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

… established partnerships with companies across subsectors including smart homes, robots, wearables, and consumer electronics. Its coverage extends to more than 50 device categories, including white goods, brown goods, kitchen and bathroom appliances, home control panels, home robots, embodied robots, smartphones, and PCs.
AISpeech has worked with major brands across those categories. In smart home appliances, it has served Midea, Hisense, Haier, and Changhong. In home robots, it has worked with Ecovacs, <a href="/company/dreame">Dreame</a>, and Roborock. In embodied robots, it has worked with Agibot, Galbot, and Magic Atom. In consumer electronics, it has maintained long-term partnerships with <a href="/company/lenovo">Lenovo</a>, Transsion, and Oppo.
AISpeech’s financial fundamentals have improved compared with the period covered in its previous filing. From 2019 to 2022, revenue rose from RMB 114.7 million (USD 16.9 million) to RMB 423.2 million (USD 62.3 million), with revenue of RMB 236.7 million (USD 34.8 million) in 2020 and RMB 307.4 million (USD 45.2 million) in 2021. Over the same period, gross margin fell from 72.17% to 57.91%.
During the 2023–2025 reporting period for the current filing, AISpeech’s revenue rose from RMB 539.1 million (USD 79.4 million) to RMB 600.8 million (USD 88.4 million), then to RMB 687.7 million (USD 101.2 million). Its gross margin improved from 53.69% in 2023 to 57.81% in 2024 and 63.24% in 2025.
AISpeech’s net loss attributable to shareholders of the parent company was RMB 111.8 million (USD 16.5 million) in 2023, widened to RMB 140.4 million (USD 20.7 million) in 2024, then narrowed to RMB 57.0 million (USD 8.4 million) in 2025. Excluding nonrecurring items and the impact of share-based payments, its attributable net loss narrowed from RMB 218.8 million (USD 32.2 million) in 2023 to RMB 144.3 million (USD 21.2 million) in 2024 and RMB 48.3 million (USD 7.1 million) in 2025. The figures show a narrower loss profile in 2025, while the company said its ability to continue as a going concern and its path toward profitability have improved.
The business mix also shifted in 2025. Smart mobility solutions contributed RMB 275.7 million (USD 40.6 million) in revenue, or 40.08% of the total. The smart office segment contributed RMB 243.5 million (USD 35.8 million), or about 35.4%, while the smart IoT segment contributed RMB 168.6 million (USD 24.8 million), or 24.51%.
AISpeech plans to raise RMB 1.555 billion through the IPO. The proceeds are earmarked for three projects: RMB 680 million (USD 100.1 million) for AI software and hardware-software integrated solutions, about RMB 325 million (USD 47.8 million) for AI smart device product R&amp;D and upgrades, and approximately RMB 550 million (USD 81 million) for R&amp;D center construction.
This <a href="https://xueqiu.com/2854806290/390969599">article</a> was adapted based on a feature originally written by Stone Jin and published on IPO Zaozhidao. KrASIA is authorized to translate, adapt, and publish its contents.
Note: RMB figures are converted to USD at rates of RMB 6.79 = USD 1 based on estimates as of May 28, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

AISpeech targets Star Market IPO as demand for conversational AI grows

Artificial Intelligence

Deep Tech

China

… strategy.
Within this system, Geespace’s role is not only that of a constellation operator, but also that of a global satellite communications service provider. Its real value lies in becoming part of the foundation for global critical infrastructure.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3807083606892294">article</a> was written by Xiao Xi for 36Kr.
Note: RMB figures are converted to USD at rates of RMB 6.79 = USD 1 based on estimates as of May 28, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Geespace’s next test is building a business case for satellite infrastructure

