign is recorded for Limitless, Bee, Humane, the Plaud NotePin, or for any device that failed to ship, so this series shows successes only. The platform ecosystem is also visibly sceptical: independent commentators publicly flagged the iZYREC campaign as a scam before it fulfilled.
• Gatekeeper / enrollment path: platform review, then a public campaign that either hits its target or returns the money.
• Years to first dollar and blocker: at the campaign's close. The blocker is that these platforms are structured around physical goods: all three campaigns funded an object, and the sources record no software campaign in this category at all.

Organisational Payment Rails

Where the money in this field visibly compounds. Both rails below bypass the stores entirely and both are gated by something other than the product: one by a security review, the other by being a supplier rather than an application.

Team and enterprise seats invoiced direct | Live, and the growth engine of every scaled incumbent

• Statutory / regulatory basis: a commercial contract, with a data processing agreement under Article 28 of the General Data Protection Regulation (GDPR) where the buyer is European, and a Business Associate Agreement under the Health Insurance Portability and Accountability Act (HIPAA) where the buyer is a United States healthcare provider.
• Dollar size: 20 to 98 US dollars per user per month across the field. Otter lists Business at 30 per user per month, or 19.99 billed annually, and an Enterprise tier reported at 98; Granola Enterprise begins at 35 per user per month; Jamie sells Teams at 39 euro per seat per month; tl;dv lists Business at 65 per seat; MeetGeek Business at 39; Voicenotes Enterprise at 24 with single sign-on, provisioning, consent management and a contractual bar on model training (https://otter.ai/pricing, https://voicenotes.com/pricing).
• Who qualifies / the buyer: any vendor that can pass a security review. The buyer is an information technology, security or finance function inside a company, not the person who will press record.
• Precedent: this is where the category's growth is. Otter is reported at roughly 100 million US dollars of annual recurring revenue, structured to push individuals toward team deployments; Read AI's valuation is anchored to penetration of 75 percent of Fortune 500 companies rather than to the 100,000 new consumer accounts a week it also reports; Granola's 125 million US dollar round was raised on an enterprise-context thesis; Fireflies reports use inside a large share of the Fortune 500. Both revenue figures here are tracker estimates.
• Gatekeeper / enrollment path: SOC 2 Type 2 and ISO 27001 are described as the non-negotiable baseline; a data processing agreement and stated regional data residency follow for European buyers; a Business Associate Agreement is available from Otter and Fireflies but gated behind high-tier enterprise plans. One vendor, Jamie, holds ISO 27001 and SOC 2 Type II, processes exclusively on European servers, deletes audio the moment the transcript exists and contractually guarantees no model training, and sells that posture as the product. The gate can also slam: Chapman University imposed a blanket ban on Read AI in August 2025 over data security, privacy and consent concerns.
• Years to first dollar and blocker: first revenue at the first signed contract. The blocker is the review itself, and a second one behind it: the individual and small-team buyer described in the same sources does not ask for certifications at all, but for on-device processing, so the evidence that opens this rail is not the evidence that convinces the person recording the meeting.

Usage-metered speech platform licensing | Live, and a different business from the application above it

• Statutory / regulatory basis: a self-serve developer agreement with a negotiated enterprise tier above it.
• Dollar size: AssemblyAI publishes a base rate of 0.15 US dollars per hour of audio with upcharges for speaker diarization, entity detection and sentiment; elsewhere in the same evidence base AssemblyAI is recorded at 0.37 US dollars an hour once those are on. Deepgram is quoted twice at different floors in one document from the same cited source, at 0.0043 to 0.0077 and at 0.0048 to 0.0077 US dollars a minute, and the disagreement is reported here rather than resolved. OpenAI's hosted transcription is quoted at 0.006 US dollars a minute, which is 0.36 an hour. Application vendors also gate programmatic access behind their own higher tiers, reported at 50 to 80 US dollars a month (https://convertaudiototext.com/blog/best-transcription-with-api-access).
• Who qualifies / the buyer: any operator of a tuned speech pipeline. The buyer is another developer or an enterprise data team, and it pays by volume.
• Precedent: AssemblyAI and Deepgram sell nothing else. Fireflies and Sonix sell application programming interface access as an enterprise upsell rather than as the business.
• Gatekeeper / enrollment path: none for self-serve; a commercial negotiation above it. Deepgram grants 200 US dollars of free credit at signup, described as roughly 400 hours of processing.
• Years to first dollar and blocker: immediate on self-serve. The blocker is that this is the supplier side of the same market, competing on price per minute against companies whose prices are already published, and it reaches no meeting-goer at all.

