Alibaba

ByteDance

CATL

Huawei

Meituan

Tencent

Xiaomi

Most Followed

Market News

Listings (IPO)

Earnings Reports

Venture Roundup

Big Tech

Automotives

Retail and Commerce

Collections

News

Features

Insights

Press Releases

Anta

DeepSeek

Geely

Li Auto

Mixue Bingcheng

Pop Mart

Going Global

Latest

Collaborate

Earnings

Auto

Retail & Commerce

StepFun has raised more than RMB 5 billion (USD 700 million) in a Series B+ funding round, according to <a href="https://news.pedaily.cn/202601/560382.shtml">PEdaily</a>. The deal ranks among the largest private financings in China’s <a href="/industry/ai">artificial intelligence</a> sector so far this year.
The round drew a mix of state-owned funds, industrial players, and financial investors. New participants include the Shanghai State-owned Capital Investment (SSCI) Leading Fund, China Life Private Equity Investment, PDVC, Xuhui Capital, Wuxi Liangxi Fund, Xiamen ITG Group, and Huaqin Technology. Existing backers <a href="/company/tencent">Tencent</a>, Qiming Venture Partners, and 5Y Capital also took part.
Founded in Shanghai in 2023, StepFun last raised capital just over a year ago, completing a Series B round in December 2024 with participation from SSCI, Qiming Venture Partners, Tencent, and 5Y Capital.
The latest raise comes as China’s AI sector shows signs of consolidation. Several leading AI developers, including Z.ai (Zhipu AI) and <a href="/company/minimax">MiniMax</a>, have turned to public markets, while private funding has increasingly concentrated around a few perceived frontrunners. Data from Tracxn show that Chinese AI firms closed fewer funding deals in 2025 than a year earlier, though average deal sizes rose. Large technology companies such as <a href="/company/alibaba">Alibaba</a> and Tencent have also become more selective, appearing to favor in-house development over external investments.
<style>@media (min-width:769px){.embed-container{height:480px!important}}@media (max-width:768px){.embed-container{height:540px!important}}</style>
<div class="embed-container"><iframe src="https://flo.uri.sh/visualisation/27373627/embed" frameborder="0" allowfullscreen> frameborder="0" scrolling="no" sandbox="allow-same-origin allow-forms allow-scripts allow-downloads allow-popups allow-popups-to-escape-sandbox allow-top-navigation-by-user-activation"></iframe></div>
StepFun’s round stands out for the range of its investors:
<ul>
<li aria-level="1">China Life Private Equity Investment, the investment arm of China Life Insurance, is known for its conservative approach to risk, making its participation noteworthy.</li>
<li aria-level="1">Xiamen ITG Group’s investment is reportedly its first in a large-model developer, signaling intensifying competition among local governments to gain AI exposure.</li>
<li aria-level="1">Huaqin Technology’s involvement points to potential hardware-software integration, as the smartphone original design manufacturer supplies brands including Oppo and Vivo and produces devices such as tablets, laptops, wearables, and automotive electronics.</li>
</ul>
The company has also expanded its leadership team. Yin Qi, CEO of Megvii and founder of Afari, has been appointed chairman of StepFun. He joins CEO Jiang Daxin, chief scientist Zhang Xiangyu, and CTO Zhu Yibo as the company accelerates its work in large model development.
StepFun’s strategy centers on Step 3, its flagship large language model, alongside development of multimodal systems and cloud-edge integration. Step 3 is designed to process text and visual inputs simultaneously, a capability known as multimodal reasoning, and is positioned around high decoding efficiency and broad hardware compatibility. Benchmark results have prompted comparisons with domestic peers, including DeepSeek’s R1.
Beyond cloud-based models, StepFun is expanding into applications that integrate AI directly into devices. In the automotive sector, <a href="/company/geely">Geely’s</a> Agent OS system, installed in vehicle models such as the Galaxy M9, is powered by StepFun’s end-to-end voice large model. StepFun expects its models to be deployed in more than one million vehicles this year, according to <a href="https://news.pedaily.cn/202601/560382.shtml">PEdaily</a>.
StepFun’s end-to-end voice large model, enables emotionally intelligent voice interactions, adapting tone and content based on user emotion and intent.
The company has also explored vision-language-action models that link perception and decision-making, and it continues to delve into on-device AI. Its Step-GUI series, released in November 2025, is described as capable of achieving state-of-the-art performance while maintaining strong general capabilities when running directly on-device.
Taken together, the scale of StepFun’s latest raise and the breadth of its backers underscore a shift in China’s AI funding landscape, where capital is increasingly flowing to a smaller number of companies seen as having the technology, leadership, and partnerships to compete at scale.