Space Tech

… better, those reactions must be observed rather than assumed.
And as deployment scales, another challenge emerges: not every new mile is equally informative. At a certain point, the problem is no longer just collecting more real-world data. It is identifying which interactions most clearly reveal where the model’s assumptions are weakest.
At the frontier of performance, the bottleneck is not simply data. It is knowing which data matter next. More precisely, it is knowing which assumptions are failing, and what evidence from reality would correct them.
A good world model should therefore do more than absorb feedback. It should help expose its own blind spots. It should make it possible to ask not only “What went wrong?” but also “What kind of real-world evidence would most improve this part of the system?”
That is where the next generation of world models should advance beyond simulation alone. A mature world model should not just passively improve after deployment, but also help direct the next round of learning. It should surface the interactions it models poorly, indicate where its own fidelity breaks down, and help determine what real-world evidence would most improve the training environment itself.
In that sense, the most important advance is not just more simulation. It is a development loop in which reality continuously corrects the model, and the model helps identify what aspects of reality the system most needs to learn from next. At that point, a world model may become more than generative, and more than merely trainable. It becomes increasingly self-correcting.
That is when a world model becomes more than generative. It becomes testable.
Without that property, more simulation may only amplify hidden errors. With it, simulation becomes something more useful: a controlled environment for hypothesis testing, counterfactual evaluation, directed learning, and targeted data collection.
Autonomous driving will not be decided by whichever model class sounds most fashionable. The real requirement is harder. Whatever the onboard architecture turns out to be, whether modular, end-to-end, interpretable end-to-end, or some hybrid form, the overall system must learn from interaction faster than it accumulates new complexity in the real world.
That depends, increasingly, on the quality of the world model behind it.
whereRelatedArticle
The success of large language models has made it tempting to believe that every AI-related domain will yield to the same recipe of more data, more compute, and larger models. But driving is not language. Language describes the world, while driving enters it. A system can sound coherent and still misunderstand distance, momentum, intent, or risk.
For self-driving cars, the decisive question is no longer how large the model is. It is whether the world model used is sufficient to train it not only to represent the world well enough to learn from it, but also to identify where that representation is failing and improve in the right direction. It is the accuracy, interactivity, diagnosability, and testability of the world model that trains it.
This article was written by Lou Tiancheng, founder and CTO of <a href="/company/pony-ai">Pony.ai</a>, and edited for brevity and clarity.

Why world models must do more than simulate: Pony.ai CTO

East Asia

Pony.ai

Transportation

Autonomous Vehicles

Mobility

Ecosystem

… Swastatika Sentosa, both MSCI-listed stocks on the IDX, have been included on the high shareholding concentration list.
Hasan Fawzi, OJK's chief capital market supervisor, said several Indonesian stocks could face adjustments in the MSCI index. "Several stocks have been indicated and will undergo adjustments, either in their weighting or possibly be removed from the MSCI. This has been preceded by a response from investors," Fawzi said.
Fawzi welcomed MSCI's move, describing it as an early warning that has been well received by investors.
This article first appeared on <a href="https://asia.nikkei.com/business/markets/trading-asia/index-providers-reshape-asian-financial-markets-with-exacting-standards">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>.
Note: KRW figures are converted to USD at rates of KRW 1504.94 = USD 1 based on estimates as of May 22, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Index providers reshape Asian financial markets with exacting standards

