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

… partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.

BYD adds to China's expanding automotive footprint in Pakistan

East Asia

China

Transportation

Electric Vehicles

Mobility

Central Asia

As fireworks lit up the sky during the Lunar New Year, another kind of frenzy was unfolding in China’s large model industry.
At one end, tech giants were spending heavily on “red packet” giveaways and free beverage promotions to draw users to their <a href="/industry/ai">artificial intelligence</a> apps. At the other, the so-called “little tigers,” a colloquial term for China’s leading large model startups, were seeing their valuations rise as fresh capital flowed in.
Among them, the biggest deal so far this year likely belongs to Moonshot AI.
<a href="https://www.bloomberg.com/news/articles/2026-02-17/china-ai-startup-moonshot-seeks-10-billion-value-in-new-funding">Bloomberg</a> previously reported that Moonshot is close to extending an <a href="https://kr-asia.com/moonshot-ai-lines-up-new-funding-at-usd-4-8-billion-valuation">existing funding round</a>, originally estimated at USD 500 million. Already backed by Alibaba, Tencent, and 5Y Capital with a total round size of USD 700 million, the extension could lift Moonshot’s valuation to about USD 100 billion or more.
36Kr has learned that, in addition to investors already disclosed, the round extension may include existing backer Gaorong Capital and a new entrant, Cathay Capital. For Cathay, this could be its first investment in a large model company.
Sources familiar with the matter said Moonshot’s fundraising has attracted interest from multiple institutions, including overseas funds with European ties. As of press time, Moonshot AI had not responded to requests for comment.
“Scramble” may be the most accurate word to describe this round. One person close to the deal told 36Kr that since the second half of 2025, when Moonshot began raising funds, limited partners at several institutions have been pressing their general partners to participate.
Recent rounds have also drawn investors who had previously stayed on the sidelines of the large model race. Cathay Capital is one example. As recently as May 2025, managing partner Duan Lanchun told 36Kr that the firm had not invested in general-purpose large models or core infrastructure, preferring to focus on applied deployment and ecosystem development.
This may be the industry’s biggest wave of bandwagoning since early 2023.
The surge has been fueled largely by the rising market capitalizations of Z.ai and MiniMax, whose Hong Kong IPOs have revived confidence in the sector.
Since going public at the start of the year, both companies have seen steady gains. On February 12, after each launched a next-generation model, their market caps reached new highs. As of February 17, Z.ai’s intraday valuation surpassed HKD 220 billion, while MiniMax topped HKD 260 billion. Both have risen roughly four to five times their IPO valuations.
Strong returns in the secondary market have drawn primary investors toward pre-IPO companies such as Moonshot, eager to secure positions ahead of a potential listing.
The Hong Kong listings have also boosted global visibility for China’s large model developers. Notably, European capital, which had shown limited interest in Chinese tech, has now taken a closer look at Moonshot.
According to a source, new investors in the ongoing round had to move quickly to secure allocations. In effect, the new tranche, priced at a valuation above USD 100 billion, was driven by institutions that missed out in the previous round.
“The market caps of Z.ai and MiniMax have provided a valuation anchor for large model companies in the secondary market,” another source said.
Previously, Moonshot and StepFun were valued more conservatively. Even at USD 100 billion, Moonshot would still lag behind Z.ai, at around USD 128 billion, and MiniMax, at about USD 153 billion.
As the secondary market validates the value of large model developers, Moonshot and StepFun have adjusted their fundraising strategies. Since mid-2025, both have adopted rolling fundraising models to accelerate valuation growth.
For Moonshot, the shift has been dramatic. It first announced a USD 500 million round by the end of 2025, which quickly expanded to USD 700 million. The latest extension, potentially pushing its valuation to or past USD 100 billion, would represent more than a twofold increase in just two months.
whereRelatedArticle
Across China, large model companies are now racing to raise funds in both primary and secondary markets. Meanwhile, the tech giants—<a href="/company/bytedance">ByteDance</a>, <a href="/company/alibaba">Alibaba</a>, and <a href="/company/tencent">Tencent</a>—have entered a new phase: spending aggressively to capture users.
During the Lunar New Year, these firms sponsored national broadcasts and poured money into “red packet” campaigns and beverage giveaways, hoping to make their AI apps household names.
For the remaining startups, the more pressing task is staying in the top tier of foundational model R&amp;D.
In a recent interview, StepFun chairman Yin Qi said developing base models requires annual spending of RMB 3–5 billion (USD 420–700 million) just to remain competitive.
For startups, buoyant valuations and investor enthusiasm offer a rare window of optimism. Whether that optimism can translate into lasting technological leadership, however, remains an open question.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3688162369908611">article</a> was written by Zhou Xinyu for 36Kr.

