Alibaba

ByteDance

CATL

Huawei

Meituan

Tencent

Xiaomi

Most Followed

KrASIA Connection

Venture Roundup

Earnings Reports

China's Auto Sector

Inside China's Startups

100 AI Creators

View All

Collections

News

Features

Insights

Ecosystem

Press Releases

DeepSeek

Li Auto

Luckin Coffee

Mixue Bingcheng

Pop Mart

Xpeng

On the Radar

Events

Latest

About

Consumer

Robotics

Fintech

Collaborate

<a href="https://klingai.com/global/">Kling AI</a>, an <a href="/industry/ai">artificial intelligence</a> platform for creatives, has announced a major upgrade with the launch of its Video 2.6 model, which now supports native audio generation. The release reflects the company’s push to simplify and consolidate the content production workflow by offering a more complete end-to-end solution.
Video 2.6 integrates the generation of video, dialogue, sound effects, and ambient audio into a single step. This removes the need for creators to use separate tools for visuals and sound and then manually synchronize them, eliminating a bottleneck that has long limited the efficiency of AI-driven media production.
According to Kling, the model’s enhanced semantic understanding enables it to interpret a wide range of inputs, from simple text descriptions and spoken commands to complex, multi-scene storylines. This helps ensure that the visuals and audio it generates align with a creator’s intended direction.
A key upgrade is the system’s expanded audio capabilities and tighter audiovisual synchronization. Video 2.6 can generate a broad spectrum of sounds, including lifelike human voices for speech, singing, and rapping, as well as detailed environmental effects such as breaking glass, crackling fire, and ocean waves. The model also supports granular control through prompts, letting creators specify emotions, tone, rhythm, and even volume—for example, shifting from a whisper to a dramatic scream.
By consolidating these features into a single workflow, Kling expects Video 2.6 to reduce production costs and shorten turnaround times for studios, influencers, and other creative professionals. The company sees the platform as essential for creators who want to deliver immersive visual and auditory experiences with less reliance on post-production.
This article was published in partnership with <a href="https://klingai.com/global/">Kling AI</a>.

Kling AI launches Video 2.6 with native audio generation to speed end-to-end content creation

