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

<a href="/company/li-auto">Li Auto’s</a> fourth-quarter results for 2025 suggest stabilization after a difficult year, but not a full recovery.
The Chinese <a href="/industry/ev">electric vehicle</a> maker returned to a slim quarterly profit in the December period. It posted net income of RMB 20.2 million (USD 2.9 million) after a third-quarter loss, while vehicle margin improved to 16.8% from 15.5%. Free cash flow also turned positive at RMB 2.5 billion (USD 363.3 million), compared with negative RMB 8.9 billion (USD 1.3 billion) in the previous quarter.
The year-on-year picture remained weak. Revenue fell 35% to RMB 28.8 billion (USD 4.2 billion), vehicle sales declined 36.1% to RMB 27.3 billion (USD 4 billion), and the company remained in the red with an operating loss of RMB 442.6 million (USD 64.3 million).
That makes the quarter more useful as evidence of operational control than of restored earnings power. Li Auto did show it could recover from a weak third quarter without sacrificing gross margin entirely, with gross margin rising to 17.8% from 16.3% in Q3. The comparison with last year remains more important. Gross margin was 20.3% in the fourth quarter of 2024, and Q4 2025 operating performance remained well below that prior-year base.
The company’s own disclosures suggest caution when interpreting the quarter-on-quarter improvement. Li Auto said the sequential rise in vehicle margin was mainly due to the estimated Li Mega recall cost recorded in the third quarter, partly offset by a lower average selling price caused by product mix changes after the start of Li i6 deliveries. In other words, part of the rebound reflected a cleaner comparison base rather than a clear improvement in underlying demand or pricing.
Expenses provided some support. Q4 operating expenses fell 1.3% from the prior quarter, helped by a 4.4% decline in selling, general, and administrative expenses to RMB 2.6 billion (USD 377.8 million). R&amp;D spending remained high at RMB 3 billion (USD 436 million) as Li Auto continued investing in <a href="/industry/ai">artificial intelligence</a> and product development. That mix aligns with management’s current message: the company is trying to preserve technology spending while tightening its sales and operating model.
<div class="tradingview-widget-container">
<script type="text/javascript" src="https://s3.tradingview.com/external-embedding/embed-widget-symbol-overview.js" async>{"lineWidth":2,"lineType":0,"chartType":"candlesticks","fontColor":"rgb(106, 109, 120)","gridLineColor":"rgba(46, 46, 46, 0.06)","volumeUpColor":"rgba(34, 171, 148, 0.5)","volumeDownColor":"rgba(247, 82, 95, 0.5)","backgroundColor":"#ffffff","widgetFontColor":"#0F0F0F","upColor":"#22ab94","downColor":"#f7525f","borderUpColor":"#22ab94","borderDownColor":"#f7525f","wickUpColor":"#22ab94","wickDownColor":"#f7525f","colorTheme":"light","isTransparent":false,"locale":"en","chartOnly":false,"scalePosition":"right","scaleMode":"Normal","fontFamily":"-apple-system, BlinkMacSystemFont, Trebuchet MS, Roboto, Ubuntu, sans-serif","valuesTracking":"1","changeMode":"price-and-percent","symbols":[["NASDAQ:LI|1D"]],"dateRanges":["1d|1","1m|30","3m|60","12m|1D","60m|1W","all|1M"],"fontSize":"10","headerFontSize":"medium","autosize":true,"dateFormat":"MMM dd, yyyy","width":"100%","height":"480","noTimeScale":false,"hideDateRanges":false,"hideMarketStatus":false,"hideSymbolLogo":false}</script>
</div>
The transcript adds a clearer frame for 2026. Management is positioning the year around three levers: tighter store-level execution in its direct sales network, a new L-series cycle led by the all-new Li L9, and a steadier ramp in battery EV models including the i6, i8, and the upcoming i9. Founder and CEO Li Xiang said the company is still targeting 20% year-on-year sales growth in 2026, even as he described the period as the most competitive year yet in China’s high-end automotive market.
That ambition still runs ahead of the near-term numbers. Li Auto guided first-quarter 2026 deliveries of 85,000–90,000 vehicles and revenue of RMB 20.4–21.6 billion (USD 3.0 billion–3.1 billion), both below the level recorded a year earlier. Management is clearly asking investors to look beyond the current trough toward a new product cycle. The guidance, however, does not yet show a clear path toward recovery.
The balance sheet gives Li Auto room to make that argument. The company ended 2025 with RMB 101.2 billion (USD 14.7 billion) in cash, and Q4 operating cash flow was positive at RMB 3.5 billion (USD 508.6 million). That cushion matters because Li Auto is not attempting to cut its way back to growth. Management said 2026 R&amp;D spending should remain around RMB 12 billion (USD 1.7 billion), with roughly half tied to AI-related initiatives including chips, computing power, and autonomous driving.
Still, the full-year numbers show how much ground the company must recover. Total revenue fell 22.3% in 2025 to RMB 112.3 billion (USD 16.3 billion). Operating margin turned negative 0.5% from positive 4.9% in 2024. Net cash from operations swung to an outflow of RMB 8.6 billion (USD 1.2 billion) from an inflow of RMB 15.9 billion (USD 2.3 billion) a year earlier, and free cash flow dropped to negative RMB 12.8 billion (USD 1.9 billion) from positive RMB 8.2 billion (USD 1.2 billion).
The result leaves Li Auto with a credible stabilization story, but not yet a clear turnaround. The fourth quarter showed that margins and cash flow can recover when comparisons become cleaner and execution improves. What it did not yet show is that the company has rebuilt the earnings base it had a year earlier, or that the expected 2026 product cycle is already visible in reported growth.
Note: RMB figures are converted to USD at rates of RMB 6.88 = USD 1 based on estimates as of March 13, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Li Auto ends 2025 on a steady note, but turnaround remains in progress