Investment

Southeast Asia

… href="https://fortune.com/2025/11/02/amazon-rufus-ai-shopping-assistant-chatbot-10-billion-sales-monetization/">Fortune</a>, Rufus has accumulated more than 250 million users, and shoppers who use AI are 60% more likely to complete a purchase than ordinary users. But some in the industry have pointed to Rufus’s instability, weaker performance when handling questions outside Amazon’s internal database, and relatively low conversion rate. Part of this has been attributed to Amazon’s proprietary large model, which may not yet match the sector’s leading models.</li>
<li aria-level="1">The third path is deep integration between large models and e-commerce ecosystems. <a href="/company/alibaba">Alibaba</a> is the representative case. The full integration of Qwen and Taobao is the first deep combination of a leading e-commerce platform and an AI application.</li>
</ol>
Ultimately, which path a company chooses depends on its resource base.
Model companies lack transaction scenarios and fulfillment capability, so they can only seek external partners. E-commerce giants have complete supply chains and data, but their large model capabilities may be limited. Players that have both large models and real-world commerce have a better chance of building a complete AI shopping experience.
whereRelatedArticle
Compared with the US market, China’s longstanding advantage in AI lies in its richer and more mature application scenarios. A huge user base, diverse formats, and rapid market acceptance create complex and varied implementation settings.
China’s major technology companies also have relatively comprehensive application layouts in consumer scenarios. In AI shopping, Alibaba has one of the most complete lifestyle service ecosystems, while Qwen has entered the ranks of internationally competitive large models. With this advantage, Alibaba has chosen to integrate the full AI shopping chain within its own ecosystem, building a barrier that may be difficult to replicate.
Alibaba has more than 20 years of e-commerce capabilities accumulated through Taobao, including search, price comparison, order placement, logistics tracking, and after-sales returns and exchanges. This integration allows those capabilities to be repackaged as “skills” that AI can call.
Inside the Qwen app, users can now complete every step from product recommendation to order placement, fulfillment, and after-sales service, rather than interacting only through shallow redirected links. This is one of the most comprehensive AI shopping service forms at present.
For the model, knowing when to search, when to compare prices, when to place an order directly, and when to advise users to think again requires a large amount of training from real shopping scenarios.
That points to a more important data asset. Taobao’s catalog of about four billion products and more than 20 years of accumulated real shopping scenario data can help Qwen understand purchase intent in user conversations and make recommendations.
Chinese venture capitalist Zhu Xiaohu once said that once the basic capabilities of large models become a relatively stable platform, the essence of competition will shift to engineering implementation capability and building closed data loops. This is precisely where Chinese companies tend to excel.
This reflects a deeper shift in large model competition. As the focus of AI development turns toward value at the application layer, competition is no longer only about model intelligence. It is also about the ability to turn technology into real products, meet specific user needs, and commercialize. As data becomes more important, high-quality data will tend to become closed, turning into each company’s moat.
The significance of this may extend beyond shopping itself. Since January this year, Qwen has been connected successively to service capabilities within Alibaba’s ecosystem, including Taobao Shangou (also known as Taobao Instant Commerce), Fliggy, Amap, and Alipay. The full integration with Taobao further fills a key gap in consumer scenarios.
The same idea is appearing among competitors. <a href="/company/bytedance">ByteDance’s</a> Doubao is accelerating integration with Douyin E-commerce. JD.com and <a href="/company/meituan">Meituan</a> have each launched independent AI shopping apps, trying to build barriers through vertical scenarios.
Whether users move from testing these tools to using them regularly will take time to answer. But one thing is becoming clearer: if people gradually get used to shopping through conversation, e-commerce, and perhaps the broader consumer market, may no longer look the same.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3808506543382020">article</a> was written by Xiao Xi for 36Kr.