Kuaishou

East Asia

Kling AI

China

Artificial Intelligence

Deep Tech

Shengshu Technology and <a href="/company/kuaishou">Kuaishou</a> each introduced new multimodal <a href="/industry/ai">artificial intelligence</a> systems this week that emphasize consistency across characters, scenes, and shots. Shengshu expanded its Vidu Q2 model with a full image generation stack, while Kuaishou launched Kling O1, a unified engine for text, images, and video. Although the releases differ in scope, both aim to solve the same problem: making AI systems reliably maintain continuity across images and video.
This marks a shift from earlier stages of generative video development. In 2024, comparisons between models focused on realism, speed, and prompt execution. Tests by KrASIA that year showed that many systems could produce visually striking clips but struggled with consistency, especially across multiple shots. Shengshu’s and Kuaishou’s latest updates point toward a new priority: whether models can handle the structured, multi-shot sequences that creative teams work with every day.
<h3>A push toward reliable video creation</h3>
Even as video generation models improve in visual fidelity, creators using them continue to encounter issues with structural coherence. Characters can shift between frames, props may disappear, and styles sometimes change without clear cause. Even short sequences can lose continuity once the camera moves or a scene changes.
Vidu Q2 and Kling O1 both frame this as the core barrier to wider adoption, especially in short-form video, advertising, and early-stage film production. In previous tests, <a href="https://www.vidu.com/">Vidu’s</a> reference-to-video feature showed how subject and style consistency could improve results. <a href="https://app.klingai.com/global/">Kling</a>, meanwhile, often produced realistic stills but struggled with temporal consistency during fast action. The new releases target these weaknesses.
<h3>Why Vidu and Kling are prioritizing continuity</h3>
Unlike general-purpose AI labs, Shengshu and Kuaishou operate in ecosystems where short-form video and commerce-driven content are central. Their users depend on workflow stability as much as visual quality:
<ul>
<li aria-level="1">Shengshu’s earlier Vidu iterations emphasized fast generation, cost efficiency, and template-driven tools to help creators produce frequent short clips with consistent characters or brand assets.</li>
<li aria-level="1">Kuaishou, whose platform depends on video-led engagement across advertising, live streaming, e-commerce, and short dramas, follows a similar logic. Kling O1 is built not just to generate new video but to edit and refine existing footage, aligning with how creators work on the platform.</li>
</ul>
This shared background explains why both companies now emphasize models that can connect still images, references, and video sequences without introducing character drift or structural inconsistencies.
<h3>What the latest releases add</h3>
Vidu Q2’s new image generation stack expands it from a video-focused tool into a unified system for still and moving images. It supports text-to-image generation, multi-reference identity preservation, and detailed image editing. Shengshu said these features build on Vidu’s strength in varied visual styles while helping creators keep characters consistent across assets.
Vidu Q2 now produces high-resolution outputs up to 4K, with image generation times as fast as five seconds, depending on complexity. Crucially, the still images can be used as references for video sequences, allowing creators to move from character design to motion without recreating assets.
Kling O1 represents a significant architectural shift from earlier versions. Built on a “multimodal visual language” framework, it can process text, images, subjects, and video clips in a single workflow. The system supports combined editing and generation tasks, such as inserting a subject while adjusting the background.
One of the model’s key features is what Kuaishou describes as “director-like memory,” intended to maintain character identity during dynamic camera movements and multi-subject scenes. It supports three- to ten-second generation lengths, and first- and last-frame control is expected to match the same range. Unlike earlier versions that required manual masking or frame-by-frame fixes, Kling O1 aims to deliver pixel-level edits through natural language prompts.
<h3>How the new releases compare with other video models</h3>
The generative video landscape is starting to split into clearer segments:
<ul>
<li aria-level="1">Film-oriented models like <a href="/company/openai">OpenAI’s</a> Sora, Google’s Veo, and Runway’s Gen-4.5, which focus on longer shots and realistic physics.</li>
<li aria-level="1">Short-form and stylized generators, where Vidu Q2 fits, optimized for social content, advertising, and animation.</li>
<li aria-level="1">Unified creation-and-editing engines, represented by Kling O1, combining reference-based generation, video editing, and restructuring.</li>
<li aria-level="1">Avatar-driven systems such as HeyGen and Synthesia.</li>
<li aria-level="1">Previsualization and storyboarding tools that rely on reusable characters and stable asset pipelines, a direction now shared by Vidu Q2 and Kling O1.</li>
</ul>
Across these categories, consistency involves several technical challenges: maintaining character identity across shots and angles, preserving props and spatial layouts, keeping color and style coherent during edits, handling multi-subject scenes, and supporting image-to-video transitions without drift. These capabilities determine whether AI-generated video can move from experimentation to regular production.
Vidu Q2’s unified image-video system and Kling O1’s multimodal engine both attempt to standardize this within their respective workflows.
The releases of Vidu Q2 and Kling O1 indicate that competition in generative video is moving away from headline-grabbing realism and toward workflow stability and continuity. Rather than trying to solve every use case, models are beginning to specialize in formats and production styles where they can deliver reliable, repeatable results.
For platforms built around short-form video and high-volume content creation, the next phase of AI video generation may depend less on the quality of individual clips and more on whether models can maintain coherent characters, scenes, and assets across the broader creative process.

Shengshu and Kuaishou unveil new AI systems to improve generative video consistency