East Asia

China

Transportation

Electric Vehicles

Mobility

Dreame's entry into the chip race is becoming increasingly multifaceted. In addition to the previously reported <a href="https://0722vps.aieehr01.top/dreame-backed-nxmind-enters-commercialization-with-tianqiong-chips-targets-orbital-computing">Tianqiong chip series</a>, NXMind, a chipmaker incubated by Dreame, outlined several additional chip development programs at this year's Appliance &amp; Electronics World Expo.
Fu Haiyang, head of NXMind, told 36Kr that the results of these five chip initiatives are intended not only for Dreame's internal use but may eventually be offered to external partners.
According to information shared at the launch event, one initiative involves a smartphone system-on-chip (SoC) called Chixiao 01, designed with an emphasis on improved graphics performance. The processor will support full-scene ray tracing at 120 frames per second at 1.5K resolution.
The Chixiao 01 will also feature a proprietary neural processing unit architecture capable of delivering <a href="/industry/ai">artificial intelligence</a> computing power of up to 200 TOPS (tera operations per second). By comparison, Qualcomm's latest flagship processor, the Snapdragon 8 Elite, offers around 80 TOPS of AI computing performance.
NXMind is also developing an integrated autonomous driving chip designed for Level 4 self-driving systems. The company aims to deliver computing power of 2,000 TOPS per chip, roughly three times the capability of most mainstream smart driving chips currently on the market. That target would position the product in direct competition with Nvidia's flagship autonomous driving platform, the Drive Thor series.
According to the company, the chip will integrate Dreame's technological capabilities in areas such as computer vision and LiDAR (light detection and ranging), and will be manufactured using a two-nanometer process node.
NXMind is simultaneously developing a processor for personal computers. The chip is expected to deliver computing power of 1.5 petaflops and will feature a unified memory architecture that supports multi-device interconnection and networking.
These efforts join the Tianqiong chip series, which is designed to support robotic vacuum cleaner control. According to Fu, Tianqiong chips will use a heterogeneous computing platform that combines multi-core CPUs, dedicated NPUs, and an independent microcontroller unit. The chip will be manufactured using a 22-nanometer process and is expected to debut in Dreame's robotic vacuum cleaners scheduled for release in the second half of this year, enabling the machines to navigate more complex environments with greater precision in obstacle detection and avoidance.
As previously reported, NXMind is also developing the Yaotai series of space computing modules designed for deployment in computing satellites. The company aims to achieve industry-leading reliability, lifespan, performance, and stability.
The long-term goal is to build a space-based supercomputing cluster consisting of up to two million computing satellites, roughly twice the size of Elon Musk's SpaceX Starlink constellation.
whereRelatedArticle
Despite the scale of its ambitions, however, sources told 36Kr that among the five chip categories announced, only the general <a href="/industry/robotics">robotics</a> chip has so far entered the tapeout and mass production stage.
Another person familiar with the matter told 36Kr that Fu, the head of NXMind, does not have a prior background in the semiconductor sector, and Dreame's chip team currently numbers only a few hundred people.
Chipmaking is a capital-intensive and highly competitive business. NXMind is targeting several areas simultaneously, including robotics SoCs, smartphone processors, autonomous driving chips, and space-based computing infrastructure, in an attempt to provide foundational computing power across industries such as robotics, smartphones, automotive, and aerospace.
At the same time, these efforts are intended to support Dreame's broader product ecosystem.
One advantage for Dreame lies in its downstream presence. Its established robotics hardware products and ecosystem provide an existing market for some of these chips, while its core business continues to generate relatively stable cash flow.
Even so, overtaking global incumbents across five vastly different semiconductor segments at the same time remains an uphill challenge for most companies to achieve alone.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3719316885304961">article</a> was written by Qiu Xiaofen for 36Kr.