Rails Where No Purchase Decision Is Made

Bundled distribution. These rails admit only the owner of the container, and they sit here, among the rails that set prices rather than below them, because the free bundled feature caps what everything above it can charge whether or not a third party can ever reach it. A competitor reaching a market this way never appears on a comparison page and pays nothing to acquire the user.

Pre-installed operating-system feature | Live, and only for the platform owner

• Statutory / regulatory basis: none. The capability ships with the device.
• Dollar size: zero to the user, and there is no subscription. Google Pixel Recorder records, transcribes in real time, labels speakers and generates summaries with an on-device model, entirely offline and free. Apple Voice Memos transcribes on device from iOS 18 on iPhone 12 and later, and Apple Notes records, transcribes and summarises through the on-device model framework, free and offline. The compute cost is borne by the buyer's own hardware, which is why no minute cap is needed.
• Who qualifies / the buyer: the platform owner. Nobody buys the feature; it is funded as hardware margin and ecosystem lock-in.
• Precedent: both platforms ship it now. What neither is documented as producing is action items as a distinct artifact: one review of the platform recorder finds it does not reliably isolate action items into a checklist separate from the summary paragraphs, and the Apple recorder is reported with no speaker labels, no summary and no action items. That gap is undocumented rather than demonstrated.
• Gatekeeper / enrollment path: none available to anyone else.
• Years to first dollar and blocker: not applicable to a third party. The blocker is that this rail sets the floor price for everyone else's: the core interaction ships free and pre-installed on both platforms.

Bundling inside a productivity or communications suite | Live, and only for the suite's owner

• Statutory / regulatory basis: none. The capability is a feature of a subscription the customer already holds.
• Dollar size: the feature carries no price of its own. ByteDance's Lark (Feishu) includes Lark Minutes at 300 free transcription minutes a month for individuals and unlimited processing on paid enterprise tiers running 6 to 39 US dollars per user per month (https://www.larksuite.com/en_us/plans). Alibaba's DingTalk A1 gives 1,000 free minutes a month, with professional tiers at 599 RMB (83.90 US dollars) annually and premium at 1,499 RMB (209.90). In the West the same capability sits behind a licence rather than a price: Microsoft Teams intelligent recap, which summarises and suggests action items, requires a Teams Premium add-on licence or a Microsoft Copilot licence (https://support.microsoft.com/en-us/office/use-copilot-in-microsoft-teams-meetings-0bf9dd3c-96f7-44e2-8bb8-790bedf066b1); Google Meet's automatic notes, which carry a summary, decisions and next steps, require an eligible Google Workspace edition or Google artificial intelligence plan held by the meeting organiser (https://support.google.com/meet/answer/14754931); Zoom's assistant is included with a paid plan and caps summaries at three hosted meetings a month at the entry tier (https://www.zoom.com/en/products/ai-assistant/).
• Who qualifies / the buyer: the suite owner. The buyer already bought the suite and makes no separate decision.
• Precedent: Lark records an in-person meeting from its own mobile application and routes the transcript, summary and action items automatically into chat channels and corporate customer relationship management bases. ByteDance also partnered with the hardware manufacturer Anker to produce the AI Recording Bean, a ten-gram wearable that syncs offline audio into Lark, treating hardware as an accessory to the software platform.
• Gatekeeper / enrollment path: for a third party, none exists: this rail is reached by being distributed by the container's owner, and the sources record no third-party meeting-notes product distributed this way in a Western suite. The adjacent route that does exist is a listing in a customer relationship management or conferencing marketplace, which is distribution rather than bundling.
• Years to first dollar and blocker: not applicable to a third party. The blocker is ownership of the container, and the historical consequence is documented: Vowel's value proposition collapsed when Zoom and Microsoft Teams shipped free native summaries.