… short-term trend. What’s your view of its long-term outlook?</h3>
MC: It depends on how the brand operates. If you don’t take products, service, store experience, and channels seriously, you’ll fall once the trend fades. But in any category, not just ancient gold, if you outperform others in these areas, you’ll thrive even without a trend.
And I don’t think the current popularity is necessarily a good thing. I’ve seen too many companies expand wildly during booms, only to collapse later. Some brands overprice and mislead consumers. That will eventually backfire.
When a boom comes, instead of letting profit swell, let your capabilities grow. When the boom passes, you’ll still be strong. I agree that some peers may see declining performance after the trend cools.
<h3>36Kr: Is the ancient gold boom related to rising gold prices?</h3>
MC: I don’t think so. Roughly 90% of traditional brands are suffering because of rising gold prices. Many stores have closed, because higher prices suppress consumption.
Customers have been waiting far too long for great products. There just weren’t any before.
<h3>36Kr: Are buyers of ancient gold the same demographic as luxury jewelry buyers?</h3>
MC: Very much so. Previously it was mostly luxury buyers. Now, many customers who used to buy foreign luxury watches buy from Lamchiu as well, or buy both.
We likely have one of the highest proportions of male customers in our category. They are typically people who travel, own yachts, collect watches or antiques, or work in finance.
<h3>36Kr: Lamchiu’s live stream breakthrough happened during the pandemic in 2020. Do you feel you entered online channels too late?</h3>
MC: I have thought before what it might have been like if we had gone online earlier. But over the years, I’ve grown more humble as a designer. I’ve come to feel I was born for this work, so it feels like a kind of mission. And whenever we reached a key moment, things naturally fell into place.
So I now believe that early success isn’t always good. Doing what you are meant to do, and letting the rest unfold, is the best outcome, both for me and for the brand.
<h3>36Kr: The industry is growing so fast. Does that make you want to move faster too?</h3>
MC: Those who want to come, will. I never rush. Lamchiu will grow at a pace suited to its capacity. If you keep staring at the scale of others, you’ll never be happy.
<h3>36Kr: So why raise capital now?</h3>
MC: Lamchiu isn’t capital-intensive. We collect deposits and deliver the goods a year later. Unlike traditional brands, we don’t need to stock inventory, so we barely have capital pressure.
This round wasn’t about the money. We admire Western management systems in manufacturing, capacity building, and training. Chen Dong has deep expertise in these areas, and we want to integrate those resources.
<h3>36Kr: How long did it take for Dayone to invest?</h3>
MC: Nearly a year.
<h3>36Kr: What held you back initially?</h3>
MC: I was reluctant to raise capital. I thought it meant losing freedom, being pushed to grow fast and chase money. So when investors approached me, I refused to meet them.
But Chen reached out more than ten times. Even when his wife was giving birth, he was still talking to me and trying to persuade me.
Over time, I realized his comments were extremely objective. He analyzed Lamchiu’s strengths and weaknesses, and explained why capital and modern management are necessary.
So from rejection to admiration, it took almost a year. As Chen joked, pursuing me was harder than wooing his wife.
<h3>36Kr: What convinced you?</h3>
MC: Not one thing, but everything, bit by bit. His sincerity. His persistence. Every conversation was genuine. Everything he said about operations, management, and planning made sense to me.
<h3>36Kr: What advice from Chen has helped you the most?</h3>
MC: He helped clarify our cultural identity.
<h3>36Kr: What do you think about competing with global luxury brands?</h3>
MC: We have deep respect for the world’s luxury houses. We’ve learned a lot from them, in terms of management, structure, and store design. We don’t see them as competitors. They are our teachers.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3575489738488964">article</a> was written by Li Xiaoxia for 36Kr.

Lamchiu’s handmade gold pieces take months to craft, but deter neither buyers nor investors

Retail

Consumer Economy

… supplier to Nintendo as well as clients in the automotive, space, and consumer electronics segments.
"However, what remains uncertain is how long this momentum will last and how the trend will evolve," Wu said. "Whenever supply gets tight, some customers start placing orders with multiple suppliers to secure supplies, which can lead to double booking somewhere in the supply chain."
This article first appeared on <a href="https://asia.nikkei.com/business/technology/tech-asia/rampant-ai-demand-throws-the-memory-chip-market-into-turmoil">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>.