China’s investors double down on AI frontrunners as StepFun raises RMB 5 billion

Artificial Intelligence

Deep Tech

Foundation and Large Models

China

Chinese chipmaker Sunrise has raised nearly RMB 3 billion (USD 420 million) in funding over the past year, according to <a href="https://mp.weixin.qq.com/s/aDcwDtjt1f-LW2vB7LCx0w">China Venture</a>. The report did not specify whether the capital was raised in a single round or across multiple financings.
The investor base spans multiple categories. Industrial and strategic backers include Sany Group’s HX Capital, 4Paradigm, CP Robotics, and GCL Technology. Venture capital and private equity firms participating include IDG Capital, Gaorong Ventures, Infini Capital, CICC Capital, Puhua Capital, Green Pine Capital Partners, and Soul Capital. Financial institutions include E Fund Management and ICBC Investment, while state-affiliated capital includes CCT Mixed Reform Fund.
Sunrise will use the funding primarily to develop its next generation of <a href="/industry/ai">artificial intelligence</a> chips, expand large-scale production, and build the software tools and partner ecosystem required to deploy its technology commercially.
Founded in 2020, Sunrise designs chips for AI inference, the process of running trained AI models to generate outputs such as answers, images, or text. Unlike model training, which occurs intermittently, inference takes place every time an AI system is used. As a result, inference has become a growing cost center for companies operating AI services, with chip efficiency increasingly shaping margins and user experience as usage scales.
Rather than developing general-purpose processors that handle both training and inference, Sunrise has focused exclusively on inference workloads. The company believes this narrower approach allows it to optimize for lower power consumption and lower cost per AI response, priorities for data centers and enterprise customers deploying large models at scale.
The company was spun out of SenseTime, which spent several years developing chip technology internally before separating the business.
whereRelatedArticle
Sunrise is led by two co-CEOs with long industry backgrounds. Wang Yong has more than two decades of experience in chip design, including roles at AMD and <a href="/company/baidu">Baidu’s</a> Kunlun chip unit. <a href="https://kr-asia.com/baidu-veteran-joins-sensetime-spinoff-sunrise-to-compete-in-ai-chip-race">Wang Zhan</a>, a founding member of Baidu, oversees commercialization and fundraising. The company is estimated to have a headcount of about 300 people.
Sunrise has already brought two chips into mass production. Its first product, S1, targets image and video recognition tasks. A second chip, S2, supports large model inference and is compatible with <a href="/company/nvidia">Nvidia’s</a> CUDA ecosystem, which lowers switching costs for customers. The company plans to release a third chip, S3, in 2026. Sunrise has reportedly said the chip will focus on reducing the cost and energy required to run large AI models, with support for FP8 and FP4 low-precision inference.
Despite its short operating history, Sunrise has notably attracted backing from industrial players, venture firms, financial institutions, and state-linked funds. The fundraising reflects investor interest in both the company’s leadership and a broader push to lower the cost of running AI models in an increasingly competitive inference chip market.