Carrier bundling into the native dialer | Live, and only for the carrier

• Statutory / regulatory basis: none. The service is an attribute of a mobile network subscription.
• Dollar size: free to the subscriber, funded as retention. SK Telecom's Adot is built into the native mobile dialer for the carrier's own subscribers, passed ten million subscribers by mid-2025, and records, transcribes, translates and summarises at no extra charge. The carrier's consolidated revenue is 43.8 trillion Korean won (approximately 32.8 billion US dollars), so the feature's cost is invisible inside it.
• Who qualifies / the buyer: a mobile network operator. The subscriber already pays for the plan.
• Precedent: Adot, which bypasses both application stores entirely and has since been extended into a multimodal service linking smart glasses. Naver's Clova Note occupies the same national market on its own large language model. No instance of a third-party application being carrier-bundled in this category appears anywhere in the sources.
• Gatekeeper / enrollment path: for a third party, a carrier partnership, of which there is no example here.
• Years to first dollar and blocker: not applicable to a third party. The blocker is that the carrier owns the dialer, which is also the one surface where cellular call capture is possible at all.

Investor Capital and the Exit

Not revenue, and included because it is where most of the money named in this field actually sits, and because the exit is the only mechanism by which any of it has ever been returned. Almost every figure below is a private-market tracker's estimate rather than a filed number; the tracker is named wherever one is.

Venture equity | Live, heavily used, priced at software multiples

• Statutory / regulatory basis: private share purchase agreements. Nothing is filed publicly, which is why the same round appears at different dates and sizes in different trackers.
• Dollar size: Granola is the ceiling: a 14 million US dollar Series A in October 2024 (reported elsewhere as 20 million led by Spark Capital, a disagreement recorded here rather than resolved), a 43 million Series B in May 2025 led by NFDG at a 250 million valuation, and a 125 million US dollar Series C led by Index Ventures with Kleiner Perkins at a 1.5 billion US dollar post-money valuation, for 192 million total per Tracxn (https://techcrunch.com/2026/03/25/granola-raises-125m-hits-1-5b-valuation-as-it-expands-from-meeting-notetaker-to-enterprise-ai-app/). Read AI: 21 million Series A in April 2024 led by Goodwater Capital and Madrona, then 50 million Series B in October 2024 led by Smash Capital at a 450 million valuation, 81 million total. Otter: approximately 70 million raised, investors including Spectrum Equity and Horizons Ventures. Fireflies: approximately 19 million from Canaan and Khosla Ventures, with a 1 billion US dollar valuation announced on the company's own blog in June 2025 (https://fireflies.ai/blog/fireflies-1-billion-valuation). Notta Co., Ltd: 31.8 million, Series B December 2025, Granite Asia, GL Ventures and Mizuho. MeetGeek: 2.56 million, Early Game Ventures and Techcelerator. On the hardware side the same rail is thin: Friend raised a 2.5 to 3 million US dollar seed in July 2024 from Pace Capital, Caffeinated Capital and Andreessen Horowitz and reported roughly 348,000 US dollars of revenue from 3,000 units; the largest completed hardware round in the corpus belongs to a different form factor, Even Realities, at 150 million pre-Series B in July 2026 led by Meituan and Tencent at a 1 billion valuation.
• Who qualifies / the buyer: any company an investor will price. The buyer is a fund, and it is buying a share of a later sale.
• Precedent: dense, and skewed to software. Investors have funded meeting applications to unicorn valuations while standalone consumer audio hardware has attracted very little, notwithstanding that hardware produced the category's largest revenue.
• Gatekeeper / enrollment path: introduction and negotiation. There is no application and no published criterion.
• Years to first dollar and blocker: at the round's close. The blocker is that the software half of this category is already capitalised, one company at a 1.5 billion US dollar valuation with 192 million raised, so a new entrant is priced against funded incumbents rather than against an empty field.