Moonshot AI sparks investor frenzy as Z.ai and MiniMax reset valuation benchmarks

Artificial Intelligence

Deep Tech

Foundation and Large Models

… funds toward clinical applications targeting movement disorders, refractory neurological diseases, and sensory impairments. —<a href="https://36kr.com/p/3681333437247106">36Kr</a></li>
<li aria-level="1">Huafeng Integrated Circuits, a semiconductor solutions provider, has completed its Series A round, raising RMB 300 million (USD 42 million). Investors include the Beijing High-Precision and Cutting-Edge Industry Development Investment Fund, Lochpine Capital, a subsidiary of <a href="/company/catl">Contemporary Amperex Technology</a> (CATL), Zhongchuang Juyuan Fund, GF Xinde Investment, Wisdom Capital, and Nanotrend Capital. Separately, it signed RMB 2.3 billion (USD 322 million) in syndicated loan agreements led by the Agricultural Bank of China, with participation from Postal Savings Bank of China, China Construction Bank, Bank of China, and other institutions. The company will use the proceeds to advance its semiconductor packaging technology and expand industrial-scale production. —<a href="https://36kr.com/p/3680183800213380">36Kr</a></li>
<li aria-level="1">Nuwa Agricultural Technology, an agricultural robotics company incubated by Skyworks, has secured an eight-figure RMB investment from Xiamen Guosheng Fund. Founded in December, it focuses on applying embodied robotics to livestock farming. The capital will support R&amp;D, product engineering upgrades, and scaled deployment. —<a href="https://36kr.com/p/3681060245581448">36Kr</a></li>
<li aria-level="1">Kingwills, an advanced materials company specializing in flash-spun materials, has completed a Series C+ round jointly led by industry funds affiliated with Sinochem Capital and CRHC Fund. It will use the proceeds to accelerate technology development and strengthen its global market position. —<a href="https://36kr.com/p/3678625523311496">36Kr</a></li>
<li aria-level="1">Philinks, an optical switching chip developer, has raised RMB 40 million (USD 5.6 million) in an angel round led by Zhongying Capital, with participation from BCT Capital. The company is developing nanosecond-level optical switching systems based on arrayed waveguide grating router technology. It plans to use the funds to scale mass production, expand its team, and develop next-generation products. —<a href="https://36kr.com/p/3678316098134914">36Kr</a></li>
<li aria-level="1">BCIFlex, an invasive BCI company, has raised RMB 300 million in a Series A+ round led by Casstar. Existing investors, including Legend Capital Management, Beijing Pharmaceutical and Health Industry Investment Fund, IDG Capital, Shunwei Capital, Lenovo Capital, Yuanbio Venture Capital, HongShan, <a href="/company/meituan">Meituan's</a> DragonBall Capital, <a href="/company/baidu">Baidu</a> Ventures, Apricot Capital, and Jiayin Capital, also participated. The company will use the funds to support clinical trials and further iterate its BCI systems. —<a href="https://36kr.com/p/3677251288654473">36Kr</a></li>
<li aria-level="1">Galaxea AI, one of China’s leading robotics companies, has closed a RMB 1 billion (USD 140 million) Series B round, bringing its total funding to nearly RMB 3 billion (USD 420 million). Investors in the round included Jinding Capital, BAIC Capital, Hone Capital, Loyal Valley Capital, Qianhai Ark, and InnoVen Capital, alongside follow-on participation from Cathay Capital, DragonBall Capital, Today Capital, Xiang He Capital, and GL Ventures. The company will use the capital to accelerate product commercialization and expand its ecosystem. —<a href="https://36kr.com/p/3678199520846464">36Kr</a></li>
<li aria-level="1">Agilink, a spinoff of Agibot, has completed a new funding round worth a nine-figure RMB sum. The round was led by an undisclosed internet technology company, with participation from Baidu Ventures, Yunfeng Capital, Synstellation Capital, Joyson Electronics, Longcheer, SAIC Financial Holdings, C Capital, and other financial institutions. Existing shareholders BlueRun Ventures and Hillhouse Capital joined during the oversubscription phase. Agilink will use the proceeds to iterate its products and scale manufacturing capacity for high-precision robotic hands. —<a href="https://36kr.com/p/3678179258425990">36Kr</a></li>
<li aria-level="1">Regend Therapeutics, a regenerative medicine company founded in 2015, has completed a RMB 350 million (USD 49 million) Series C round. New investors include Universal Capital, Hefei Hi-tech VC, Hongtai Aplus, Hefei Industry Investment, and Gongqingcheng Fuhui, with follow-on participation from FIRHealth Capital and Tasly, among others. The company focuses on stem and progenitor cell therapies for organ repair and enhancement. It will use the funds to advance clinical development and commercialization of its cell and gene therapy pipeline. —<a href="https://36kr.com/p/3677602224841353">36Kr</a></li>
</ul>
whereRelatedArticle
<h3>Sleek EV, Polybee, Windrose, and others made recent headlines:</h3>
<ul>
<li aria-level="1"><a href="https://www.sleekev.com/">Sleek EV</a>, a Singapore-headquartered electric scooter manufacturer operating in Thailand, raised USD 8.5 million in the first close of its Series A round. The round was led by Kymco Capital, the investment arm of Kwang Yang Motor, alongside more than 20 investors, several of them from China.</li>
<li aria-level="1"><a href="https://polybee.co/">Polybee</a>, a Singapore-based agritech company, has raised USD 4.3 million in a seed round. The round was co-led by Paspalis Capital and Elev8.vc, with participation from Seeds Capital and angel investors.</li>
<li aria-level="1"><a href="https://windrose.tech/">Windrose Technology</a>, an electric truck manufacturer, launched a Series C round targeting a total raise of more than USD 100 million, with the C1 tranche led by a fund managed by China Mobile. The company also secured credit lines from several state-owned banks.</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: Grab to acquire Stash, Gobi Partners backs Valiance Health, Humain invests in xAI, and more