SenseTime spinout Sunrise raises RMB 3 billion for next-gen AI inference chips

SenseTime

East Asia

… overlooked assets in the AI race. While Alibaba Cloud’s infrastructure and its open-source Qwen model have drawn attention, T-Head fills a remaining gap in Alibaba’s broader AI strategy.
The timing also coincides with a surge in investor interest in Chinese chipmakers, which has driven a flurry of public listings.
In August, Cambricon Technologies, one of the few semiconductor firms listed on China’s A-share market, briefly surpassed Kweichow Moutai in intraday market capitalization, despite reporting revenue of RMB 4.6 billion (USD 644 million) in the first half of 2025. Its valuation has since stabilized at around RMB 600 billion (USD 84 billion).
That rally helped trigger a broader rush to market. Since late last year, several Chinese GPU companies have gone public in quick succession, further inflaming an already overheated semiconductor sector:
<ul>
<li aria-level="1">Moore Threads became the fastest company to receive IPO approval on the STAR Market, completing the process in 88 days before listing on December 5, 2025. It is now valued at about RMB 300 billion (USD 42 billion), after reporting RMB 780 million (USD 109.2 million) in revenue over the first nine months of 2025.</li>
<li aria-level="1">MetaX followed 12 days later, debuting on the STAR Market with shares rising nearly sevenfold on its first day of trading. The company now has a market capitalization exceeding RMB 240 billion (USD 33.6 billion) and reported nine-month revenue of RMB 1.24 billion (USD 173.6 million).</li>
<li aria-level="1">On January 2, Biren Technology listed in Hong Kong, becoming the first domestic GPU company to do so. Its market value now exceeds RMB 70 billion (USD 9.8 billion), despite half-year revenue of RMB 58.9 million (USD 8.2 million).</li>
<li aria-level="1">A week later, Iluvatar CoreX also debuted in Hong Kong, reaching a valuation of roughly RMB 40 billion (USD 5.6 billion) on first-half 2025 revenue of RMB 324 million (USD 45.4 million).</li>
<li aria-level="1">Another contender, Enflame Technology, has completed IPO counseling and is preparing a STAR Market listing. Its prospectus shows half-year revenue of RMB 2.83 billion (USD 396.2 million).</li>
</ul>
Together, these five companies command a combined market capitalization of about RMB 1.25 trillion (USD 175 billion), roughly 4% of Nvidia’s current valuation.
T-Head, by contrast, operates across a broader range of computing segments, giving it a stronger base to deliver end-to-end solutions. Already positioned as a market leader in several categories, its eventual listing is likely to further influence how investors assess China’s domestic semiconductor sector.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3650412256731265">article</a> was written by Qiu Xiaofen for 36Kr.

Alibaba clears public listing plan for chip unit T-Head as China’s semiconductor push accelerates

Semiconductors

… the Li Mega have fallen below that level.
Li Auto is working to add Sunwoda as a second battery supplier for the Li i6 to ease delivery constraints. 36Kr reported that Sunwoda can now supply up to 3,000 battery packs per month, supplementing output from CATL (Contemporary Amperex Technology) and improving delivery capacity.
Li Auto is also restructuring its offline retail network. Some showrooms are appointing store managers to lead demonstrations, raising expectations for product knowledge and sales capability among frontline staff. Employees who fail to meet those standards may be reassigned, according to people familiar with the changes.
At the same time, a number of mall-based stores are set to close as leases expire, reflecting weak performance. One insider said many of these locations were opened rapidly during Li Auto’s expansion phase and no longer meet efficiency benchmarks.
Whether these adjustments will be enough remains an open question. For Li Auto, 2026 is shaping up to be both its most turbulent and most consequential year yet, a test of whether it can rebuild momentum and restore investor confidence.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3647785422376584">article</a> was written by Xiao Man for 36Kr.

Li Auto targets 40% growth in 2026 as it steers its focus back to range-extended cars

Transportation

Electric Vehicles

Mobility

… assigning tasks based on employees’ skills and working styles. Until then, DeepWisdom prioritizes generalists. Fewer than 15 core employees operate across frontend, backend, algorithms, and testing to minimize handoffs.
Many hires have come from the company’s user community and internal hackathons rather than traditional recruiting. Wu believes performance under real constraints reveals more than resumes.
DeepWisdom is preparing to open its first overseas office in Silicon Valley, in a building informally known as “the lucky office,” which has previously housed companies such as Google, PayPal, and Logitech. The building also happens to share a wall with Stanford University.
“For a long time to come,” Wu said, “the top tier of AI companies will still be competing on human efficiency.”
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3640426144812417">article</a> was written by Zhou Xinyu for 36Kr.