Revenue-financed operation, no outside capital | Live, with named precedent at scale on both sides

• Statutory / regulatory basis: none. This is the absence of a rail rather than a rail, and it is listed because the field's two most instructive cases both run on it.
• Dollar size: Plaud shipped over a million and a half devices (its own count as reported in January 2026) on approximately 5 to 6 million US dollars of total disclosed venture funding, most recently a 4.75 million US dollar convertible note led by Carbide Ventures in April 2025, per Sacra and Tracxn (https://sacra.com/c/plaud/). Fathom reached 16 million US dollars of annual recurring revenue by the end of 2023 with no venture backing, per the tracker GetLatka. Wave AI Note Taker, built by a single independent studio, is estimated by the tracker Fluxbuilder at 360,000 US dollars of monthly recurring revenue on 25,000 monthly downloads, which annualises to roughly 4.3 million. Voicenotes is recorded as unfunded and bootstrapped. Meetily is open source and free for individuals, charging 120 to 360 US dollars a year only for managed team deployments.
• Who qualifies / the buyer: anyone whose first transaction is cash-positive. The buyer is the ordinary customer of whichever rail above the company uses.
• Precedent: four named cases, two of them at scale. Note what the two largest have in common and the application does not: Plaud's first transaction is a device sale that arrives in full before any service is delivered, and Fathom's bootstrap ran on a product sold to teams.
• Gatekeeper / enrollment path: none.
• Years to first dollar and blocker: at the first paid transaction. The blocker is named directly in the sources and is the single most consequential idea in this field's money layer: the 159 US dollar device may be structurally a self-liquidating customer acquisition channel, cash-positive at the point of sale, funding software development without dilution, with the physical object acting as a sunk-cost anchor that lowers software churn relative to a deletable application. This is the reasoning of one analysis and it carries no source and no churn number for either side. The funding and shipment figures under it are Sacra's; the argument on top of them is not evidence. It is recorded here because if it holds, the mechanism that pays for finding the customer in this category is the hardware margin, and a software-only entrant has nothing in its place. The application rail does carry a cash-in-advance mechanism of its own, which the prices throughout this document imply without naming: an annual plan is collected in full at signup and remitted by the store in one payment less commission, so Otter Pro at 8.33 a month billed annually is 99.96 US dollars arriving before a minute is transcribed. That is a fraction of a 149 to 299 US dollar device sale, and it arrives on the same terms.

Trade sale to a platform owner | The category's only demonstrated exit

• Statutory / regulatory basis: private acquisition agreements. No filing carries terms for two of the three.
• Dollar size: only one price was ever disclosed. HP acquired the assets of Humane in February 2025 for a reported 116 million US dollars, against roughly 230 million raised and a peak venture valuation reported at 850 million to 1 billion; Humane shut its cloud servers on 28 February 2025, rendering every AI Pin sold non-functional, with refunds limited to a 90-day return window (https://gizmodo.com/humane-bricks-its-ai-pin-as-it-gets-acquired-by-hp-for-116-million-2000565528). Meta acquired Limitless in December 2025 and immediately halted new device sales while continuing to support existing users; terms were not disclosed. Amazon acquired Bee in July 2025 and continued the product line; the acquirers disclosed no terms, and a figure of 7 million US dollars appears in one tracker-derived table, given in that document's body as the acquisition price and in its own table as total funding raised, so it is not usable as a price.
• Who qualifies / the buyer: in this category, so far, only platform owners. Three of three buyers were Meta, Amazon and HP.
• Precedent: three acquisitions inside thirteen months, all of hardware, all by platform owners, plus one team acqui-hire on the software side: Vowel, which raised 17.8 million US dollars including a 13.5 million Series A in September 2021, shut down in July 2023 after a signed term sheet was withdrawn and after Zoom and Microsoft Teams shipped free native summaries, and whose team was taken on by Zapier in March 2024.
• Gatekeeper / enrollment path: none that can be applied for.
• Years to first dollar and blocker: at the deal. The blocker is that the pattern has a direction and no price: three of three says who buys, and one disclosed figure out of three says nothing about what a software-only business in this category is worth. The reading that undisclosed terms imply acqui-hires or distress sales is one analysis's own inference and carries no source.