Dreame unveils Chixiao 01 smartphone processor among five new chip initiatives

Dreame

Deep Tech

Semiconductors

StairMed, a brain-computer interface (BCI) developer, has secured RMB 500 million (USD 72.7 million) in a funding round led by <a href="/company/alibaba">Alibaba Group</a>, with participation from SDIC Unity Capital. Existing investors including <a href="/company/tencent">Tencent</a>, FountainBridge Capital, OrbiMed, Oriza Seed, Qiming Venture Partners, Lilly Asia Ventures, Source Code Capital, and SSCI Leading Fund also joined the round. Springhill Fund served as financial adviser.
With this deal, StairMed has raised a total of RMB 1.1 billion (USD 159.9 million) over the past year.
Last December, the company unveiled its second-generation high-throughput wireless invasive BCI system, WRS02, which increases electrode channels to 256. Beyond restoring motor control, the system is also designed to explore language reconstruction as a potential future application.
By early this year, using a self-developed surgical robot, StairMed completed clinical implantation of its 256-channel BCI system and validated brain-controlled interaction functions. The company plans to begin multicenter registration clinical trials later in 2026, aiming to enroll and implant approximately 40 patients within the year.
The year 2026 could mark an important stage for BCI development. While technical systems are approaching real-world deployment, policy developments are also providing momentum. Last year, China’s National Healthcare Security Administration released guidance aimed at establishing pricing frameworks for neurological medical services. The guidance includes BCI-related procedures as independently billable medical services, signaling institutional support for integrating BCI technologies into mainstream healthcare.
In a previous interview, Zhao Zhengtuo, co-founder of StairMed, said the clinical value of BCI-based therapies could be verified within the next three to five years, particularly in improving patients’ quality of life and ability to work.
“In five to ten years, [BCIs] will begin to show potential in consumer health,” Zhao said. “Because BCIs enable direct communication between humans and machines, they could redefine human-computer interaction and dramatically enhance efficiency.”
For BCI platforms, however, technological limits extend beyond hardware. Ultra flexible neural electrodes, integrated system design, and clinical validation are all important, but the depth of neuroscience research and understanding of the brain ultimately determines how far these systems can advance.
whereRelatedArticle
In that process, data may become a decisive factor.
“The company that accumulates the largest and richest brain datasets will be best positioned to define the next wave of applications and technologies,” Zhao said. Once large volumes of data are collected from a single brain region performing specific tasks, researchers could train what he described as a “foundation model” for brain data, similar to how large <a href="/industry/ai">artificial intelligence</a> models are trained today.
Such a model could then be applied to new patients, potentially improving the initial decoding performance and brain control capabilities of their BCI systems.
The participation of technology companies such as Alibaba could accelerate that effort. Capabilities in multimodal large models, computing infrastructure, intelligent hardware, and ecosystem development could eventually combine with StairMed’s hardware and clinical translation experience, potentially supporting the development of next-generation BCI systems and related applications.
From the beginning, StairMed envisioned BCI technology eventually reaching consumer-grade applications. The company chose to start in healthcare because it provides both a necessary regulatory pathway and an area with urgent real-world demand, offering a practical environment for refining the technology.
“Our hope is not just to restore motor or sensory function, but also to push the boundaries of human control,” said Li Xue, another co-founder of StairMed. “In the past, we controlled limbs with our brains. In the future, we might control external devices via BCIs, potentially surpassing normal human limits.”
Early results already suggest that possibility. Paralyzed patients implanted with BCI devices have reportedly shown advantages in specific tasks, such as cursor control speed, after training.
Looking ahead, the convergence of BCIs, AI, and advanced hardware could allow people to control intelligent agents and external devices with complex intent through thought alone.
In StairMed’s long-term vision, the brain could connect with multiple external systems through BCIs, bringing the concept of human–machine integration closer to practical use.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3720572708665989">article</a> was written by Hai Ruojing for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.88 = USD 1 based on estimates as of March 13, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Alibaba leads RMB 500 million investment in BCI startup StairMed