Venture Capital

Private Equity

Investment

Merger &amp; Acquisition

Singapore

Malaysia

Japan

Saudi Arabia

Southeast Asia

Middle East

A quiet fault line runs through China’s music consumption market.
On one side sits a large pool of potential users. Musical instrument consumption in China has been growing at an annual rate of more than 12%.
On the other side is a steep conversion cliff. Learning a traditional folk guitar typically requires clearing three hurdles: memorizing chords, mastering finger placement, and controlling rhythm. At any of these points, many beginners give up within two months of buying an instrument, consigning it to the recommerce platform Xianyu (also known as Idle Fish), or sliding it under the bed.
For years, instrument makers have oriented their products toward professional players, structuring entry-level instruction around graded examination systems. Yet for a much larger group of consumers, those who simply want to express themselves through music in a fleeting moment, that demand has largely gone unmet.
LiberLive positions itself along this divide. The stringless guitar, a product category that has existed for just four years, is now appearing on mainstream consumer electronics shelves.
According to the company, LiberLive’s new U1 stringless guitar topped Tmall’s sales rankings on its first day of launch. In its first month on the market, the U1 and C2 series occupied three of the top spots on the category’s bestseller list. C series shipments have surpassed 500,000 units cumulatively, based on official figures released by the company.
The U1’s design attempts to distill elements of a professional musician’s toolkit and adapt them for casual users.
Compact and portable, the U1 features a 60-centimeter body, weighs 1.5 kilograms, and adopts a detachable folding structure. With a single press and strum using a magnetic pick, users can generate real-time chord progressions. Indicators light up where the user presses, lowering the learning barrier. The companion app’s “smart mode” can automatically match more complex chords.
At its core, the interaction model turns performance into an immediate dialogue between user and instrument. During a commute, at an outdoor campsite, or in a parent-and-child moment at home, the U1 is designed to fit into fragmented daily scenarios, separating musical expression from traditional venue and skill constraints.
If music theory is no longer a prerequisite for playing, another question emerges. Users can produce sound with the U1, but how do they shape it into the emotion they want?
The device includes a function intended to address that gap. By tapping the body of the guitar, users can switch among seven preset emotional styles. The same chord progression can be rendered as upbeat or subdued. The sound library expands to nearly 40 instruments, integrating piano, harp, and ukulele tones into a single device. Three built-in drum machine modes support manual switching, allowing a solo user to layer multiple tracks. Environmental, rhythm-based, and pet-inspired sound effects further simplify how users trigger musical interactions.
These functions are powered by <a href="https://kr-asia.com/liberlive-unveils-liberaos-a-high-dynamic-audio-system-for-its-c2-stringless-guitar">LiberAOS</a>, the company’s proprietary high-dynamic audio sampling and synthesis system. A magnetic pick captures fingertip movements, which are processed through what the company describes as a tri-core heterogeneous computing architecture to deliver near-zero-latency feedback, meaning sound responds as soon as the pick strikes. A high-volume sampled audio engine restores harmonic overtones and resonance details, intended to preserve a sense of realism in synthesized tones. The dual-frequency speaker system includes a 10-watt mid-to-high frequency unit and a 15-watt bass unit, which the company says achieves Hi-Fi-level acoustic standards within a compact body. These technical claims are based on company descriptions.
whereRelatedArticle
From perception to computation to sound output, multiple layers of technical configuration operate behind the scenes. Yet when a user picks up the U1, there is no need to understand the parameters. The intended experience is immediate and intuitive: a playable sound without formal training.
Around the hardware, an ecosystem is taking shape. The U1 supports nearly 40 instrument tones, including piano, harp, and ukulele, enabling varied performance needs with one device. It integrates three drum machine modes, allowing a single user to approximate a band effect. The dedicated app hosts millions of user co-created scores, according to the company, and sing-and-play videos can be exported and shared with one tap. The newly launched U1 dock and U1 quick-release strap are designed to reduce setup time before playing.
The timing of the U1’s launch appears deliberate. As sheet music libraries, sound packs, and user communities accumulate around the same hardware, LiberLive’s positioning is shifting from supplying stringless guitars to offering a faster pathway to music expression.
By reframing music performance as an interactive, emotion-driven, real-time activity rather than solely a skill-based system, LiberLive is attempting to flatten the learning curve, starting with the guitar. The next instruments to be reimagined could include the piano, the drum kit, or other tools long associated with high learning costs.
The broader instrument market is also adjusting its value anchor. As hardware parameter competition approaches saturation, incremental users are shifting their decision criteria from technical performance to immediacy. Legacy brands such as Yamaha and Roland continue to emphasize narratives of long-standing craftsmanship. LiberLive is taking a different approach, focusing less on making instruments more technically sophisticated and more on moving them from the professional shelf into the everyday flow of ordinary life.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3681146132836226">article</a> was written by Huang Nan for 36Kr.

