Eyes on the smart screen, AI earbuds in ears, AI glasses on the bridge of the nose, an AI recording card at hand, an omnidirectional microphone in the middle of the table, and functions like ARS, TTS, summarization, abstraction,generalization, to-do lists, Q&A, and real-time search powered by large models like Gemini, GPT, Tongyi Qianwen, and Wenxin Yiyan embedded within these devices.
Because AI hardware equips the office with “digital ears” and an “AI brain,” meetings, once a “disposable consumable,” have been transformed into productive fields for knowledge sedimentation and accumulation.
In a very short time, the types and forms of AI office hardware on the market have rapidly innovated: from AI headphones that can seamlessly translate over a dozen languages, to recording cards as thin as a bank card yet capable of discerning power dynamics, to elegantly floating microphones resembling art installations—these全新的AI hardware are making a “saturated investment” attempt to liberate workers from tedious information processing.
Among them, sales of AI recording cards have already exceeded one million units. This seems to indicate that products aimed at efficiency improvement and work model innovation represent a more certain “rigid demand” market within the broader innovation hardware track, where leading players are also more likely to achieve substantial revenue and profits.
But the final outcome is clearly not yet determined. Just like all popular product directions, new competitors will emerge, prices will continuously drop, and users will become increasingly discerning. More crucially, the questions of whether office hardware should be sold to individuals or enterprises, and who owns the data—the employee or the company—will lead to different commercial paths based on the choices made.
01 AI Hardware Conquers the Conference Room
The emergence of AI office hardware stems from the need for information processing in complex work scenarios. How does a cross-border boss seamlessly communicate in a multinational meeting with five foreign clients using Chinese, English, Japanese, Korean, French, and Spanish? How can a sales manager track the visit situations of every frontline salesperson and be thoroughly familiar with all customer needs? How can remotely working teams completely preserve collaborative information? How can product owners adjust progress随时 amidst complicated R&D and production processes?
In this pursuit of “frontline information,” various B-end AI tools have emerged, equipping business activities with “digital ears” and a “brain”: AI headsets, AI recorders, AI glasses, AI microphones, and AI wearable tools have sprung up like bamboo shoots, flooding into conference rooms, factories, large meetings, and various other work scenarios. They silently complete the collection and sedimentation of massive amounts of raw information and, leveraging powerful AI capabilities, capture key topics, identify action items, and归纳 disagreements and consensus from tens of thousands or even hundreds of thousands of words of raw audio and transcripts, refining chaotic piles of information into clear reference guides.
The first to catch fire was the Plaud Note from a Shenzhen hardware startup. This product is as thin as a bank card, can attach to the back of a phone, and blends seamlessly into communication scenarios—whether you’re on a phone call, in an online meeting, or having an offline conversation, a light press of a button instantly starts recording. Unlike ordinary recorders, Plaud AI supports multi-language high-precision transcription (112 languages) and uses large models to automatically generate meeting summaries, mind maps, and to-do lists.
Plaud Note was launched for overseas markets in June 2023. As of June 2025, global sales exceeded 1 million units.
In 2025, the popularity of AI recording cards spread from abroad to China. In March, Mobvoi launched the TicNote AI recording card, featuring a self-developed AI agent, and it has become one of Mobvoi’s best-selling hardware products. In August, DingTalk launched the DingTalk A1, deeply integrated into the DingTalk office ecosystem. It can not only generate verbatim transcripts, summaries, minutes, and other text documents but also create report documents with one click based on built-in professional templates within DingTalk.
Besides recording cards, AI headphones are also frequent participants in meetings. Their functions have evolved from early recording, noise cancellation, and transcription to real-time translation, meeting minutes, health monitoring, and even progressing towards the role of a “digital companion” with personality traits.