Biotechnology

Health Tech

… consumption patterns are undergoing structural change. Spending is increasingly driven by self-reward, quality upgrades, and emotional or lifestyle experiences rather than purely functional needs.
These emerging consumption patterns have fueled economic activity while creating new supply chains. Supporting such companies in the capital market could diversify the industrial structure of China’s stock market and allow investors to participate in the growth of new consumer trends.
<h3>A broader exit channel could reshape expectations</h3>
For investors, reforms could have far-reaching implications for both venture capital and public market valuations.
Historically, venture capital firms have been cautious about backing emerging consumer brands because even successful companies in these sectors often struggled to qualify for A-share listings due to unstable earnings or unconventional business models.
A more flexible ChiNext listing framework would provide clearer expectations for exit opportunities. This could encourage investors to back earlier-stage startups and smaller companies that may not yet be profitable but demonstrate strong growth potential.
Over time, such a shift could stimulate greater capital inflows into emerging consumer sectors and create a cycle linking industrial innovation with capital formation.
Another factor shaping the discussion is valuation.
Historically, consumer companies listed in the A-share market have traded at higher valuation multiples than their counterparts in Hong Kong.
At present, consumer stocks in the mainland market often trade at price-to-earnings ratios above 30 times, compared with roughly 18 times for consumer companies in Hong Kong.
For example, Tsingtao Brewery trades at roughly 18 times earnings in the mainland market, while China Resources Breweries trades at about 13 times in Hong Kong. Beingmate trades at around 46 times earnings in the mainland market, whereas Feihe, known internationally as Firmus, trades at under 12 times in Hong Kong.
Several mainland-listed liquor companies command even higher valuations. Yanghe trades at roughly 37 times earnings, Shuijingfang around 32 times, and Shede Spirits about 111 times.
Part of this divergence reflects differences in market structure. Hong Kong’s market includes a larger share of discretionary consumer companies, including retail, food service, and apparel businesses, which tend to experience larger valuation swings during economic cycles. By contrast, the mainland market has historically been dominated by staple consumer companies whose valuations are generally more stable.
These valuation gaps suggest that if emerging consumer companies gain a pathway back to the mainland market, their potential market capitalization could expand significantly. That prospect could enhance shareholder returns and attract long-term institutional capital.
For the A-share market itself, the impact could also be structural.
For years, China’s consumer sector in the mainland stock market has been dominated by traditional blue chip companies such as liquor makers and home appliance manufacturers, leaving the valuation framework relatively static.
The arrival of new-generation consumer companies could reshape that landscape. As more consumer brands listed in Hong Kong potentially return to the mainland market, the valuation structure of China’s consumer sector may gradually evolve.
Companies with durable competitive advantages and sustainable growth prospects would likely command more stable valuations, while speculative stocks driven primarily by hype could see their premiums fade. Over time, that shift could move China’s consumer equity market toward a more disciplined and sustainable pricing framework.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3718189771863427">article</a> was written by Wang Hanyu for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.88 = USD 1 based on estimates as of March 12, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