Retail equity crowdfunding | Live, one named precedent in this category

• Statutory / regulatory basis: a regulated online investment offering. The specific exemption Fathom's offering relied on is not recorded in the sources behind this document.
• Dollar size: over 3.2 million US dollars from 2,148 retail investors at a 73 million US dollar valuation, as part of a 17 million US dollar Series A that reserved 10 percent of the allocation for the company's own users (https://growthturbine.com/blogs/25-biggest-wefunder-success-stories-updated-2025).
• Who qualifies / the buyer: a company willing to disclose and to take on many small shareholders. The buyers are individual retail investors, in this case drawn from the product's existing user base.
• Precedent: Fathom, on the Wefunder platform, and no other instance in this category anywhere in the sources. Fathom went on to raise a 43 million US dollar Series B in May 2025, so the retail round did not substitute for institutional capital.
• Gatekeeper / enrollment path: the funding portal, which runs the offering and the disclosure.
• Years to first dollar and blocker: at the offering's close. The blocker is sequence: this offering followed a product with proven revenue and an existing user base to sell to, and it is not visible in the sources as a route to a first product.

Non-Dilutive Cost Offsets and Public Programmes

Money that is not revenue and not equity. In a regulated field this is where evidence generation gets paid for; here it offsets infrastructure cost or funds a research plan, and no instance of any programme below being used by a company in this category appears anywhere in the sources.

Cloud provider startup credits | Live, a cost offset and not revenue

• Statutory / regulatory basis: a supplier's published startup programme terms.
• Dollar size: standard tiers at Amazon Web Services, Google Cloud and Microsoft Azure are described as granting 100,000 to 350,000 US dollars of computing credit valid for one to two years. That figure carries no citation in the source that states it and is recorded here as unconfirmed. A specific and citable smaller instance: Deepgram grants 200 US dollars of free application programming interface credit at signup, described as roughly 400 hours of processing.
• Who qualifies / the buyer: early-stage companies meeting a supplier's published criteria. Nobody pays anything; a cost is deferred.
• Precedent: the mechanism is standard across software startups. No named company in this category is recorded as having taken it.
• Gatekeeper / enrollment path: an application against published criteria, frequently routed through an accelerator, an investor or a partner programme.
• Years to first dollar and blocker: at acceptance. Two blockers: credits expire, and against this field's cost structure they buy a bounded number of recorded hours. At the assembled marginal cost of 0.21 to 0.62 US dollars per recorded hour, a 100,000 US dollar credit is on the order of 160,000 to 480,000 recorded hours, which is a runway for a cohort and not a business model.

Small Business Innovation Research Phase I | Open, no precedent found in this category

• Statutory / regulatory basis: the United States federal Small Business Innovation Research programme, administered by participating agencies against their own solicitations.
• Dollar size: Phase I awards of up to 275,000 US dollars for proof-of-concept development.
• Who qualifies / the buyer: small United States businesses. The awarding agency is the payer and it funds a work plan, not a product.
• Precedent: none found in this category. The only positioning advice in the sources is that framing a transcription tool as accessible technology for the hearing impaired, or toward a defence application, increases the probability of an award, which is an assertion with no example behind it.
• Gatekeeper / enrollment path: registration with the federal contracting systems, then a competitive proposal against a named agency solicitation with fixed deadlines.
• Years to first dollar and blocker: on the solicitation's award cycle. The blocker is subject matter: the programme funds a research question with a public purpose, and a consumer meeting recorder has to be rewritten as one before it is eligible in substance rather than in form.