With the U1, LiberLive aims to address music instrument dropout

Music

Creative

… sentiment.
<h3>Micropolis secures additional robot orders after Saudi pilot</h3>
Micropolis Robotics completed a pilot deployment of security robots at an Aramco refinery and the Riyadh National Power Plant in cooperation with Saudi Arabia’s Ministry of Interior, according to <a href="https://www.investing.com/news/company-news/micropolis-completes-saudi-pilot-receives-order-for-270-more-robots-93CH-4481726">Investing.com</a>. The pilot validated the autonomous operation of its unmanned ground vehicles in high-security environments.
Following the pilot, the company received a letter of intent from QSS Robotics for an additional 270 robots, increasing its potential total order volume to 770 units. The Ministry of Interior is evaluating broader nationwide deployment, with the technology positioned to support Saudi Vision 2030.
<h3>CrowdStrike and Saudi Aramco advance cybersecurity cooperation</h3>
Cybersecurity firm CrowdStrike has signed a MoU with Saudi Aramco to expand cybersecurity capabilities in Saudi Arabia, <a href="https://www.investing.com/news/assorted/crowdstrike-signs-mou-with-aramco-for-saudi-cybersecurity-expansion-432SI-4487673">Investing.com</a> reported.
The collaboration will focus on long-term cybersecurity investment, protection of critical infrastructure, and the acceleration of AI-related initiatives. The parties are also exploring the establishment of a regional headquarters in Saudi Arabia and the deployment of local cloud capabilities, aligned with Saudi Vision 2030’s goals to enhance national cybersecurity and support AI-driven economic development.
<h3>Saudia in talks with Boeing and Airbus on aircraft order</h3>
Saudi Arabia’s national carrier Saudia is in preliminary talks with Boeing and Airbus regarding a potential purchase of at least 150 narrowbody and widebody aircraft, according to <a href="https://www.bloomberg.com/news/articles/2026-02-05/boeing-airbus-in-talks-with-saudi-arabia-for-historic-jet-order">Bloomberg</a>.
If finalized, the deal would represent the airline’s largest aircraft order to date. The procurement is intended to partially renew its fleet and expand overall capacity, supporting Saudi Arabia’s strategy to develop a global aviation and tourism hub. In recent years, Saudia has placed multiple large aircraft orders and implemented management restructuring, service upgrades, and technology enhancements as part of its transformation plan.
<h3>QIA renews commitment to fund-of-funds program</h3>
Mohammed bin Abdulrahman Al Thani, Qatar’s Prime Minister and minister of foreign affairs of Qatar, said Qatar Investment Authority (QIA) plans to expand its fund-of-funds program by USD 2 billion to further support startups and venture capital, according to <a href="https://www.agbi.com/entrepreneurs/2026/02/qia-ups-startup-backing-with-2bn-funding-pool/">AGBI</a>.
He added that Qatar will combine capital with technological capabilities, including dedicated computing resources for local startups. Data cited indicate that startup incentives in Qatar increased eightfold over the past year, with the number of registered startups doubling and applications exceeding 6,000.
<h3>QIA becomes founding investor in Hongkong Land’s Singapore fund</h3>