From MetaGPT to Atoms: DeepWisdom leads China’s push into “vibe coding”

Ahead of the Detroit Auto Show, US President Donald Trump paid a visit to the city, declaring that he was "standing up for the American auto worker like no president has ever stood up before."
Stocks tied to the automotive sector were not in a welcoming mood. Between January 9 and 15, General Motors slid 3%, Ford Motor fell 4%, and Stellantis dropped 11%, due in large part to announced losses related to <a href="/industry/ev">electric vehicles</a> following the Trump administration’s turn away from EVs.
At the show, the main topic of conversation on the floor was not Trump, but the question of when Chinese automakers will arrive in the US.
The discussion was spurred by the CES technology show held a week earlier in Las Vegas.
The event had enough Chinese automakers to feel like a car show, despite the fact that they do not make or sell vehicles in the US An executive at major Chinese EV company <a href="/company/geely">Geely</a> said the company is weighing an expansion into the US, with an announcement expected to come in the next 24–36 months.
Speaking in Detroit on January 13, Trump said to "let China come in," indicating he would not block them from manufacturing in the US.
The show, by contrast, felt as if it was keeping China out. EVs were scarce, and many US automakers put non-electric models front and center. But consumers seem to have their sights set further out.
<a href="/company/byd">BYD</a> cars are everywhere in Brazil, said an engineer in the country who came to Detroit to see GM models that use parts he worked on. Chinese automakers could start making cars in the US, the engineer said.
BYD already dominates EV sales in Brazil, accounting for 72% of the total in 2025. It opened its first South American plant there in July.
Joe McCabe, CEO of AutoForecast Solutions, argued that it is only a matter of time before Chinese players come into the US, saying that it is consumers who decide whether products are good.
Much like how Japanese and South Korean automakers emphasized the quality of their cars when they first entered the country, Chinese companies are scrambling to figure out how to connect with consumers to gain market share, McCabe said. That led them to exhibit at CES despite the tensions between China and the US.
There are vehicles in the US that consumers know well without realizing their connection to China. A prime example is Sweden's Volvo, which is owned by Geely. The Chinese company is considering using Volvo production facilities in the US for its expansion into the country.
One word Trump kept coming back to in Detroit was "affordable," which in the US denotes cars prices around USD 20,000–30,000.
Although the US now levies a 100% tariff on Chinese vehicles, automakers there have models that could double in price and still come in below USD 30,000. The fact that they are produced in China is what makes those low prices possible, but if these companies can manufacture in the US as well, they could pose a threat.
According to Peipai Zhao, president of the North American operations of Chinese LiDAR (light detection and ranging) sensor maker RoboSense, Chinese automakers considering entering the US are weighing three options: improving quality to attract consumers, partnering with US companies, or setting up a different company with no connection to China. Either way, affordability looks likely to be key.
This article first appeared on <a href="https://asia.nikkei.com/business/automobiles/chinese-evs-spot-on-ramp-to-us-market-with-affordability">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.