European Innovation Council Accelerator | Open, no precedent found in this category

• Statutory / regulatory basis: the European Innovation Council, the European Union's instrument for high-risk, high-impact innovation.
• Dollar size: grants of up to 2.5 million euro, with blended equity available alongside. Localised national digitisation grants in member states such as Germany and France subsidise software development separately.
• Who qualifies / the buyer: companies established in the European Union or an associated country. The European Union is the payer.
• Precedent: none found in this category.
• Gatekeeper / enrollment path: a competitive call with a written application, an interview stage and reporting obligations against an approved work plan.
• Years to first dollar and blocker: on the call's award cycle. The blocker is establishment: the programme requires a European Union entity, and the acceptance rate and evaluation criteria are not recorded in the sources behind this document.

Israel Innovation Authority research and development grants | Open, no precedent found in this category

• Statutory / regulatory basis: the Israel Innovation Authority's research and development grant tracks.
• Dollar size: up to 50 percent of an approved research and development budget for early-stage companies. The award is repayable, which separates it from every other entry in this section: the authority's published royalty rules take a share of revenue until the grant is repaid with interest, and transferring the funded knowledge or the manufacturing out of Israel triggers a repayment capped at a multiple of the grant. The specific royalty percentage and the interest basis are not established here.
• Who qualifies / the buyer: companies located in Israel, or companies partnering directly with Israeli firms. The authority is the payer and funds a share of approved spending.
• Precedent: none found in this category.
• Gatekeeper / enrollment path: an application to the authority's research committee against an approved budget.
• Years to first dollar and blocker: on the committee's cycle. The blocker is establishment and matched spending: the grant pays a share of money the company has already committed. Repayment or royalty terms are not recorded in the sources behind this document.

Dead Ends: Structurally Closed Rails

Grouped by the pattern that closes them, so the pattern is recognisable elsewhere.

Closed by platform policy rather than by law.

Paid cellular call capture by a third-party application | Closed by: Apple and Google platform policy

• What blocks it: iOS exposes no public interface to the audio stream during an active cellular call, and Google Play policy has banned use of the accessibility service interface for call recording since May 2022. Wiretap law is platform-agnostic and is not what closes this; the stores are.
• Is there any pivot that reopens it? Not within an application. The routes that exist go around it: the dedicated recorder ships a vibration conduction sensor that reads chassis vibration through the phone's back glass, doing in hardware what platform policy forbids software to do; one application-side product instructs users to switch to speakerphone so the microphone can hear both sides; and a carrier that owns the dialer can do it natively, which is what makes the carrier rail above possible. The evidence for the vibration sensor itself rests on one aggregator page and a user manual reference, with no teardown, patent or vendor specification behind it.
• Precedent of others hitting this wall: the buyer evidence is the precedent. In a convenience sample of 64 explicit purchase-rationale statements, 28.1 percent, the largest single category, is capturing a cellular phone call, quoted verbatim as "Otter doesn't record phone calls (thanks, iOS)". Those buyers did not walk past an equivalent application; for that job there is none, and the revenue attached to that demand is unreachable from any software rail.

Closed by the category's own contracts and by pending litigation.

Selling or licensing the recorded corpus | Closed by: biometric privacy statutes plus the market's own contractual guarantees