Rampant AI demand throws the memory chip market into turmoil

Semiconductors

AI Infrastructure

FinVolution Group’s Philippines business has secured a credit facility with HSBC in the Philippines in the seven-figure PHP range. The proceeds will support the expansion of accessible financial services for underserved yet creditworthy Filipinos.
The facility is designed to strengthen financial inclusion by providing timely and sustainable access to individuals with limited or thin-file credit histories. This initiative reflects FinVolution’s commitment to advancing credit technology and building a more inclusive financial infrastructure across its markets.
FinVolution is a global fintech company with operations in China, Indonesia, the Philippines, and Pakistan. As of the third quarter of 2025, it has connected 38 million borrowers with more than 130 financial institutions worldwide, enabling its partners to extend beyond traditional service boundaries through technology-driven innovation.
Tiezheng Li, CEO of FinVolution Group, said:
<blockquote>“This facility is an important step in our ongoing collaboration with institutional partners to expand responsible credit access in growth markets. It also reinforces our commitment to promoting responsible and inclusive financial services for underserved populations.”</blockquote>
The credit facility marks a milestone in bridging traditional banking and fintech platforms, demonstrating how collaboration between banks and fintech companies can deliver scalable and sustainable solutions for inclusive economic growth.
<h3>About FinVolution Group</h3>
FinVolution Group is a fintech company that connects millions of consumers and small and medium enterprises with financial institutions.
Founded in 2007 and listed on the New York Stock Exchange under the ticker code FINV since 2017, FinVolution has been at the forefront of the pan-Asian credit technology industry, developing technologies in credit risk assessment, fraud detection, big data, and artificial intelligence.
This press release was published in partnership with <a href="https://en.finvgroup.com/">FinVolution Group</a>.

FinVolution secures HSBC credit facility to expand financial services in the Philippines

<h3>Singapore-based SynaXG raises over USD 20 million in inaugural round</h3>
<a href="https://www.synaxg.com/">SynaXG Technologies</a>, a deep tech company building wireless infrastructure for the <a href="/industry/ai">artificial intelligence</a> era, has raised over USD 20 million in its inaugural funding round. The round was jointly led by January Capital, Vertex Ventures, and Qualgro.
Founded by CEO Xin Huang, SynaXG develops AI-native software and physical infrastructure to support next-generation wireless networks. The company works with global partners, including chipmakers, network equipment providers, and telecom operators. According to SynaXG, demand is increasing as AI-powered applications such as <a href="/industry/robotics">robotics</a> require real-time adaptability and stable connectivity.
The capital will help accelerate the company’s product development, expand its engineering teams, and deepen collaborations with operators and enterprise partners globally. SynaXG is also preparing for a Series A round to support upcoming commercial rollouts and broader adoption of AI-native radio access network systems across international markets.
<h3>CapitaLand acquires LXA Capital to scale residential mortgage business</h3>
LXA Capital, a residential mortgage investment company, has been acquired by CapitaLand Investment (CLI), marking a step in CLI’s effort to institutionalize residential mortgages as an asset class across Asia. The transaction terms were not disclosed.
In a statement shared on social media announcing the acquisition, LXA said joining CLI will allow it to pursue its strategy on a larger platform and create new synergies for investors.
Founded to build institutional-grade mortgage investment capabilities in the region, LXA added that the acquisition strengthens its ability to deliver value through expanded resources, broader market access, and integration with CLI’s global real assets ecosystem.
whereAdBanner.Uptake
<h3>Recent deals completed in China:</h3>
<ul>