Chinese EVs spot on-ramp to US market with affordability

Longcheer Technology made its Hong Kong market debut on January 22, adding to a growing roster of mainland technology companies turning to the city’s capital markets.
The smart device manufacturer sold about 52.3 million shares in the offering, which was heavily oversubscribed, according to market filings. The share sale launched on January 14 at HKD 31 (USD 4) per share, with the stock rising beyond HKD 35 (USD 4.5) on its debut before retreating to close at HKD 32.1 (USD 4.1) on the first day of trading. Longcheer’s Hong Kong shares were trading at around HKD 31.2 (USD 4) at the time of publication, slightly above the initial offer price.
Media reports said Longcheer plans to use the IPO proceeds to expand production capacity, increase investment in R&amp;D, and pursue potential strategic acquisitions.
The listing marks Longcheer’s second public market debut, following its Shanghai Stock Exchange IPO in March 2024. In that offering, the company issued about 60 million shares at RMB 26 (USD 3.6) per share, raising roughly RMB 1.6 billion (USD 224 million). The Shanghai listing followed a prolonged effort to access China’s A-share market, after earlier attempts in 2015 and 2017 were halted.
Longcheer’s Hong Kong debut adds to a steady stream of mainland technology IPOs in the city, including listings by Biren Technology, <a href="/company/minimax">MiniMax</a>, and Z.ai, also known as Zhipu AI. The trend reflects Hong Kong’s <a href="https://kr-asia.com/hong-kongs-ipo-market-roars-back-to-life-leading-the-world-in-2025-listings">growing role as an alternative venue for public fundraising</a> amid ongoing geopolitical uncertainty. Mainland inflows have been a key factor in pushing the Hong Kong Stock Exchange back to the top of global IPO rankings last year, ahead of Nasdaq and the New York Stock Exchange, which were long considered benchmark destinations for Asian companies seeking public capital.
whereRelatedArticle
The company has prior experience with offshore markets. It was listed on the Singapore Exchange in 2005 under the ticker BJL, before being taken private and delisted in late 2020.
Founded as a provider of smart technology products and solutions, Longcheer focuses on the design, R&amp;D, manufacturing, and servicing of terminal devices, including smartphones, tablets, wearables, home appliances, and mixed reality hardware. It operates as an ODM, or original design manufacturer, for major Chinese brands such as <a href="/company/huawei">Huawei</a>, <a href="/company/xiaomi">Xiaomi</a>, and <a href="/company/honor">Honor</a>.
According to a Frost &amp; Sullivan study on the smart device manufacturing industry, Longcheer ranked as the world’s largest smartphone ODM by shipments, delivering 172.9 million units and accounting for a 32.6% global market share. It ranked second in consumer electronics ODM shipments, with 219.1 million units and an estimated 22.4% share. In smart glasses manufacturing, Longcheer placed second globally, shipping 2.1 million units for a 21.7% share, and counts Ray-Ban Meta among its customers.
Chinese media outlet <a href="https://finance.sina.com.cn/stock/t/2026-01-22/doc-inhiecca2839053.shtml">Sina</a> reported that Longcheer operates four main manufacturing bases: two in Huizhou and Nanchang, along with overseas facilities in Vietnam and India. The company also maintains regional offices in the US, South Korea, and Japan.
Financially, Longcheer reported revenue of RMB 29.3 billion (USD 4.1 billion) in 2022, RMB 27.2 billion (USD 3.8 billion) in 2023, and RMB 46.4 billion (USD 6.5 billion) in 2024. Gross profit rose from RMB 2.4 billion (USD 336 million) in 2022 to RMB 2.7 billion (USD 378 million) in 2024, although gross margin declined to 5.8% in 2024 following the revenue surge.
Net profit totaled RMB 561.5 million (USD 78.6 million) in 2022, RMB 602.7 million (USD 84.4 million) in 2023, and RMB 493.4 million (USD 69.1 million) in 2024.
For the first nine months of 2025, Longcheer reported revenue of RMB 31.3 billion (USD 4.4 billion), down 10.3% year-on-year, while net profit rose 17.7% to RMB 507 million (USD 71 million). Net profit excluding non-recurring items fell 18.4% to RMB 271 million (USD 37.9 million) over the same period.
In the third quarter of 2025, revenue declined 9.6% year on year to RMB 11.4 billion (USD 1.6 billion). Net profit increased 64.5% to RMB 152 million (USD 21.3 million), while net profit excluding non-recurring items rose 56.0% to RMB 92.4 million (USD 12.9 million).