Why AI shopping can finally move from recommendations to checkout

Taobao

Tmall

Alibaba Cloud

E-commerce

Consumer Economy

… AI inference servers including the ThinkSystem SE455i, SR650i, and SR675i for enterprise inference workloads.
Liquid cooling is another part of Lenovo’s AI infrastructure offering. Revenue from Lenovo Neptune liquid cooling rose 300% from a year earlier in the previous quarter. Lenovo’s latest results show that its annual server manufacturing capacity exceeds 70,000 racks across AI, compute, and storage systems, including more than 11,000 direct liquid-cooled racks designed for AI workloads.
For AI data centers, liquid cooling affects energy consumption, deployment density, and operating costs. Lenovo said Neptune liquid cooling can help data centers bring power usage effectiveness below 1.1 in high-intensity AI computing environments.
In storage, Lenovo’s acquisition of Infinidat, completed in early April, strengthened ISG’s high-end enterprise storage capabilities.
The software layer is also becoming more important. Lenovo Group executive vice president and China president Liu Jun said ISG’s growth has been driven not only by hardware, but also by the Wanquan platform, which can centrally manage and schedule diverse computing resources. Under the same hardware conditions, it can help customers improve performance by more than 20%, according to Liu.
In compute-as-a-service, Lenovo TruScale has become a growth driver for the Solutions and Services Group, or SSG. It shifts traditional hardware procurement into a subscription model based on usage, addressing cost pressure, idle resources, and technical complexity.
In endpoint AI applications, the Qira personal AI agent supports cross-device collaboration and task transfer. It is part of Lenovo’s effort to move beyond hardware into personal AI interfaces, while supporting its broader hybrid AI strategy across devices, edge environments, cloud systems, and private enterprise infrastructure.
Ken Wong, Lenovo Group executive vice president and president of SSG, said during an earnings call that the primary purpose of the AI factory is to improve AI production efficiency. It can significantly reduce the cost of each token while increasing the business value generated by each token.
“This means Lenovo is upgrading into an end-to-end AI service provider,” Wong said. “It can help customers truly build AI compute systems, bring AI into real business scenarios, continuously optimize operations, and ultimately turn AI into business value.”
<h3>ISG’s next test is profitability</h3>
Lenovo’s AI factory model is contributing to ISG’s revenue and profit recovery. It also shows how Lenovo is trying to shift more of its profit base from one-off hardware sales toward higher-value infrastructure and services.
A Huatai Securities research report cited by IPO Zaozhidao said one reason for ISG’s profit recovery is its expansion from cloud service providers to higher-margin enterprise and small and midsize business customers. Compared with cloud service providers, these customers tend to have stronger demand for bundled solutions and may offer more opportunities to integrate Lenovo’s SSG services.
At Lenovo Tech World 2026, the company highlighted its global manufacturing footprint and supplier relationships as competitive advantages.
That matters in an environment where memory, CPUs, and other key components remain in tight supply. A resilient supply chain can help Lenovo improve order predictability and manage logistics costs by shifting production across its global factory network. That capability may be harder for pure ODMs (original design manufacturers) to replicate. For cloud vendors and enterprise customers, predictable delivery can be as important as price.
Lenovo could also gain market share and improve margins during the current storage and CPU supply upcycle if it secured inventory earlier than competitors.
The link between ISG and SSG is central to Lenovo’s strategy. In fiscal 2025/26, SSG revenue exceeded USD 10 billion for the first time. Its operating profit margin reached 22.4%, its revenue growth was about four times the overall IT services industry growth rate, and it recorded year-on-year growth for 20 consecutive quarters.
For Lenovo, AI infrastructure can become more than a hardware sale. ISG can bring in customers through AI factory deployments, while SSG can add managed services, software, and ongoing support. That structure gives Lenovo a clearer path to recurring revenue if customers continue to expand their AI systems.
This <a href="https://xueqiu.com/2854806290/390480323">article</a> was adapted based on a feature originally written by MD and published on IPO Zaozhidao. KrASIA is authorized to translate, adapt, and publish its contents.
Note: HKD, RMB figures are converted to USD at rates of HKD 7.83 = USD 1 and RMB 6.80 = USD 1 based on estimates as of May 26, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Lenovo’s “AI factory” approach starts to show results