<li aria-level="1">Seno, a material tech company developing ultra thin film delivery systems, has raised USD 700,000 in a pre-seed round. The investor details were not disclosed. Seno is engineering a modular dissolvable film platform intended for future applications in consumer products, wellness, and functional films. The capital will support its R&amp;D expansion, formulation and stability testing, quality and safety system upgrades, and early technical validation as the company prepares for upcoming international evaluations. Seno operates in the US and across Asia.</li>
<li aria-level="1">Anywit Robotics, a maker of <a href="https://kr-asia.com/didi-invests-in-anywit-robotics-to-push-lifelike-human-robot-interaction">head-and-face robot components</a>, has raised an eight-figure RMB sum in a pre-Series A round from several industry investors, including <a href="/company/didi">Didi’s</a> corporate venture capital arm. The capital will support refinements to its standardized expressive head products and upgrades to its emotional interaction models. Winsoul Capital, the lead investor in Anywit’s angel round, served as the financial advisor. —<a href="https://36kr.com/p/3576840991554434">36Kr</a></li>
<li aria-level="1">Borland, a jeweler known for <a href="https://kr-asia.com/jewelry-house-borland-raises-funding-to-tap-chinas-ancient-gold-renaissance">filigree inlay and ancient gold craftsmanship</a>, has raised RMB 100 million (USD 14 million) in a Series A round. The round was led by Challenjers with participation from Kering and Shunwei Capital. The company plans to use the funds to strengthen its branding, expand omnichannel retail operations, upgrade its supply chain capabilities, and invest in talent development. —<a href="https://36kr.com/p/3575489738488964">36Kr</a></li>
<li aria-level="1"><a href="https://beanstar.com/">Beanstar Coffee</a>, a chain coffee franchise brand, has secured an eight-figure RMB sum in a Series B round led by a Suzhou municipal government-managed fund. The company said the capital will support the expansion of its offline stores and branding efforts. Beanstar previously raised angel funding from Shunwei Capital in 2022, followed by a Series A round in 2024 led by Zhengxuan Capital, with Shunwei Capital participating again. —<a href="https://36kr.com/p/3572067116579975">36Kr</a></li>
<li aria-level="1"><a href="https://www.rycc.net/">Ruiyun Cold Chain</a>, a logistics platform, has raised nearly RMB 100 million in a Series A+ round led by Guangzhou Industrial Investment and Capital Operation Holding Group. Founded in 2020, the company provides end-to-end cold chain logistics services supported by digital infrastructure, combining a nationwide offline network with an online freight capacity platform. The new capital will be used to strengthen its digital capabilities and advance its international expansion plans. —<a href="https://36kr.com/p/3569392468032649">36Kr</a></li>
</ul>
whereRelatedArticle
<h3>Roojai, Cove, Amperesand, and more made recent headlines:</h3>
<ul>
<li aria-level="1">Roojai, a digital insurer in Thailand, raised USD 60 million in a Series C round. The investment was co-led by funds under Apis Partners Growth and Asia Partners, with participation from existing shareholders HDI International, Primary Group, and the International Finance Corporation.</li>
<li aria-level="1"><a href="https://cove.sg/">Cove</a>, a co-living solutions provider, is set to acquire Casa Mia Coliving, a Singapore-based operator with about 500 furnished rooms in central districts. The deal expands Cove’s total inventory to more than 2,000 rooms in Singapore and 8,000 rooms across the Asia Pacific.</li>
<li aria-level="1"><a href="https://amperesand.io/">Amperesand</a>, a power infrastructure company serving artificial intelligence data centers and other critical power applications, raised USD 80 million in a Series A round co-led by Walden Catalyst Ventures and Temasek, with participation from Industry Ventures, Acclimate Ventures, SG Growth Capital, and existing backers including Xora Innovation, Material Impact, TDK Ventures, and Foothill Ventures.</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>.
<style>.addcaption{font-size:14px!important;line-height:20px!important;color:inherit;padding-top:8px;border-top:1px solid #e8e8e8}</style>