QIA has announced it is joining Hongkong Land’s first private real estate fund, the Singapore Core Private Real Estate Fund, as a founding investor.
Launched with assets under management of SGD 8.2 billion (USD 6.3 billion), the fund focuses on prime office-led commercial properties in Singapore’s central business district. Qatar Investment Authority said the investment reflects its long-term confidence in Singapore’s position as a global business hub.
<h3>Invest Qatar partners Doha Bank to support foreign investors</h3>
Invest Qatar signed a cooperation agreement with Doha Bank during Web Summit Qatar 2026 to improve market access for foreign investors and strengthen Qatar’s position as a global investment destination, <a href="https://www.gulf-times.com/article/719940/business/invest-qatar-doha-bank-partner-to-offer-tailored-financial-services-to-foreign-investors">Gulf Times</a> reported.
Under the agreement, both parties will provide “one-stop services” for foreign enterprises, including business matching, knowledge sharing, and corporate referrals.
<h3>WareOne launches logistics marketplace platform</h3>
Qatari startup WareOne has launched a digital logistics platform described as an “Airbnb-style” marketplace for logistics and warehousing space, according to <a href="https://www.gulf-times.com/article/719948/business/qatar-startup-delivers-30-savings-with-airbnb-for-warehouses">Gulf Times</a>.
The company said its pay-as-you-go model can reduce storage costs for small and medium enterprises by up to 30%. In the 11 months since launch, WareOne has expanded to 25 partners across Qatar, the UAE, Kuwait, and Bahrain, serving more than 40 customers.
<h3>Qatar signs multiyear cooperation agreement with TikTok</h3>
According to <a href="https://www.gulf-times.com/article/719930/qatar/state-of-qatars-government-communications-office-and-tiktok-sign-multi-year-mou-to-support-global-creative-excellence">Gulf Times</a>, Qatar’s Government Communications Office (GCO) signed a multiyear MoU with <a href="/company/tiktok">TikTok</a> during Web Summit Qatar 2026 to launch a three-year international creator support program.
Jassim bin Mansour bin Jabor Al Thani, a director at the office, said the partnership reflects a shared commitment to creating long-term value for content creators. He said Qatar is positioning Doha as a platform for creators to grow by offering space, infrastructure, and institutional support.
<h3>Qatar’s GCO signs MoU with AmCham Qatar</h3>
Qatar’s Government Communications Office also signed a MoU with the American Chamber of Commerce in Qatar during Web Summit Qatar 2026, <a href="https://thepeninsulaqatar.com/article/05/02/2026/gco-signs-mou-with-amcham-qatar">The Peninsula</a> reported.
The agreement is intended to attract more US startups, investors, and decision-makers to participate in the summit and to deepen bilateral cooperation in technology and innovation. The memorandum is part of a broader series of partnerships concluded during the event to strengthen international collaboration and support the continued development of Qatar’s.
This <a href="https://mp.weixin.qq.com/s/O2Di5s_njljyGUP3dv2v-g">article</a> was adapted based on a feature originally written and published by Al Shasia. KrASIA is authorized to translate, adapt, and publish its contents.