AI meeting headphone company Future Intelligence was established at the end of 2021, previously being the headphone business department of iFlytek. In 2022, it launched the iFlytek AI Meeting Headphone series, creating a new sub-category. As of May this year, the cumulative shipments of the iFlytek AI Meeting Headphone series exceeded 800,000 units, with revenue doubling for three consecutive years. On October 13, Future Intelligence completed its third round of funding this year.
In January 2025, Future Intelligence launched its overseas brand Viaim at CES, expanding into North American and Asia-Pacific markets. It entered the Middle East in July and announced its entry into Europe via the IFA exhibition in September. From January to July this year, Viaim’s sales in North America increased 7.2 times, and sales in the Asia-Pacific region grew 1.28 times compared to the second half of last year.
Insta360, primarily focused on sports and outdoor scenarios, also entered the meeting scene, launching an AI omnidirectional microphone in September. It features a unique floating design and can integrate with video conference cameras (Link series) to automatically locate sound sources and track speakers.
In October, Plaud launched the AI wearable recording device Plaud NotePin S. Shaped like a smooth “small pill” or “lipstick tube,” it can be turned into a necklace, bracelet, brooch, or clip-on lapel recorder using standard accessories like lanyards, wristbands, and clips, allowing users to record discreetly while walking, touring, or working.
Besides the DingTalk A1 recording card, DingTalk also launched an AI badge. It looks like a slightly thicker ID badge, with a built-in microphone, battery, and communication module. Sales or service personnel wearing it can discreetly record entire conversations with customers during daily work, automatically generating a customer report including customer profile, key concerns, and follow-up suggestions.
AI glasses are also useful tools in meetings. When you have an important speech, public address, or corporate roadshow, wearing a pair of AI glasses brings your speech prompts right before your eyes. AI glasses also support real-time multilingual mutual translation, making them a “magic tool” for international exhibitions , international conference and negotiations.
According to Counterpoint’s “Global Smart Glasses Shipments Tracker” report, global smart glasses market shipments in the first half of 2025 increased by 110% year-on-year. Meta Ray-Ban accounted for over 70% of sales, with the rest captured by new entrants like Xiaomi and TCL RayNeo—RayNeo launched the RayNeo V3 series, Xiaomi launched its first Xiaomi AI Glasses, Thunderobot launched AURA Smart Glasses, and Quark’s AI glasses are also on the way.
Data from iResearch’s “2024 China Smart Meeting Tools Market Research Report” shows that the scale of China’s smart meeting tools market reached 12.36 billion yuan in 2024, a year-on-year increase of 35.7%. It is expected that by 2025, the penetration rate of AI in the enterprise meeting and collaboration tools market will reach 60%. 83.2% of enterprises believe that AI tools like automatic meeting minutes can reduce labor costs by over 30%. Real-time speech transcription, intelligent noise cancellation, and multi-scenario adaptation have become core demands for meetings.
02 AI Hardware Rakes in the Cash
What AI brings to hardware is not simply adding an “intelligent” function, but fundamentally reconstructing the value logic, profit model, and ecosystem positioning of hardware products.
In the past, the hardware itself was the end point of value; manufacturers sold physical functions, like the noise cancellation of headphones or the recording clarity of a recorder. iFlytek’s recorder launched a lifetime free transcription service, using free software services to boost hardware sales.
Now, new hardware with AI functions not only commands higher pricing, but the AI hardware itself evolves into a carrier and entry point for services. Users are not just buying a device; they are gaining access to a continuously evolving set of capabilities driven by large models and Agents. Hardware companies design AI as a subscription-based value-added service, obtaining sustained, high-margin recurring revenue (ARR) through monthly or annual fees, shifting from “selling hardware” to “hardware + subscription services.”
Plaud, the definer of AI recording cards, prices its overseas hardware at $159, with AI services costing no less than $240 per year. Plaud founder Xu Gao revealed that Plaud’s annualized revenue is approaching $250 million, with about half coming from annual AI subscription services.