Deals in brief: SynaXG raises over USD 20 million, CapitaLand acquires LXA Capital, new China investments, and more

Venture Capital

Private Equity

Investment

Merger &amp; Acquisition

Singapore

Southeast Asia

Anywit Robotics has raised an eight-figure RMB sum in a pre-Series A funding round from several industry investors, including <a href="/company/didi">Didi’s</a> corporate venture capital arm. The capital will be used to refine its standardized expressive head products and upgrade its emotional interaction models. Winsoul Capital, the lead investor in Anywit’s angel round, served as the financial advisor for the transaction.
Founded in December 2023, Anywit develops multimodal interactive robots designed to emulate the vitality of humans. The company has released a head-and-face component for humanoid robots that features 34 degrees of freedom. According to the company, the system delivers embodied expressions comparable to those of humans, including eye contact and audio-lip synchronization.
The facial interaction component integrates a multimodal large model, a proprietary small model for emotion generation, and a motion planning system capable of handling multiple intents. This setup is intended to support interaction experiences that feel more natural to users.
<figure id="attachment_158689" aria-describedby="caption-attachment-158689" style="width: 420px" class="wp-caption aligncenter"><img loading="lazy" decoding="async" class="size-full wp-image-158689" src="https://console.kr-asia.com/wp-content/uploads/2025/12/Anywit-Robotics-Demo-Expressions.webp" alt="" width="420" height="236" /><figcaption id="caption-attachment-158689" class="wp-caption-text">Graphic source: Anywit Robotics via 36Kr.</figcaption></figure>
Cao Rongyun, founder and CEO of Anywit, told 36Kr that the company leads the market in the segments where it specializes. He added that its technology benchmarks globally against Ameca, a recognized industry leader. Anywit is also advancing its commercialization efforts. It has introduced interactive robots equipped with expressive head-and-face components for use in education and research, marketing reception, and entertainment.
Anni, a robot developed by Anywit, appeared at this year’s World Robot Conference and the 27th China Hi-Tech Fair. The company has delivered preliminary units of its educational robots, established a standardized product line, and deployed robotic teachers in primary school classrooms.
<figure id="attachment_158688" aria-describedby="caption-attachment-158688" style="width: 420px" class="wp-caption aligncenter"><img loading="lazy" decoding="async" class="size-full wp-image-158688" src="https://console.kr-asia.com/wp-content/uploads/2025/12/Anywit-Robotics-Demo-Emotions.webp" alt="" width="420" height="236" /><figcaption id="caption-attachment-158688" class="wp-caption-text">Graphic source: Anywit Robotics via 36Kr.</figcaption></figure>
Looking ahead, Cao told 36Kr that Anywit plans to expand its product exploration, with mass shipments of standardized robotic head-and-face components targeted for 2026. Drawing a parallel to the role graphical user interfaces played in bringing computers into homes, and how the iPhone’s touch screen and app ecosystem ushered in the mobile internet era, he believes mature facial expression interaction technology will be a critical breakthrough for broader robot adoption.
Anywit’s team includes graduates from the University of Science and Technology of China. The group has nearly a decade of research experience in human–machine interaction and has earned awards in several international <a href="/industry/robotics">robotics</a> competitions.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3576840991554434">article</a> was written by Qiu Xiaofen for 36Kr.