Middle East weekly: Baidu’s Apollo Go operates driverless fleet at WGS, QIA commits USD 2 billion to startups, Qatar signs agreement with TikTok, and more

Qatar

… million, compared to Thailand at USD 1.25 billion and Vietnam at USD 1.66 billion. Even tiny Singapore has a market size of about USD 290 million.
The contrast between the hype around the sector and its relatively slow growth has been reflected in the share price of Fore Kopi Indonesia, Fore's owner and the only listed company among the coffee chains. After listing on the Indonesia Stock Exchange last April at IDR 188 (USD 0.008) per share, it jumped 34% on its first day of trading and hit IDR 800 (USD 0.035) in June. But it has since fallen back to about IDR 530 (USD 0.023).
Hence, the search beyond Indonesia for new markets.
Kenangan has been in the vanguard of the overseas expansion. The chain debuted in India and Australia in 2025, adding to its presence in Singapore, Malaysia, and the Philippines. Tomoro has outlets in Singapore, China, and the Philippines, while Fore is in Singapore.
<figure id="attachment_159966" aria-describedby="caption-attachment-159966" style="width: 1920px" class="wp-caption alignnone"><img loading="lazy" decoding="async" class="size-full wp-image-159966" src="https://console.kr-asia.com/wp-content/uploads/2026/02/Fore-Coffee-Singapore-Bugis-Junction.jpg" alt="Photo shows Fore Coffee’s first outlet in Singapore, located at Bugis Junction in the city-state’s central district." width="1920" height="1080" srcset="https://console.kr-asia.com/wp-content/uploads/2026/02/Fore-Coffee-Singapore-Bugis-Junction.jpg 1920w, https://console.kr-asia.com/wp-content/uploads/2026/02/Fore-Coffee-Singapore-Bugis-Junction-720x405.jpg 720w, https://console.kr-asia.com/wp-content/uploads/2026/02/Fore-Coffee-Singapore-Bugis-Junction-1024x576.jpg 1024w, https://console.kr-asia.com/wp-content/uploads/2026/02/Fore-Coffee-Singapore-Bugis-Junction-768x432.jpg 768w, https://console.kr-asia.com/wp-content/uploads/2026/02/Fore-Coffee-Singapore-Bugis-Junction-1536x864.jpg 1536w, https://console.kr-asia.com/wp-content/uploads/2026/02/Fore-Coffee-Singapore-Bugis-Junction-480x270.jpg 480w" sizes="(max-width: 1920px) 100vw, 1920px" /><figcaption id="caption-attachment-159966" class="wp-caption-text">Fore Coffee’s first outlet in Singapore, located at Bugis Junction in the city-state’s central district. Photo courtesy of Fore Coffee.</figcaption></figure>
As they expand outside Indonesia, brands have to deal with different consumer preferences. "Indonesians like sweet, milk-based iced coffee at affordable price points, while Singapore leans towards espresso and specialty coffee, alongside strong, sweet traditional 'kopitiam' coffee," said Behera, referring to traditional coffee shops in Malaysia and Singapore. "Balancing a recognizable signature with local preferences on dairy, sweetness, and price is a key challenge."
Indonesian chains also have to contend with global behemoths like Starbucks and China's <a href="/company/luckin-coffee">Luckin Coffee</a>, as well as country-specific brands, each with their own unique selling points. Luckin Coffee's app-driven kiosk-based operating model has let it keep prices low as it sweeps across Singapore, while local players like <a title="" href="https://asia.nikkei.com/business/food-beverage/malaysian-cafe-chain-zus-coffee-expands-reach-with-canned-drinks" target="_blank" rel="noopener">Zus Coffee</a> and Cafe Amazon are entrenched in Malaysia and Thailand, respectively, with significant store footprint and prices that are sometimes cheaper.
Kenangan Coffee's Tirtanata is not too concerned. "We have grown quite quickly across countries, not because we are the cheapest; often local competitors are cheaper than us. However, we have localized the offering and used technology to increase sales per store," he said, referring to the brand's app that reels in customers with loyalty points and promotions, and modern management systems. Even in mature coffee markets like Australia, he said there is room for a "playful" brand like Kenangan, catering to Asian tastes.
Operating models also need to be tweaked. In Indonesia, most chains are mall-centric and adopt the grab-and-go store format, but Tirtanata points out that in Malaysia, stores are generally larger at 55–75 square meters, necessitating sit-down experiences. "As we branch out into shophouses, it is counterproductive to be grab-and-go," he said, "so we need to adapt because the inventory of locations available does not always support an existing business model."
"Indonesian coffee brands might find it difficult to be profitable, given operational inefficiency and supply chain management as they expand overseas," cautioned Lim at Euromonitor, referring to the challenge of the high cost of rapid expansion and consistently delivering the same product in multiple foreign markets. Fore Coffee is profitable and Kenangan has said that it is profitable in Indonesia and at the group level, but Tirtanata agreed that any new international market is not usually profitable for the first two to three years.
Despite the slow domestic growth, brands believe the Indonesian cafe market is far from saturated. "We aim to launch 60–70 new Fore Coffee stores every year, with the main focus still on Indonesia, especially in tier-two and tier-three cities," Rachmattulah said.
Tirtanata has similar plans. "We will still be Indonesia-centric since our sales per store are growing well and we have room for adding new stores. We will open about 70 stores in Malaysia next year, but also 300-odd stores in Indonesia." He added that while the target is to open 3000 stores by 2028, he expects 70–75% of the group's revenues to come from Indonesia even then.
Along with new stores, brands are also expanding their offerings. Fore, for example, has launched Fore Donuts, a donut chain positioned as complementing their core coffee business.
Despite the slow domestic growth, Lim is bullish on Indonesian coffee chains' prospects. "Consumer reception has been strong due to the unique local taste and wide variety of coffee beans," he said. "The brands are launching continuous product innovation to target young consumers, and it will drive their popularity in local and international markets."
This article first appeared on <a href="https://asia.nikkei.com/spotlight/asean-money/indonesian-coffee-chains-brew-stronger-overseas-ambitions">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>.