Lenovo

Enterprise Services

Hardware

AI Infrastructure

<a href="/company/huawei">Huawei</a> has proposed a new semiconductor development framework called the “Tau scaling law,” framing it as an alternative method to improve chip performance as Moore’s law becomes harder to sustain.
The proposal was introduced on May 25 at the IEEE International Symposium on Circuits and Systems in Shanghai by He Tingbo, chairwoman of Huawei’s scientist committee and president of its semiconductor business department. In her keynote, she described Tau scaling as a shift from geometric scaling, the traditional method of shrinking transistors, to time-based scaling, where chip performance is improved by reducing signal delay across devices, circuits, chips, and systems.
“Losing geometric scaling does not mean losing time scaling,” He said.
She added that one core technology under this approach is LogicFolding, an architecture designed to shorten critical wiring paths, reduce signal propagation delay, and improve transistor density and circuit performance. The company said Kirin chips scheduled for launch in fall and winter 2026 will be the first to adopt LogicFolding.
Dubbed “Her’s law” by He’s peers, Tau scaling is less about shrinking the chip, as Moore’s law is, and more about shortening the journey inside it. By shifting the emphasis to architecture and signal efficiency, the approach could reduce dependence on the most advanced lithography tools, though not eliminate it.
<figure id="attachment_162206" aria-describedby="caption-attachment-162206" style="width: 1920px" class="wp-caption alignnone"><img loading="lazy" decoding="async" class="size-full wp-image-162206" src="https://console.kr-asia.com/wp-content/uploads/2026/05/Huawei-He-Tingbo-IEEE-ISCAS-2026.jpg" alt="Photo shows He Tingbo, chairwoman of Huawei’s scientist committee and president of its semiconductor business department." width="1920" height="1080" srcset="https://console.kr-asia.com/wp-content/uploads/2026/05/Huawei-He-Tingbo-IEEE-ISCAS-2026.jpg 1920w, https://console.kr-asia.com/wp-content/uploads/2026/05/Huawei-He-Tingbo-IEEE-ISCAS-2026-720x405.jpg 720w, https://console.kr-asia.com/wp-content/uploads/2026/05/Huawei-He-Tingbo-IEEE-ISCAS-2026-1024x576.jpg 1024w, https://console.kr-asia.com/wp-content/uploads/2026/05/Huawei-He-Tingbo-IEEE-ISCAS-2026-768x432.jpg 768w, https://console.kr-asia.com/wp-content/uploads/2026/05/Huawei-He-Tingbo-IEEE-ISCAS-2026-1536x864.jpg 1536w, https://console.kr-asia.com/wp-content/uploads/2026/05/Huawei-He-Tingbo-IEEE-ISCAS-2026-480x270.jpg 480w" sizes="(max-width: 1920px) 100vw, 1920px" /><figcaption id="caption-attachment-162206" class="wp-caption-text">He Tingbo, chairwoman of Huawei’s scientist committee and president of its semiconductor business department. Photo courtesy of Huawei.</figcaption></figure>
Traditional semiconductor progress has long been measured by how small transistors can be made and how many can be packed onto a chip. Huawei’s proposition focuses instead on how efficiently signals and data move. In theory, shorter signal paths can reduce latency and improve performance, even if the underlying manufacturing process does not match the most advanced global nodes.
According to He, high-end chips based on Tau scaling could reach transistor density equivalent to 14 angstroms, or 1.4-nm, processes by 2031.
This does not necessarily mean Huawei expects to manufacture true 1.4-nm chips by then. A more precise reading is that it aims to achieve density comparable to future 1.4-nm-class processes through architecture, circuit design, and system-level optimization.
With LogicFolding, Huawei intends to switch its Kirin chips from a single-layer to a double-layer structure. By stacking and rearranging parts of the chip, Huawei believes the vertical design can increase density and enhance performance, even if the underlying manufacturing process has not advanced by the same amount under geometric scaling.
“Before LogicFolding, it took three years to lift transistor density from 126 to 155 million transistors per square millimeter,” He said. “In 2026, LogicFolding takes it all the way to 238 million transistors per square millimeter, in one single step.”
The global benchmark is already moving in that direction. TSMC unveiled its A14 process, a 1.4-nm-class technology, in 2025 and said it is planned to enter production in 2028. Other leading chipmakers, including Intel and Samsung, are also pursuing next-generation nodes around the 2-nm and angstrom-class range.
Huawei’s target is therefore ambitious, though it would still need external validation through commercial chips, measurable density gains, production yields, and real-world performance.
whereRelatedArticle
The announcement comes as China continues to face constraints in advanced chipmaking due to US and allied export controls that have limited the country’s near-term ability to produce chips beyond the 7-nm node. Huawei and others in China’s broader semiconductor sector have been looking for ways to sustain chip advances amid these restrictions, with localization of semiconductor design and manufacturing among Beijing’s priorities.
The proposal is therefore significant because it builds on Huawei’s broader role in China’s <a href="/industry/ai">artificial intelligence</a> and semiconductor self-reliance push. Huawei has often been cited as <a href="https://merics.org/en/report/chinas-drive-toward-self-reliance-artificial-intelligence-chips-large-language-models">a central player in China’s AI chip self-reliance drive</a>, including through its work with domestic chipmakers, while studies note that Chinese AI chips still generally lag Nvidia in performance.
Tau scaling also sits alongside other approaches aimed at extending computing performance beyond Moore’s law, including neuromorphic computing and quantum computing.