Chinese smart device maker Longcheer joins wave of Hong Kong listings

Hardware

Internet of Things

Hong Kong

Since <a href="/company/pop-mart">Pop Mart’s</a> stock hit a record high of HKD 339.8 (USD 43.6) per share on August 26 last year, it has fallen more than 40% over the past five months. Industry analysts largely attribute the decline to the fading premium surrounding its once-dominant IP, Labubu.
Data from the Dewu app shows that the Labubu “Big Into Energy” blind box now trades at an average price of RMB 100.25 (USD 14), down 23% from its 90-day peak. Compared with its record high of RMB 639.32 (USD 89.5) last June, shortly after Pop Mart launched the “Wacky Mart” series, the decline exceeds 84%.
In some cases, prices have fallen even further. On Tmall’s subsidy-driven channel, unopened blind boxes are selling for as little as RMB 86 (USD 12), below Pop Mart’s official retail price.
Anxiety in the consumer market has spilled into capital markets. According to data from Wind, the number of Pop Mart shares sold short has risen sharply since August 2025, at one point nearing 10,000 shares.
On January 19, Pop Mart announced plans to repurchase 1.4 million shares at prices ranging from HKD 177.7–181.2 per share (USD 22.8–23.2), for a total outlay of HKD 251 million (USD 32.2 million). The buyback, its first since 2024, was widely interpreted as a signal of confidence from management.
Fueled by celebrity exposure, Pop Mart has become one of China’s most successful global consumer exports, with Labubu emerging as its standout asset. According to the company’s earnings report, The Monsters series, which includes Labubu, generated RMB 4.81 billion (USD 673.4 million) in revenue in the first half of 2025, accounting for nearly one-third of total revenue.
Around the same period, however, Pop Mart began tightening control over the secondary market by cracking down on scalpers and increasing supply. In a media interview, founder Wang Ning said that after September 2024, Labubu’s monthly sales could reach ten million units worldwide. “We’re trying in every possible way to keep it from being maliciously speculated on,” he said.
Dewu data supports that assessment. After “Wacky Mart” launched, Pop Mart restocked Labubu vinyl and plush toys six times. The expanded production quickly pushed down prices of the mini Labubu edition, undermining scalpers’ hoarding strategies.
At the same time, Pop Mart has dialed back online hype by deploying robotic vending machines to steer traffic toward physical stores. In retrospect, the company appears to have largely regained control over pricing.
As a result, consumers can now buy standard editions at or near retail prices, while hidden or limited-edition versions command modest, more rational premiums. From a business perspective, the strategy helps Pop Mart protect margins while maintaining the credibility of its premium IP portfolio.
In a research note, Deutsche Bank said Labubu’s production capacity expanded from 10 million units in the first half of 2025 to a monthly average of 50 million by year’s end. The firm estimated the expansion could lift Pop Mart’s adjusted net profit to RMB 14.5 billion (USD 2 billion) in 2025.
Still, Labubu’s rapid rise and correction have reignited debate over Pop Mart’s reliance on blockbuster IPs and the sustainability of such product cycles.
Notably, even as resale premiums have declined, scarcity has increased in select collaborations. In October last year, Labubu partnered with LVMH’s heritage brand Moynat. Shortly afterward, Twinkle Twinkle collaborated with bubble tea chain Heytea. Around the same time, Pop Mart hired Wu Yue, former president of Louis Vuitton China, a move widely seen as signaling deeper diversification and a push upmarket.
On the same day it announced the buyback, Pop Mart also unveiled a Molly-themed smartphone developed with <a href="/company/honor">Honor</a>. Based on the Honor 500 specifications, the device features Molly artwork on the back, along with a customized interface, boot animation, and photo watermark built around the IP.
By most measures, Pop Mart remains among the industry’s most efficient creators of blockbuster IP. In the first half of last year, The Monsters, Molly, Skullpanda, Crybaby, and Dimoo each generated more than RMB 1 billion (USD 140 million) in revenue. Thirteen additional artist IPs, including Hirono, Twinkle Twinkle, and Hacipupu, surpassed RMB 100 million (USD 14 million) in sales. Even so, Pop Mart’s ambition to become a global, IP-centric platform will hinge on its ability to balance scale with perceived scarcity.
Pop Mart is not alone in using buybacks to steady investor sentiment. Other Chinese consumer companies, including Kweichow Moutai and Midea Group, have also announced repurchase plans, reflecting expectations of a gradual recovery in consumption.
A telling comparison has emerged. As Labubu’s market price has converged toward retail, Feitian Moutai has, for the first time, traded at its official retail price of RMB 1,499 (USD 209.9). While the products serve very different consumers, both shifts suggest a correction of speculative excesses as supply and demand rebalance.
Pop Mart’s repurchase plan has received broad investor approval. Morgan Stanley said the company’s long-term growth story remains intact and projected net profit of RMB 12.6 billion (USD 1.8 billion) for 2025. Other analysts noted that the buyback could help establish a price floor, supported by expected strength in 2025 earnings and momentum into the first quarter of 2026.
As of publication, Pop Mart shares were trading at HKD 197.6 (USD 25.3) at the market open on January 22, up more than 9% since the January 19 buyback announcement.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3648979523706501">article</a> was written by Xie Yunzi for 36Kr.