China signals ChiNext reforms to tilt listings toward consumer companies

NXMind, a company within <a href="/company/dreame">Dreame's</a> ecosystem, officially unveiled its Tianqiong chip series on March 11. It said the chips have entered mass production and will soon be deployed across Dreame's robotics product lineup.
Demand for computing power is expanding at a pace that exceeds the historical trajectory of Moore's law amid the <a href="/industry/ai">artificial intelligence</a> boom. According to a <a href="https://openai.com/index/ai-and-compute/">2018 analysis by OpenAI</a>, compute used in the largest AI training runs has doubled every 3.4 months since 2012, increasing by more than 300,000-fold over that period. As large models continue to scale in parameter size, and edge intelligence evolves from basic perception to more complex decision-making, computing power has emerged as a central constraint in the next phase of AI competition.
At the same time, global computing supply faces structural bottlenecks. Developers are approaching the physical limits associated with Moore's law, slowing improvements in transistor density. Capacity for advanced semiconductor manufacturing processes also remains constrained, creating uncertainty in chip supply. In addition, terrestrial data centers face limits related to energy consumption targets, cooling efficiency, and land availability, making it difficult to sustain the pace at which AI computing demand is expanding.
According to a <a href="https://www.iea.org/reports/electricity-2024">2024 report</a> by the International Energy Agency, global data center electricity consumption could exceed 1,000 terawatt-hours by 2026, roughly equivalent to Japan's annual electricity usage. Rising energy demand, cooling constraints, and site availability have become three major factors limiting the expansion of computing infrastructure.
As a result, competition in the semiconductor sector is undergoing a structural shift. Edge chips are evolving from single-purpose components into system-level intelligent hubs, with highly integrated system-on-chip designs becoming key enablers of embodied intelligence. At the same time, the physical deployment of computing infrastructure is expanding beyond Earth's surface toward architectures that integrate space, air, and ground networks. Computing nodes in low Earth orbit (LEO) are emerging as a new frontier, with global technology companies exploring deployments in this domain.
Wu Zhongze, former vice minister of China's Ministry of Science and Technology, said at this year's Appliance &amp; Electronics World Expo that chipmakers are quietly reshaping how the world operates. In recent years, China has promoted deeper integration across innovation chains, industrial supply chains, capital networks, and talent pipelines within the chip and computing sectors to support long-term industry development.
From a broader perspective, the rise of edge intelligence is redefining chip design paradigms. As embodied intelligence and humanoid robots move from laboratory research toward industrial deployment, chips are no longer just computing components. Instead, they serve as the backbone connecting perception, understanding, decision-making, and execution.
Robots operating in complex and dynamic environments must respond in real time, placing strict requirements on latency, power consumption, and reliability. Traditional general-purpose chip architectures often struggle to balance performance and efficiency under these conditions. This has accelerated the development of specialized system-on-chip architectures, where heterogeneous computing units integrate perception fusion, decision planning, and motion control within a single chip. The result is an edge intelligence system capable of operating locally on the device.
NXMind's Tianqiong chip series reflects this shift. The chip features a heterogeneous computing platform composed of a multi-core CPU, a dedicated neural processing unit, and an independent microcontroller unit.
NXMind said the chip will be deployed by Dreame to support functions such as LiDAR (light detection and ranging), AI-driven vision fusion, and binocular obstacle avoidance algorithms. These capabilities are intended to improve navigation and obstacle avoidance in complex home environments.
The company's differentiation strategy builds on Dreame's existing technology base. Rather than developing a chip platform independently from the ground up, NXMind draws on Dreame's accumulated algorithm expertise, supply chain ecosystem, and operational data generated from product shipments. According to the company, this enables a collaborative design approach that links chip architecture with algorithms.
Scenario-defined chips have increasingly become a strategy for semiconductor companies seeking to establish competitive barriers.
Meanwhile, the structure of computing infrastructure is evolving. As ground-based data centers encounter physical and environmental constraints, space-based computing deployments are being explored as an alternative approach.
LEO provides a vacuum environment that naturally supports heat dissipation, along with continuous access to solar energy. In theory, this could enable higher-density computing deployments with lower cooling costs. Space-based computing nodes could also provide onboard processing for global LEO satellite internet systems, space-based remote sensing data, and deep space exploration missions, reducing delays associated with transmitting data back to ground stations.
According to 36Kr, NXMind's first Yaotai space computing module is scheduled to launch into orbit in March. The launch is intended to begin construction of a supercomputing center in LEO and initiate preliminary verification of an in-orbit computing network. The move also suggests that NXMind is transitioning from an R&amp;D phase toward early commercialization.
NXMind's current business portfolio spans smartphone processors, autonomous driving chips, robotics chips, space-based computing technology, and personal AI computers. Together, these initiatives form a broad product ecosystem.
This trajectory reflects a wider shift across the AI sector. Computing power is expanding from centralized cloud infrastructure toward distributed edge environments, while also extending into integrated architectures that connect space, air, and ground systems.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3718486572037507">article</a> was written by Huang Nan for 36Kr.

Dreame-backed NXMind enters commercialization with Tianqiong chips, targets orbital computing