After entering the domestic market in September, Plaud also continued the subscription model: a free membership (300 minutes of free transcription per month and limited summary templates), a 299 RMB/year professional membership (1200 minutes/month transcription + over 3000 global summary templates), and an 899 RMB/year premium membership (unlimited minutes + over 3000 global summary templates).
The Insta360 Wave is priced at 2198 RMB, bundled with 300 minutes of audio-to-text service. It also offers an advanced AI service for 58.17 RMB/month, providing 900 minutes of transcription, AI summarization, speaker diarization, industry-specific lexicon, and AI Q&A. Exceeding the transcription limit allows purchasing additional minutes incrementally.
Future Intelligence’s iFlytek AI Meeting Headphone Pro 3 has an official price of 1499 RMB, targeting office and business scenarios with AI-driven features and master tuning. It hasn’t launched AI subscription services domestically yet but has corresponding AI subscriptions overseas, offering tiered plans based on different transcription minute allowances. In the view of CPO Liu Da, the domestic payment habit is still being cultivated: “Our post-80s and 90s generations grew up with free services; their willingness to pay for services is strengthening. The post-00s might be more willing to pay for services.”
In this process, choosing B-end or C-end customers becomes a watershed in the business model. We note that Future Intelligence currently focuses on To C; Plaud primarily serves the C-end but has started engaging with B-end clients; Insta360 Wave caters to both; DingTalk offers different products for different customer segments.
Liu Da mentioned that a large automotive 4S chain store had proposed a customization request, hoping to use AI headphones to record salesperson-customer interactions and integrate the data with internal OA systems and databases. Liu Da believes that the To B business requires customization for enterprises. While technically feasible and capable of generating decent revenue, it could cause the company to lose focus and sacrifice product and technological differentiation.
Plaud mainly serves C-end users overseas, targeting roles like lawyers, finance, healthcare, and education that involve extensive dialogue-based work. Domestically, while serving the C-end, it is also engaging with B-end users, such as healthcare clients, offering some lightweight customization, but To C remains the primary focus.
The Insta360 Wave, launched at the end of September, targets both B-end and C-end. An internal Insta360 staff member stated that a version specifically designed for enterprises is still in planning and development. For the currently released version, if an enterprise purchases it and binds the devices to an account managed by the enterprise, the audio and documents will be stored under that enterprise-managed account. Subsequent plans for the enterprise version include designing for employee usage scenarios, allowing employees to control and delete their own recordings and AI summary files.
DingTalk simultaneously launched the C-end oriented AI recording card DingTalk A1 and the B-end oriented AI badge. The recordings and transcripts from the former are synchronized within the DingTalk app. The latter has predefined workflows; besides providing individual users with summaries, it regularly aggregates all team recording data and automatically pushes summaries to relevant leaders. Enterprise customers can also customize push times, management permissions, etc., within the workflow.
We observe that To C revenue relies on product premium and subscription services, while To B revenue relies on solutions and volume procurement.
The core of the To C model is product-driven, requiring manufacturers to adhere to user experience and data privacy while deeply cultivating vertical scenarios. Hence, Future Intelligence customizes Agents for the legal细分领域, and Plaud develops “non-interruptive” smart tagging features. To B, however, must solve enterprise pain points, integrate into enterprise workflows, and undergo corresponding customization. Once embedded, it easily creates stickiness and high switching costs.
The choice between To B and To C essentially answers “Who is your payer, and who do you ultimately want to become?” Both have pros and cons, and both can birth large companies. Various enterprises are making explorations based on their own DNA, market timing, and long-term vision.
Furthermore, considerable profit margins have attracted a rush of players, leading to simultaneous price wars and ecosystem wars, rapidly heating up the competition arena.