Indonesian coffee chains brew stronger overseas ambitions

Retail

Consumer Economy

Indonesia

Coffee

… your dataset quality scores 60 out of 100, the generated data might only reach 55.”
With better models, automakers can simulate scenarios that are rare or difficult to capture in real life, producing synthetic training videos that speed up iteration. “The efficiency gains over traditional retraining are significant,” a supplier engineer said. “Model updates could accelerate by a full generation.”
Still, these benefits remain mostly theoretical. World models are a clear improvement over older simulators but remain far from perfect.
<h3>Algorithms not yet ready, and hallucinations persist</h3>
These are still early days. One automaker engineer told 36Kr that Chinese-made world models can currently generate 30–60 seconds of video at most. Yet object consistency remains weak, and maintaining spatiotemporal coherence across multiple views is difficult.
At their core, world models are generative systems, that means they could hallucinate. “The hardest challenge is ensuring generated content stays realistic,” a supplier product manager said. “If you generate a person, how do you ensure their movement follows real-world physics?”
When a simulator produces physically impossible scenes, downstream models can learn false correlations, degrading vehicle performance. For instance, if generated vehicles move sideways, the model might assume a car can instantly switch lanes, prompting real-world overreactions.
If a simulator cannot model causal relationships, such as how rain affects braking or how glare distorts sensor readings, its edge cases become fictional, and optimization efforts are wasted.
Some see the main constraint in data and computing power. But Xia Zhongpu, former head of Li Auto’s end-to-end driving development, sides with Yann LeCun: the real limitation is algorithmic. In vision, self-supervised learning has yet to mature as smoothly as in language.
Language models scale rapidly because text carries dense semantic meaning; each word encodes information. Visual data is far sparser. For driving, only a small fraction of pixels influence decisions. “Today’s algorithms can’t yet extract enough decision-relevant information from images,” Xia said. “A picture has millions of pixels, but maybe a few dozen matter for driving.”
Without breakthroughs in world model algorithms, debates about data sufficiency or compute capacity are secondary. Because the underlying technology remains experimental, automakers’ investments are still exploratory. Even some executives admit uncertainty about the next step.
If world models eventually become powerful enough, and hardware catches up, they could, in theory, run directly on vehicles. For now, though, Chinese automakers treat them mainly as simulation tools. Their real-time decision-making capabilities remain limited.
That explains the current disconnect: everyone talks about world models, but users notice little difference. Most systems still live in the cloud, serving as training supervisors rather than on-road decision-makers.
“Deploying a world model on the end side is the hardest part,” Xia said.
So far, no company has applied a world model directly to vehicles. Xia added that if large models can eventually capture the physical world, predict its evolution, and influence it beneficially, not only autonomous driving but robotics as a whole could be transformed.
For now, however, the world model remains more teacher than driver, waiting in the cloud for its turn behind the wheel.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3668863813559170">article</a> was written by Xiao Man for 36Kr.

What exactly is a “world model” in smart driving?