Didi invests in Anywit Robotics to push lifelike human-robot interaction

Didi

Embodied Intelligence

… are the teacher’s problems, search and exploration are the student’s problem-solving process, and the reward model is the teacher’s feedback.
Choosing the right difficulty is crucial. Make it too advanced and the student gives up, but if it’s too easy, they learn nothing. The same applies to reinforcement learning: data volume alone doesn’t help. Appropriateness does.
<h3>36Kr: How does reinforcement learning relate to embodied agents?</h3>
WY: The relationship goes two ways. One is locomotion, where reinforcement learning is already mature and doesn’t require pre-training.
The other involves long-horizon reasoning and planning, usually combined with large pre-trained models. That area only became popular after ChatGPT.
These two aspects form a spectrum from high-frequency control for short tasks to abstract reasoning for complex ones.
Traditional reinforcement learning for control doesn’t need pre-training, think quadruped robots that can run and jump. Tiny neural networks trained in simulation can directly control real robots without pre-training.
In such tasks, reinforcement learning trains the network to output control signals for each joint, enabling motion over durations as short as seconds, not hours. By contrast, models like ChatGPT and <a href="/company/deepseek">DeepSeek's</a> R1 use reinforcement learning after pre-training to enhance reasoning.
Large models that employ reinforcement learning can think for minutes or hours, use common sense, break complex problems into subtasks, and call tools. But so far, this success remains in the digital realm, not the physical.
In between lies the vision-language-action (VLA) model, which is often discussed in embodied intelligence research.
<h3>36Kr: How do we move from VLA to fully embodied intelligence?</h3>
WY: VLA applies pre-training to the physical world. Researchers gather massive data to pre-train models that not only complete short tasks like running or jumping but also generalize to minute-long activities like folding towels and pouring water.
To reach longer-range tasks like cooking or cleaning, robots must perform for hours, combining fine control with abstract reasoning and human interaction, just like digital agents, except it’s in the physical world.
So I see embodied agents as systems that merge locomotion or VLA as the “small brain” controlling motion, with language models enhanced using reinforcement learning as the “big brain.”
Unlike digital agents, physical agents still get less attention. Most people focus on hardware aspects such as gripping accuracy, object sorting, and so on. But altering the physical world is always harder.
Given my focus, I’m working on stabilizing long-duration reasoning before combining it with physical control.
<h3>36Kr: How will reinforcement learning and VLA eventually integrate?</h3>
WY: Our current plan is a layered structure. As I said at WAIC, the higher you go, the more human knowledge you need. Likewise, the lower, the less.
Lower layers handle instinctive reactions like grabbing a cup based on tactile feedback or simple physics. Upper layers need prior knowledge. So there’s a natural division between digital and physical agents.
I don’t think VLA will be the final paradigm, because it isn’t large enough in scale to become a fully capable agent. We’ll perfect the digital agent format first while others explore the embodied side, then merge them when the timing is right.
<h3>36Kr: At the September conference, you said your AReaL team seeks an entirely new, minimalist organizational model. Why?</h3>
WY: In the internet era, building a product required four or five people which typically includes a frontend developer, a backend developer, and a product manager.
In the AI era, one person might be sufficient to handle all that.
Previously, small companies outsourced many tasks. Now, AI can streamline not just internal work but also outsourcing itself.
If a team can run heavily on AI, its capabilities will naturally scale outward, because if AI can serve us, it can serve others, too. That's a new product opportunity.
Our AReaL team has only six members, with some external support. Counting everyone, we could still make it leaner. I want the team to stay minimalist, and that’s why it has always been small.
<h3>36Kr: But large companies have complex structures. How do you achieve that simplicity within one?</h3>
WY: First, a modern agent-focused team must use many agents every day itself.
Second, I combine algorithm and infrastructure teams into a single full-stack unit.
Traditional structures separate algorithms from systems and add data collection teams, creating a segregation dynamic whereby algorithm teams are the “clients,” while engineers become the contractors doing the “dirty work.” That division kills innovation. Once you’re used to being the former, you avoid the grunt work. And if you’re in the latter, you lose creative space.
OpenAI didn’t magically invent new algorithms, it simply perfected the details.
So to excel in infrastructure and data, you need to dig deep. With that groundwork, algorithms can shine. That’s why algorithms and infrastructure must be co-designed and co-evolved. A small, highly capable team can collectively fulfill this.
Large organizations, say with 200 people, can’t avoid silos. Limited communication bandwidth leads to rigid roles and inefficiency. So a compact, full-stack setup and high innovation go hand in hand. Forget the 200-person org chart. In the AI era, it’s all about going from zero to one, so take bold, radical approaches and build anew.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3566367968459910">article</a> was written by Fu Chong for 36Kr.