Huawei proposes “Tau scaling law” to extend chip advances beyond Moore’s law

Semiconductors

… company.</figcaption></figure>
whereAdBanner.Uptake
<h3>Recent deals completed in China:</h3>
<ul>
<li aria-level="1"><a href="https://www.heygears.com/">HeyGears</a>, a 3D printing company, has raised more than RMB 300 million (USD 44.1 million) in a Series C round. Legend Capital Management and Fortune Capital co-led the round, with participation from Gopher Asset Management and CAS Investment Management. It will use the funds for photopolymerization technology R&amp;D, global supply chain development, talent acquisition, and the launch of next-generation products. —<a href="https://36kr.com/p/3824073646624899">36Kr</a></li>
<li aria-level="1"><a href="https://www.link-touch.com/">Link-Touch</a>, a six-axis force sensor company, has secured a nine-figure RMB sum in a Series C++ round co-led by SAIC Financial Holdings and Shang Qi Capital. China Fortune-Tech Capital, Charoen Pokphand Robot, and Houwei Capital also participated. The funds will be used for product R&amp;D, digitalized production, and international expansion. —<a href="https://36kr.com/p/3824053061652616">36Kr</a></li>
<li aria-level="1">EBKernel, an embodied intelligence company, has raised a nine-figure RMB sum in a new round led by industrial investors, alongside undisclosed funds and existing shareholders. It will use the funds for technology R&amp;D, team expansion, and market development. —<a href="https://36kr.com/p/3819931562414467">36Kr</a></li>
<li aria-level="1">OmniDimension, an embodied intelligence chip developer, has raised RMB several hundred million in a seed round jointly led by Zhongguancun Development Group and Qihang Investment Management. Shanghai Engine Fund, Stony Creek Capital, Biwin Storage Technology, Yanchuang Group, Haiyi Equity Investment, and Tanyuan Venture Capital also joined the round. —<a href="https://36kr.com/p/3821371042877575">36Kr</a></li>
<li aria-level="1"><a href="http://autoagents.ai">AutoAgents.ai</a>, an agentic AI startup, has completed a pre-Series A round. New investors include Fanchuang Capital, Zhongguancun Development Group, and Tanyuan Venture Capital, while an Orient Securities subsidiary and KylinHall Partners returned as existing backers. The funds will go toward procuring compute, expanding the team, and developing and operating its product ecosystem. —<a href="https://36kr.com/p/3816909945029760">36Kr</a></li>
</ul>
whereRelatedArticle
<h3>FusionAP, Melazyme, TME, and others made recent headlines:</h3>
<ul>
<li aria-level="1">FusionAP, a Malaysia-based semiconductor packaging company, raised USD 2 million in a pre-seed round co-led by Vertex Ventures Southeast Asia &amp; India and Southern Capital Group.</li>
<li aria-level="1"><a href="https://melazyme.com/">Melazyme</a>, a precision fermentation company, raised USD 2 million in a seed round led by Thailand-based SeaX Ventures, with participation from Stellaris Venture Partners and Plug and Play Ventures.</li>
<li aria-level="1">Tencent Music Entertainment (TME) completed its acquisition of Ximalaya on May 18, making the online audio platform a wholly owned subsidiary.</li>
</ul>
If there are any news or updates you'd like us to feature, get in touch with us at: <a href="mailto:hello@kr-asia.com">hello@kr-asia.com</a>.