<a href="/company/tencent">Tencent</a>, long known for its measured and cautious style, has begun making a series of assertive moves in <a href="/industry/ai">artificial intelligence</a>.
On March 6, it launched a promotional campaign offering free OpenClaw installation sessions assisted by its employees, drawing thousands of people to queue outside its office building. On March 9, it unveiled WorkBuddy, a general AI agent. A day later, on March 10, it began internal testing of QClaw, an AI assistant. These moves appear to be only the beginning.
36Kr has learned that Tencent's WeChat is experimenting with developing its own independent AI model. The model has already completed foundational capability development and received an internal codename, with plans for a public rollout later this year.
In the near term, the model will be integrated into WeChat's mini program ecosystem, allowing developers to build various AI agents. Over the longer term, WeChat aims to explore applications that embed AI more deeply into its social ecosystem. For example, by drawing on users' long-term behavioral data within WeChat, the company is exploring how AI could improve efficiency and enhance the user experience inside the app.
When DeepSeek emerged a year ago, Tencent was among the fastest companies to respond. It integrated Yuanbao with <a href="/company/deepseek">DeepSeek</a>, quickly driving a surge in daily active users. For some time afterward, however, Tencent remained relatively quiet in the AI agent race. It was not until Doubao's daily active users surpassed 100 million later in the year that Yuanbao began responding more aggressively, backed by Tencent's financial resources.
A report from QuestMobile shows that after the Lunar New Year campaign ended, Yuanbao's daily active users largely fell back to pre-event levels, failing to sustain the momentum created during the promotion period.
Sources familiar with the matter described several themes discussed during a WeChat executive meeting in December 2025, when the company reviewed its AI strategy.
First, WeChat must possess built-in AI tools that do not rely on third-party systems.
Second, WeChat essentially consists of three components: person-to-person social interaction, information acquisition, and efficiency tools. Social relationships themselves cannot be replaced by AI, meaning AI's role can only emerge in the information and productivity layers.
Earlier, during Tencent's third-quarter 2025 earnings call, president Martin Lau outlined a broader vision. In his view, WeChat's ultimate blueprint is to launch an AI agent of its own.
“WeChat's ecosystem includes a communications and social network that allows agents to understand users' needs, intentions, and interests,” Lau said. “It also has a content ecosystem, […] a mini program ecosystem that covers most use cases on the internet, a commercial ecosystem that allows people to purchase goods, and a payment ecosystem that enables transactions almost instantly. Taken together, this could become the user's ideal assistant, one that understands needs and can complete tasks within the ecosystem.”
WeChat already holds a unique advantage over Qwen and Doubao. It is the only app with more than 1.4 billion monthly active users, offering a natural pathway to adoption and usage. More importantly, that position is supported by two scarce assets: vast social networks and a mini program ecosystem that spans e-commerce, transportation, local services, and government services.
If WeChat's AI agent can connect both the social graph and the mini program service system, it could orchestrate these capabilities across its existing user base. For rivals still spending heavily to acquire users, this could pose a significant competitive challenge.
For the same reason, however, building an AI agent within WeChat may prove more complex.
The key capability of an AI agent is understanding user behavior and context. Yet the foundation of any social platform is trust. Users are willing to chat, discuss work, and exchange private information on WeChat because they assume their data is secure. Once AI becomes involved, users may grow uneasy about privacy.
According to an unnamed source cited by 36Kr, one of WeChat's central considerations during development has been ensuring that users can approach AI as a tool without feeling overly monitored, since access to personal or corporate data, often sensitive, may be required to power these capabilities. As a result, each step of development is being approached cautiously.
Since the AI wave began, Tencent's deepest internal concern has revolved around a single question: could a new breakout product emerge that might disrupt WeChat, much like how WeChat once replaced phone contact books and legacy messaging platforms?
For now, however, that moment of disruption has yet to arrive. Regardless of how technology evolves, fundamental human needs, including connection and belonging, are unlikely to disappear. In that sense, AI still cannot replace real human relationships. On that front, WeChat continues to hold an advantage that is difficult to displace.
WeChat's decision to invest in its own model therefore serves both as defense and reinforcement. It also reflects Tencent's longstanding product philosophy: the first product released is not necessarily the final centerpiece.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://www.36kr.com/p/3717835616335497">article</a> was written by Xiao Sijia for 36Kr.