Plaud defined the category with first-mover advantage and product innovation, while DingTalk entered strongly with an almost halved price (499 RMB), deeply bundling hardware with its office ecosystem to achieve a closed-loop experience like one-click conversion of meeting minutes to to-do items. Huaqiangbei white-label manufacturers, leveraging China’s powerful supply chain, drove prices down to 120-150 RMB, quickly capturing the price-sensitive market. This made AI recording cards transform from a “thousand-yuan” novelty to a “hundred-yuan” mass-market product in a very short time.
The competition for AI headphones is even fiercer. In the mass market, AI headphones priced below 500 RMB account for 60.1% of sales. The brand Sanga (SANAG) captured the top spot in online sales in the first half of 2025 with an affordable average price of 359 RMB. In the high-end market (above 1500 RMB), iFlytek and Future Intelligence performed突出, focusing on vertical scenarios like office meetings and building moats through high-precision transcription, multilingual translation, meeting assistance, and other depth functions.
As AI capabilities converge (all promoting “recording + transcription + summary + to-do generation”), to escape homogenized involution, manufacturers like Plaud and Future Intelligence are competing on more precise and professional AI services—no longer “is my transcription accuracy 1% higher than yours?” but “can I provide industry-specific, role-specific dedicated workflows for doctor consultations / lawyer depositions / product manager post-mortems?”
Both Plaud and Future Intelligence are planning diverse hardware products, not just to sell more devices, but to deploy “senses” across the user’s entire work scenario, collecting richer contextual information, enabling AI services to proactively predict, precisely intervene, and ultimately provide complete solutions.
03 The Two Faces of AI Tools
The proliferation of AI meeting tools acts like a wedge, deeply embedding into the fabric of organizations, prying open and restructuring work models and management logic.
Among these, data ownership becomes a key point for AI application: users control data in the To C model, while enterprises need to manage and control data in the To B model.
For high-turnover positions like store staff and 4S shop assistants, some companies, aiming to standardize business processes and optimize management, use tools like AI badges, AI headphones, and AI recording cards to regulate employee behavior. This also provides employees a shortcut to “skip the lengthy professional training stage and directly embed into modernized processes.”
For example, an AI badge can automatically monitor if staff use polite language like “Hello” or “Please have a seat,” and identify cold, harsh phrases like “Can’t do it” or “Look yourself.” It can also automatically check if an employee’s script follows standard procedures, contains违规promises, or omits key information, and evaluate service attitude.
However, this leads to blurred boundaries in data collection, potentially over-collecting sensitive information unrelated to work, causing AI tools to slide from “efficiency tools” towards “monitoring tools.” In management, ethical debates about means are sometimes superseded by the pursuit of effective results. As long as the final outcome is “improved efficiency” and “standardized processes,” the nature of the means can be tolerated or even encouraged at a certain stage, even if this empowerment is narrow and instrumental (only knowing how to act as the AI dictates).
This reveals the subtle workplace ecology under AI augmentation. Although sharing the concept of “AI empowerment,” one type serves the individual (“Assistant AI”), while the other serves the organization (“Supervisor AI”). The former is voluntarily purchased and used by employees, primarily enhancing personal productivity; the latter is uniformly distributed and managed by the enterprise, aiming to make the “black box” of frontline employee (e.g., sales, customer service) interactions with customers transparent and datafied, serving enterprise management and decision-making.
Efficiency and supervision are the two sides of the same AI tool coin. While it speeds up workflows, it also quietly constructs an unprecedented digital supervision system, raising concerns and challenges regarding data security and employee privacy.
This is the workplace on the eve of AGI’s arrival, a mix of deep anxiety, active transformation attempts, and a re-examination of individual value: with one hand we create people’s desire to escape mechanical labor, and with the other we are building unprecedented, sophisticated machines that datafy people and their thoughts.
The controllability pursued by organizations and the autonomy craved by individuals are both amplified to the extreme within the AI framework.
The choice of AI tools by each enterprise is essentially a vote for the “future work model” they believe in, and participation in defining the rules and boundaries of the upcoming new era of human-machine collaboration.