• What blocks it: the Illinois Biometric Information Privacy Act names a voiceprint as a biometric identifier at 740 ILCS 14/15(b), with statutory damages stated as 1,000 US dollars per negligent and 5,000 per intentional violation, and no proof of harm required. On top of the statute the market has already sold the opposite promise: Jamie contractually guarantees that customer data is never used for artificial intelligence model training, and Granola's policies state that third parties are not allowed to use personal data for model training. A vendor that has made those promises to win enterprise seats cannot then sell the corpus.
• Is there any pivot that reopens it? None found that survives both constraints at once. Anonymous ephemeral speaker clustering inside a single recording is described as the safe harbour; tying a voice vector to an identity is the transition into named biometric identification, and enrolment is also the only technical route to better attribution.
• Precedent of others hitting this wall: In re Otter.AI Privacy Litigation, consolidated before Judge Eumi K. Lee in the Northern District of California from Brewer v. Otter.ai Inc. (August 2025), which alleges among other things that Otter used recorded content to train its speech models without adequate disclosure to non-account-holders, and records that Otter defaults to an opt-out model for training on meeting data (https://www.workplaceprivacyreport.com/2025/08/articles/artificial-intelligence/ai-notetaking-tools-under-fire-lessons-from-the-otter-ai-class-action-complaint/). Three further actions are pending on adjacent theories: Cruz v. Fireflies.AI Corp. (Illinois, December 2025, voiceprint claims, consolidated with Brewer), Basich v. Microsoft Corp. (Western District of Washington, February 2026, live transcription in Teams) and Chamberlain v. Granola (Northern District of California, July 2026, against a product that processes audio on the user's own device and was sued over how it markets covert capture). None has resolved. There is no judgment, no settlement and no regulator fine anywhere in this field, so this is a cost line with no number on it; the one dollar figure attached to any of the cases is a litigation-analytics vendor's projection rather than an agreed or approved amount. A 2026 Seventh Circuit decision holding the 2024 Illinois amendment retroactive and limiting damages appears in the sources and its effect on the per-instance multiplier is unverified.

Closed by the absence of a payer layer: the product is consumer software with no regulatory classification.

Health insurance reimbursement and fee-for-service billing codes | Closed by: the absence of any coverage pathway for a non-clinical consumer product

• What blocks it: a billing code pays a covered provider for a clinical service delivered to a patient. This product has no regulatory classification at all: this project's own contract records the regulatory class as not applicable and states that the binding legal surface is recording-consent law, which constrains the product without classifying it. Nothing in this field has been submitted to a regulator, no code exists that describes producing minutes of a business meeting, and no coverage policy anywhere in the sources mentions the category. The only statutes that reach this product are wiretap and biometric ones, 18 U.S.C. 2511(2)(d) for the federal one-party consent floor and 740 ILCS 14/15(b) for voiceprints, and neither is a payment authority.
• Is there any pivot that reopens it? Only by becoming a different product. Ambient clinical documentation sold to a healthcare provider is an adjacent category with its own coverage and privacy requirements; within this one, the nearest thing that exists is a Business Associate Agreement offered by Otter and Fireflies and gated behind high-tier enterprise plans, which is a contract term on the seat rail and not a coverage rail.
• Precedent of others hitting this wall: none, because nobody in this category has tried. That is itself the finding: across the whole field, not one vendor is recorded as pursuing a payer, a code or a coverage decision.

Employer benefit plans and tax-advantaged health accounts | Closed by: the same absence

• What blocks it: these accounts and plans reimburse qualified medical expenses. A meeting recorder is not one, and no vendor markets one as such.
• Is there any pivot that reopens it? None found. The nearest adjacent case in the sources runs the other way: a national tax code that lets a Japanese business instantly expense a recorder priced under 100,000 yen as a consumable, which is a tax treatment of a business purchase rather than a benefit rail.
• Precedent of others hitting this wall: none recorded.

Emerging Patterns

Every rail that compounds in this field ends at an organisation, not at a person. The consumer subscription is the rail every vendor lists first and none of the scaled ones grows on: Otter prices individuals at 8.33 US dollars per user per month billed annually and drives growth through Business at 30 and an Enterprise tier reported at 98; Read AI's valuation is anchored to Fortune 500 penetration rather than to the 100,000 consumer accounts it adds weekly; Granola, Jamie, Wave, MeetGeek, tl;dv and Voicenotes all publish a team or enterprise tier above the individual one. The operational consequence is that the evidence that opens the compounding rail is a security review, SOC 2 Type 2 and ISO 27001 and a data processing agreement and stated residency, while the sources say plainly that the individual and small-team buyer asks for something else entirely, on-device processing. Those are two different products' worth of evidence, and only one of them is a rail with expansion in it.