What will define AI? AReaL head Yi Wu points to reinforcement learning

Foundation and Large Models

… Lab, <a href="/company/shein">Shein</a>, Insta360, <a href="/company/pony-ai">Pony.ai</a>, <a href="/company/dreame">Dreame</a>, and <a href="/company/xpeng">Xpeng Motors</a>, many of which faced skepticism or volatile growth trajectories.
When IDG invested in Bambu Lab in late 2020, consumer 3D printers were widely dismissed as niche, slow, and unreliable. IDG did not bet on the market but on the team of engineers led by founder Ye Tao. When the company’s first prototype showed that a 3D printer could be fast, reliable, and easy to use, IDG doubled down. Its conviction was eventually rewarded.
The same logic applied to Insta360. In a category dominated by GoPro, IDG backed a young founder born in the 1990s, Liu Jingkang, and ultimately became the company’s largest outside shareholder. “We believe in the new generation,” Niu said. “They have strong instincts for products, aesthetics, and global markets.”
As enthusiasm for new consumer brands cooled, revenue-based financing gained momentum by targeting businesses with stable cash flow. IDG avoided such short-term, small-ticket investments. Instead, it expanded its legacy in travel and hospitality through Trip.com, Homeinns, H World Group, Wuzhen, and Yada International by investing in longer-cycle cultural tourism assets.
One example is the Yaohu Town project, with a total investment of RMB 13 billion (USD 1.8 billion). Created by Yada founder Jiang Jianning and Wuzhen’s Chen Xianghong, it aims to build a resort town on 20,000 acres of planned land. Construction was delayed by the pandemic, but IDG stayed committed, securing new funding and local partners. Yaohu Town opened on December 28, 2024, and within one year reached its annual revenue target of RMB 365 million (USD 51.1 million).
Across more than 30 years of investing, with assets under management among the largest in the industry, IDG’s primary capability has remained consistent: a willingness to embrace innovation, risk, and uncertainty, grounded in due diligence, long-term thinking, and patience.
Partner Lian Meng, who backed Heytea and Shein, summarized the philosophy: “The most important thing in life is long-term companionship.” This applies to founders and increasingly to assets. IDG is willing to wait for value to reemerge after hype fades.
Patience also shapes its internal learning. In Luckin’s case, Niu said IDG’s understanding evolved over time. Initially, margins were the priority. Later, IDG realized that in a hypercompetitive coffee market, Luckin’s returns would come from scale, leaving room for course correction.
Lian reflected on his own shift in perspective. “I used to think ‘good’ meant products I would give to friends,” he said. That instinct helped IDG identify brands defined by strong visual identities, such as To Summer and Acne Studios. “But taste is subjective,” he added. “After <a href="/company/pinduoduo">Pinduoduo</a> and Luckin, I realized how essential it is to understand diverse consumer perspectives. And I should respect Pinduoduo and Luckin more than most people.”
<h3>Entry and evolution</h3>
Not every investment leads to success. Tiger Attitude, a bakery chain, was one of IDG’s most significant misses during the 2022 consumer boom. According to managing director and consumer lead Fan Wu, the firm’s internal review concluded that the issue was not consumer brands themselves but inflated valuations and unrealistic expectations.
The team gained new insight: unlike bakery goods, coffee consumption is not constrained by frequency. Coffee is both functional and cultural.
The introspection helped IDG stay disciplined as the consumer-brand frenzy continued, deliberately avoiding overpriced deals. “Not investing was already a win,” Wu said.
Through these cycles, IDG’s consumer strategy has evolved across three phases:
<ul>
<li aria-level="1">The first focused on global brands entering China, such as Moncler and Farfetch.</li>
<li aria-level="1">The second emphasized new channels, brands, and categories, capitalizing on <a href="/industry/e-commerce">e-commerce</a> and backing companies such as Heytea, To Summer, and Genki Forest.</li>
<li aria-level="1">The third and current phase emphasizes technology and globalization, leading to investments in Luckin, Bambu Lab, Shein, Insta360, and <a href="/company/anker">Anker</a>.</li>
</ul>
IDG supports Chinese consumer companies expanding overseas while helping international brands enter China, including Acne Studios, Gentle Monster, Rossignol, and Lakrids by Bulow.
If innovation and globalization define the third phase, acquisitions are becoming the primary tool. Wu, whose background includes Blackstone and Bain Capital, helps drive this strategy. She believes China is entering a period of lower interest rates and stable GDP growth, making stable excess returns more important. Mergers and acquisitions in the consumer sector fit this environment.
But execution is challenging. “Acquisitions require long negotiations and deep operational understanding, from due diligence through post-deal management,” she said.
IDG’s roadmap highlights two main opportunities: exits by multinational firms that lead to sales of controlling stakes and management buyouts in partnership with experienced operators. This marks a shift from minority investing toward more active control.
This fall, IDG acquired a majority stake in Lakrids by Bulow, a Danish confectionery brand, with plans for global expansion rather than a China-only strategy. “In the Moncler era, IDG took small equity positions and helped with China expansion. Now we have the confidence and ability to manage a global consumer company,” Wu said. Capabilities gained from past deals, such as acquiring Osram with MLS and Metro China with Wumart, are now flowing into the consumer sector.
Seven years ago, 36Kr asked IDG what it ultimately wanted to become. Today, the answer is clearer. As the consumer sector continues to evolve, IDG, now in its 32nd year, is becoming a more sophisticated hybrid. It aims to remain steady while embracing new variables. It seeks to serve as a bridge for global brands entering China, a global operator for Chinese brands going abroad, and an investor capable of taking either an active or minority role depending on the situation.
Its high-stakes investment in Luckin remains one of the clearest expressions of what the firm has become. And at 32, IDG is evolving into the institution it strives to be.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3561359993994113">article</a> was written by Chen Zhiyan for 36Kr.

IDG’s Luckin gamble, and how the cautious giant learned to take risks