WeChat explores building in-house AI model as Tencent ramps up AI push

WeChat

Artificial Intelligence

Chinese automaker <a href="/company/byd">BYD</a> is transforming the <a href="/industry/ev">electric vehicle</a> market, surpassing <a href="/company/tesla">Tesla's</a> sales in more than 20 countries and regions over the past five years as it embraces risk and expands globally to overcome waning sales at home.
A dramatic reversal occurred last year in the UK as BYD's annual sales surpassed Tesla's for the first time. That spring, at a dealership in west London, BYD founder and chairman Wang Chuanfu posed for a photo with a customer who bought an Atto 3 EV.
The Atto 3, which BYD chose as its first passenger export vehicle in 2021, gets its name from the word "attosecond," a unit of time equivalent to one quintillionth of a second. The company chose the name to convey its rapid pursuit of technology and performance.
In 2021, Wang often spoke about speed in company meetings.
"Toyota Motor and Volkswagen are a little slow in electrifying their vehicles, but once they make a leap, the impact will be huge," he said at the time. "BYD absolutely has to be faster." Outpacing other companies was necessity for survival.
The plan worked. Global sales expanded rapidly, and in the four years from 2021, BYD entered 113 countries and regions.
Nikkei compared BYD and Tesla's sales figures for 2020 and 2025 using data from S&amp;P Global Mobility. BYD overtook Tesla in 22 countries and regions. In addition to European nations like the UK, Spain, and Italy, the automaker also passed Tesla in Hong Kong and Singapore, where luxury vehicles are said to have a large market share. Last year, BYD toppled Tesla as the world's top EV seller.
The average price of BYD's passenger vehicles in China last year was RMB 114,000 (USD 16,478.8) according to a Chinese research company. The automaker has kept costs down by producing batteries and core components in-house. Its highly price competitive EVs have flowed into the global market in a steady drip.
One such destination is Uruguay, with a population of about 3.4 million and a new-vehicle market of just 70,000 units.
BYD laid the groundwork for its full entry into Uruguay by conducting road tests of electric buses in the early 2010s. It has since delivered large electric buses and electric taxis and built relationships with the local government and dealers.
Elsewhere in South America, at Peru's port of Chancay—about 80 kilometers north of the capital, Lima—Chinese-made vehicles are neatly lined up on the quay.
The port, built with Chinese capital as a component of Beijing's Belt and Road infrastructure initiative, began full-scale operations in June 2025. Much of the cargo bound for South America from Asia previously passed through ports in Central or North America.
The China-backed port opened a direct route. Transport times have been slashed to about 25 days, down by around 10 days.
During its expansion, BYD has encountered friction in North America and Europe. Plans for a passenger vehicle factory in Mexico, which BYD hoped to use as an avenue into North America past high tariff barriers, <a href="https://asia.nikkei.com/business/automobiles/electric-vehicles/mexico-blocks-byd-ev-factory-putting-us-tariff-talks-first" target="_blank" rel="noopener">were scrapped</a> last year as US President Donald Trump's administration put pressure on the Mexican government.
In his election campaign, Trump had blasted the construction of factories in Mexico by Chinese companies looking to sell cars in the US Mexico could have been hurt during tariff negotiations with Washington if it gave the impression of welcoming Chinese businesses.
Meanwhile, the European Union has set its own standards in a bid to keep Chinese players out, giving preference to <a href="https://asia.nikkei.com/business/automobiles/eu-to-launch-compact-ev-class-to-cut-costs-and-counter-china" target="_blank" rel="noopener">compact electric vehicles</a>.
Even when faced with such walls, BYD has little choice but to keep expanding overseas, as the rapid growth in domestic sales that fueled its investment has come to a halt.
The automaker sold about 3.5 million passenger vehicles in China last year, down roughly 10% from 2024. Sales for the July-September quarter sank 3% on the year to RMB 194.9 billion (USD 28.2 billion), the first decline in 22 quarters.
Free cash flow sank to minus RMB 44 billion (USD 6.4 billion), the largest negative figure of the past 15 or so years. Lower free cash flow risks leaving the company with less capital to invest overseas.
To avoid trade friction, BYD is shifting from exports to local production. It is slated to open a passenger vehicle factory in Hungary as early as this year, following new plants in Thailand in 2024 and Brazil in 2025.
Other Chinese players also have picked up overseas production facilities from other automakers. <a href="/company/great-wall-motor">Great Wall Motor</a> acquired the site of a former Mercedes-Benz plant in Brazil, and <a href="/company/chery">Chery Automobile</a> is buying <a href="https://asia.nikkei.com/business/automobiles/nissan-to-sell-plants-in-south-africa-to-china-s-chery-automobile" target="_blank" rel="noopener">plants in South Africa</a> from Japan's Nissan Motor.
China in 2023 surpassed Japan to become the world's top auto exporter. Japanese automakers built more than 16 million vehicles overseas in 2024, but their Chinese competitors are catching up fast.
This article first appeared on <a href="https://asia.nikkei.com/business/automobiles/electric-vehicles/byd-redraws-global-ev-map-overtaking-tesla-in-20-markets">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.0722vps.aieehr01.top/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.
Note: RMB figures are converted to USD at rates of RMB 6.92 = USD 1 based on estimates as of March 10, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