Where the compute runs decides which rails are available, before any pricing decision is made. The assembled marginal cost of a recorded hour in the cloud is 0.21 to 0.62 US dollars on one assembly and 0.26 to 0.38 on a narrower one, dominated by transcription; the summarisation call on a one-hour transcript is quoted across the same evidence base at anything from 0.002 to 0.10 US dollars, which is between two and four orders of magnitude smaller either way and negligible against the transcription half. At a 15 US dollar subscription netting about 13 after commission and a 0.40 US dollar median hour, breakeven is roughly 32.5 hours a month, and a user recording two to three hours a working day costs the vendor about 20. Move transcription onto the device and the marginal cost falls to about 0.05 US dollars an hour with breakeven past 250 hours; move everything onto the device and it reaches zero. But the same move removes the floor that lets anyone charge: every product in this field with a one-time price or a free unlimited tier is a local-processing product, and the two free unlimited products are the operating systems' own. Zero marginal cost is not a moat here; it is the mechanism by which the price of the category's core function goes to nothing.

Free tiers do not converge, and each one meters a different thing. Three products give 300 transcription minutes a month; the rest sit nowhere near it. Fireflies caps total storage rather than a monthly reset, at a figure the sources give inconsistently as 800 minutes in one place and 400 in another. Notta gives 120 minutes a month with a 3-minute per-file cap, Wave 30 minutes, Voicenotes 100 minutes a week, DingTalk 1,000 minutes a month, Fathom caps summaries at the first five calls rather than capping capture at all, and Granola substitutes a 30-day history cap for a minute cap. The consequence is that there is no category benchmark to price a giveaway against: each cap is one vendor's own cost-control instrument, set by that vendor's architecture, and reading a market signal into a shared number is reading an artifact. No cap is free to the vendor, either. A cap is a metered inference bill paid on behalf of people who mostly never pay anything: at the assembled 0.21 to 0.62 US dollars a recorded hour, a 300-minute monthly cap costs up to roughly 3 US dollars a month for every free user who consumes it, against a 15 US dollar subscription that nets about 13. The one worked cohort in this evidence base, which carries the same borrowed conversion rate as every other customer figure here and the same absence of a footnote, has 100 free signups consuming up to 200 US dollars of inference a month while the three to five who convert generate 45 to 75, loss-making before commission is applied. Every cap in the list above is that arithmetic, solved once by one vendor for one architecture.

The category's only demonstrated exit is a sale to a platform owner, and the pattern has a direction with no price. Meta bought Limitless in December 2025, Amazon bought Bee in July 2025, HP bought Humane's assets in February 2025, and only HP's 116 million US dollars was ever disclosed. Three of three tells a reader who thinks this capability belongs inside a larger product; one disclosed figure out of three tells nobody what one of these businesses is worth. The same asymmetry runs through the whole money layer: the field is measured in raises, valuations, install counts and review counts, and not one usage figure exists anywhere in it, no retention, no churn, no share of installs that record a second meeting.

Everything checkable about the money is a company figure; everything about the customer is borrowed or unlitigated. Every company number in this field carries a footnote, to a tracker or a news item, and the customer numbers carry none: free-to-paid conversion of 2 to 5 percent, customer acquisition cost of 20 to 40 US dollars, monthly churn of 5 to 8 percent and a resulting lifetime value of 180 to 300 are taken from consumer productivity software generally, and the source that supplies them states outright that category-specific financial data for private meeting applications is rarely audited publicly. They are labelled as borrowed everywhere they appear in this document and should not be treated otherwise. The same shape appears on the risk side: four class actions, no judgment, no settlement, no fine, and one projection from a litigation-analytics vendor. A rail in this field can be sized by what it pays. It cannot be sized by what it converts.

Also Found, Not Profiled

Rails the sweep checked and did not profile, with the fact that put each here. Alphabetical.

Watchlist

The moves that would change what any of these rails is worth. Dated as of 1 September 2026.