Pop Mart launches share buyback as Labubu craze fades

Retail

Consumer Economy

<a href="/company/lark">Lark</a>, also known as Feishu, is preparing to launch its first hardware product, a compact, bean-shaped recording device that uses <a href="/industry/ai">artificial intelligence</a>. Developed jointly with <a href="/company/anker">Anker Innovations</a>, the project marks an experimental move for a company that has, until now, focused exclusively on software.
According to product materials reviewed by 36Kr, the device weighs about ten grams and features a dual MEMS (microelectromechanical system) microphone array. With its charging case, the total weight reaches roughly 48 grams. It supports both Bluetooth and Wi-Fi transmission and is designed for hands-free, always-on voice capture.
AI-powered recording devices began gaining traction in 2025, when Plaud introduced its <a href="https://kr-asia.com/can-plauds-overseas-first-strategy-give-it-an-edge-in-chinas-crowded-voice-recorder-market">AI notetakers</a>, triggering a wave of similar products from Chinese manufacturers. Despite the surge, few major technology companies have entered the category with dedicated hardware.
Founded in 2017, Lark has never released a hardware product. Several people close to the company told 36Kr that the recorder is viewed internally as a trial project, with the product team continuing to prioritize software development.
From 2024 onward, consumer AI hardware began diversifying rapidly, with recorders, cameras, and smart glasses emerging as alternative entry points for interacting with AI beyond smartphones. Each form factor reflects an effort to make AI more ambient and continuously accessible.
Instead of adopting the card-style design favored by many competitors, Lark and Anker chose a smaller, bean-like form. Card-shaped recorders can attach magnetically to the back of a phone, but their size can limit comfort and portability. The bean-shaped design is intended to be lighter and suitable for extended, all-day use.
Users can clip the device to a collar or pocket, or attach it magnetically to a phone. The goal is to lower the friction of recording, allowing conversations to be captured during meetings, client visits, or commutes without opening an app or handling a phone.
Despite its small size, the device supports live meeting notes. It can generate real-time captions and AI-produced summaries, accompanied by simple visual elements such as bar charts.
Portability combined with contextual awareness is positioned as a core differentiator. According to 36Kr, the device and its charging case offer more than 32 hours of battery life on a full charge. The recorder can store up to eight gigabytes or roughly 250 hours of audio, and supports fast charging, delivering about two hours of recording from a ten-minute charge.
<h3>Competing for the next context gateway</h3>
Since 2025, advances in foundational AI models, particularly multimodal systems, have renewed momentum in voice-based interaction. In China, established players such as <a href="/company/iflytek">iFlytek</a> and Sogou have upgraded voice input tools. Meanwhile, <a href="/company/bytedance">ByteDance's</a> Doubao has introduced its own input method, embedding large model capabilities into everyday typing. Other startups, including Remio, are developing products aimed at managing users' daily context across work and personal life.
Across both input methods and wearables, the competition centers on becoming the memory gateway for AI assistants. Systems that capture richer context can deliver more personalized responses and are harder for rivals to replicate.
Per 36Kr, Lark’s device supports an end-to-end workflow from live meetings through post-meeting follow-up. During meetings, users can view AI-generated highlights in real time. Afterward, the system produces full transcripts with speaker identification, action items, and key decisions.
All recordings are stored directly in Lark’s Knowledge Base, where speech is converted into structured text. Users can then query past content conversationally through Lark’s AI assistant, for example, asking what concerns a client raised in a recent meeting. The system retrieves relevant context and returns a concise answer, reducing post-meeting administrative work.
Lark was built within ByteDance and, since its launch in 2017, has focused on high-tech industries. It has refined tools for document collaboration, transcription, and meeting support to build an early reputation for productivity.
That positioning has carried into the current AI cycle. Lark is widely used by AI-focused companies including <a href="/company/deepseek">DeepSeek</a>, <a href="/company/minimax">MiniMax</a>, Moonshot AI, and Z.ai (Zhipu AI), all of which rely on the platform for internal collaboration.
Such organizations, typically with heavy procedural and documentation demands, are also the most likely to benefit from an AI recording device.
The design logic behind Lark’s hardware reflects longstanding enterprise requests, including speaker-labeled summaries, automatic extraction of to-do items, and searchable historical records.
Industry observers generally agree that while AI recording hardware can be replicated quickly, durable differentiation depends on continued engineering refinement. Improvements in meeting templates, transcription accuracy, and post-processing workflows are likely to matter more than form factor alone.
Here, Lark holds an inherent advantage. The device runs on ByteDance's Doubao large model, whose multimodal capabilities are widely regarded as competitive within the field. Its ability to parse complex dialogue and generate accurate transcripts directly affects the quality of meeting summaries, creating a technical edge that may be difficult for competitors to match in the near term.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3645215315267202">article</a> was written by Deng Yongyi for 36Kr.

Lark steps into hardware, developing an AI-powered recorder with Anker

Lark

Anker