BYD redraws global EV map, overtaking Tesla in 20 markets

Tesla

Joyy reported its fourth-quarter 2025 results on March 10, with figures suggesting its long-awaited return to growth may be approaching, albeit on different terms than before. The company reported USD 581.9 million in revenue for the three months ended December 31, up 5.9% year-on-year (YoY) and 7.7% from the prior quarter. It was Joyy’s first quarterly revenue increase on a YoY basis since the second half of 2024. The rebound, however, was driven less by its legacy live streaming business than by a rapidly expanding advertising operation.
That shift matters because it changes both Joyy’s growth profile and its cost structure. Advertising revenue jumped 62.4% YoY to USD 145.4 million in the quarter. Live streaming revenue still declined from a year earlier, falling to USD 394.4 million from USD 422.4 million, though it rose 1.5% sequentially. Non-live streaming revenue reached USD 187.5 million, or 32.2% of group revenue, up from 23.1% a year earlier. The figures show that Joyy’s diversification is becoming more visible in the numbers.
The center of that shift is Bigo Ads. Joyy said the unit generated USD 128.1 million in fourth-quarter revenue, up 61.5% YoY and 23.3% from the previous quarter. Its Audience Network business expanded even faster. Management linked the growth to broader traffic coverage, deeper penetration across advertiser verticals, and algorithm improvements that increased delivery efficiency and advertiser spending. SDK ad requests rose 166% YoY, while demand from North America and Western Europe strengthened sequentially.
The tradeoff is already visible in profitability. Gross profit was broadly flat YoY at USD 205.6 million, pushing gross margin down to 35.3% from 37.1% a year earlier and 35.8% in the third quarter. Joyy attributed the pressure mainly to higher revenue-sharing fees and content costs tied to third-party traffic acquisition for its advertising business.
The pressure extended below the gross line. Non-GAAP EBITDA was unchanged from the previous quarter at USD 50.6 million but declined from USD 55.7 million a year earlier. Non-GAAP EBITDA margin narrowed to 8.7% from 10.1%, while non-GAAP operating income margin fell to 7% from 8.4%. Sales and marketing expenses rose to USD 81.4 million as Joyy resumed user acquisition spending after unusually low spending in the same quarter a year earlier, when a temporary app store interruption created one-off savings.
That leaves the quarter mixed beneath the headline beat. Operating income improved to USD 18.3 million from a USD 427.9 million loss a year earlier, while net income from continuing operations attributable to Joyy reached USD 54.3 million compared with a USD 304.1 million loss a year earlier. The comparison, however, is partly skewed by the absence of a USD 454.9 million goodwill impairment charge recorded in the fourth quarter of 2024. On a more comparable basis, profitability was steadier rather than sharply stronger. Non-GAAP net income from continuing operations slipped to USD 70.3 million from USD 72.4 million in the third quarter and USD 96.1 million a year earlier.
Live streaming, meanwhile, is recovering only gradually. Joyy highlighted a third consecutive quarter of sequential growth in the segment, supported by localized operations, a broader content mix, and <a href="/industry/ai">artificial intelligence</a>-driven improvements to recommendations and payment systems. Bigo paying users rose 1.5% sequentially to 1.54 million, while ARPPU reached USD 222.8. Even so, live streaming revenue remains below last year’s level, suggesting the core business has stabilized more than it has reaccelerated.
The full-year figures show the same pattern. Joyy’s 2025 revenue declined to USD 2.12 billion from USD 2.24 billion in 2024 as live streaming revenue fell to USD 1.53 billion from USD 1.79 billion. Advertising partly offset that drop, rising 37.1% to USD 442.7 million, while Bigo Ads revenue increased 38.5% to USD 398.5 million. Non-GAAP EBITDA rose 10.9% to USD 189.8 million, indicating that the company maintained earnings discipline even as its revenue mix shifted.
The balance sheet gives Joyy room to continue that transition. The company ended 2025 with USD 3.26 billion in net cash and generated USD 116 million in operating cash flow in the fourth quarter. It also returned USD 332 million to shareholders during 2025 through buybacks and dividends, and has since declared a cash dividend of USD 1.38 per ADS (American depository share) for the first quarter of 2026.
The next question is whether advertising can keep scaling without further eroding group margins. Joyy guided first-quarter 2026 revenue to USD 538–548 million, implying a sequential decline from the fourth quarter. The company also plans to refine its segment reporting beginning in the first quarter of 2026, which may provide a clearer view of how much growth is coming from advertising, commerce software, and its still-dominant social entertainment business.

Joyy revenue rises in Q4, but margins hit by advertising growth costs

Short Video

Live Streaming

Video

Creative

Content

Creator Economy

… returns.
For now, the market is rewarding the milestone. But investors will likely watch whether Nio can keep vehicle margins near current levels, absorb raw material inflation without giving back too much profitability, and grow deliveries without rebuilding its cost base too quickly. The fourth quarter of 2025 showed that Nio’s model can work under favorable conditions. It did not yet prove those conditions will last.
Note: RMB figures are converted to USD at rates of RMB 6.88 = USD 1 based on estimates as of March 11, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Nio hits first operating profit in Q4 as turnaround begins to show