Deals in brief: Secai Marche raises fresh funding, Singapore’s K25.ai secures at least USD 2 million, five China investments, and more

Venture Capital

Private Equity

Merger &amp; Acquisition

Singapore

Vietnam

Japan

… have connected with more than 840 European electricity suppliers and more than 5,000 heat pump models. These are not things that can be caught up with in one or two years.
There is another point many people overlook: user stickiness. Our app’s DAU (daily active user) rate exceeds 50%, and users open it four times a day on average. They look every day at how much electricity they generated and how much money they saved. Once this habit is formed, the cost of switching to another brand becomes very high. So I am not too worried about competition. Instead, the more imitators there are, the more it means the sector has been validated. That is a good thing for us.
<h3>36Kr: What is your biggest judgment about industry development over the next two to three years?</h3>
LB: My biggest judgment is that only AI-driven balcony energy storage will be competitive in the future. Without AI, hardware alone has no differentiation.
Why? Because hardware will become increasingly homogenized. Core components such as batteries and inverters are similar across suppliers, and costs will become increasingly transparent. But AI capability is different. It is soft and invisible, and yet it is exactly what users can perceive most clearly: how much they pay for electricity each month. That number is the most honest.
So everything we are doing now is essentially building a decision engine with AI at its core. The stronger this is, the better the user experience, the more money users save, and the stronger the stickiness. This is a positive flywheel.
<h3>36Kr: In technology R&amp;D, what are Zendure’s key plans for the future?</h3>
LB: We mainly focus on several directions.
On hardware, the first is lowering hardware costs, the second is some unique thinking around inverter architecture design, and the third is improving efficiency. Hardware ultimately comes down to efficiency and cost. When efficiency rises, heat decreases. When heat decreases, fewer heat dissipation components are needed, and hardware costs naturally fall. So what we are working hardest on is efficiency and cost.
On software, the main focus is deep application of AI. We have already launched an internal strategy to transform into an AI-native organization.
At Zendure, AI is not just an internal efficiency tool. We will use AI from the bottom up to reconstruct our product architecture. Native AI hardware and software products will also be launched in the future.
<h3>36Kr: Beyond energy storage itself, is Zendure considering products for other scenarios?</h3>
LB: We are already doing this. Our electric cargo bike will launch in the second half of this year. We are the only company in the balcony energy storage industry doing this. We have expanded from the home to short-distance outdoor mobility in the yard.
Some European users have even asked if we will make electric vehicles. It is not very realistic for now, but it shows users’ high trust in the brand. If they like you, they hope you will make everything.
We are also exploring heat pumps and air conditioners, though nothing has been finalized yet.
<h3>36Kr: Will Zendure’s design direction change?</h3>
LB: Yes. Our user base is expanding from early technology geeks to mass market users. Especially when couples make purchase decisions together, women’s aesthetic requirements are much higher. So next, we will move in a more appliance-like direction, making the appearance more refined and closer to home appliances.
<h3>36Kr: What are Zendure’s goals for the next few years?</h3>
LB: To lead Europe's plug-in home energy management system market, reach one million users, and complete an IPO. Personally, I lean more toward a Hong Kong listing, because 99% of our revenue comes from overseas. A Hong Kong listing would benefit our overseas business and brand, and would serve as a concentrated brand upgrade for our channels, energy suppliers, and installer ecosystem partners in Europe and North America.
I have communicated with some European investors. Once our brand is mentioned, they all say they are very familiar with it. Because our strategic focus over the years has been overseas, we have kept a low profile in China, so China’s capital markets do not feel our brand as strongly.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3804589375987204">article</a> was written by Xiao Xi for 36Kr.
Note: EUR figures are converted to USD at rates of EUR 0.86 = USD 1 based on estimates as of May 25, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.

Zendure aims to lead in AI-driven home energy management systems

Batteries

Energy