Autonomous Vehicles

Singapore's economy grew 5% in 2025, data released on February 10 showed, beating an earlier estimate of 4.8% growth, driven by a manufacturing boost from global demand for <a href="/industry/ai">artificial intelligence</a>, even as the city-state prepares for challenges posed by potential disruption from the technology.
Last year's result fell below the 5.3% expansion recorded in 2024. On February 10, officials upgraded the growth forecast this year to a range of 2–4%, up from 1–3% previously.
The Southeast Asian trading hub's gross domestic product expanded 6.9% in the October-December quarter versus a year earlier, higher than the <a href="https://asia.nikkei.com/economy/singapore-economy-expands-4.8-in-2025-outperforming-forecasts" target="_blank" rel="noopener">5.7% estimate</a> released last month, according to government data.
"The global economy has outperformed expectations, with most major economies turning in stronger-than-expected growth in the fourth quarter of 2025," the Ministry of Trade and Industry said in a statement on February 10. "Global trade activity remained resilient despite the US tariffs, likely reflecting effective US tariff rates that were lower than the announced headline rates."
During the three months through December, the export-driven manufacturing sector expanded 18.8% compared with a year earlier, while output in the more domestically oriented services sector grew 4.8%.
Electronics exports jumped 24.9% on the year in December, UOB Global Economics and Markets Research pointed out in a January note, owing to "robust export growth" in telecommunications equipment and integrated circuits.
Singapore has received a trade bump from electronics and non-electronics shipments, amid rising demand for AI hardware overseas. But the city-state is also bracing for disruption from the global AI wave. Authorities are keen to position the country as a regional leader in embracing the tech, while also aware that it could displace many jobs domestically as companies increasingly lean on software to drive business rather than workers.
Singapore has earmarked additional investment of more than SGD 1 billion (or about USD 790 million) over the five years to 2030, under a national AI R&amp;D plan, authorities said last month. The government plans to strengthen the city-state's position as an AI research hub by scaling up research efforts and taking on challenges within the field. The funds will be used to establish research centers to advance the field in priority areas, such as emerging AI methodologies and general-purpose AI, and to develop a pipeline of talent by partnering with researchers and institutions.
"We will strengthen our talent base through nurturing AI research expertise at all levels," the country's minister for digital development and information, Josephine Teo, said in January. "We will enhance scholarships and research opportunities for our students, so they will be well placed for competitive PhD, postdocs, and faculty openings in top institutions."
"Semiconductor manufacturers set up greenfield plants and expanded existing facilities to serve strong global demand for AI-related chip, server, and server-related products, with these investments having positive spillover effects on the precision engineering sector," Singapore's Economic Development Board, an agency tasked with drawing investments into the city-state, said on February 10.
Of the SGD 14.2 billion (or about USD 11 billion) in fixed-asset investment commitments that the city-state received in 2025, about 85% came from manufacturing-related projects, the board said in an update.
At a briefing on February 10, Augustin Lee, permanent secretary at the trade ministry, told reporters that "stronger-than-expected growth momentum" seen in the last quarter of 2025 is projected to carry into this year.
"Apart from the AI investment boom, which is expected to be sustained in 2026, expansionary fiscal policies in the advanced economies, as well as accommodative global financial conditions, should also support global growth," he said. "A stronger-than-projected upswing in the AI investment cycle could provide a greater boost to electronics demand and drive further gains in equity markets."
This article first appeared on <a href="https://asia.nikkei.com/economy/singapore-beats-estimates-with-5-gdp-growth-in-2025-amid-ai-challenges">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>.

Singapore beats estimates with 5% GDP growth in 2025 amid AI challenges

In embodied intelligence, few points draw as much agreement as the importance of data.
Large language models such as OpenAI’s GPT were trained on vast text corpora, and the so-called scaling law, the idea that model performance improves predictably with more data and compute, has become a working assumption across much of <a href="/industry/ai">artificial intelligence</a>. In the physical world, however, there is no equivalent reservoir. Whether human or robot, the volume of real-world interaction data remains far below what would be required to recreate a GPT-like inflection point for embodied systems.
That gap raises practical questions. How is the data collected? At what scale? And how is quality maintained? For companies building embodied intelligence, the answers increasingly shape their prospects.
Lumos, a <a href="/industry/robotics">robotics</a> startup founded in September 2024, has chosen to focus on the first link in that chain: data collection.
The company has launched what it describes as a backpack-mounted universal manipulation interface device called FastUMI Pro. It plans to deploy 10,000 units in 2026 across six real-world environments, industrial sites, homes, hotels, restaurants, shopping malls, and offices, to conduct systematic data collection.
To understand the strategy, it helps to clarify what UMI is. Proposed jointly by researchers at Stanford University, Columbia University, and the Toyota Research Institute, UMI is a low-cost data collection and learning framework. Unlike teleoperation-based methods, which typically bind data to a specific robot’s hardware, UMI decouples the acquisition system from the robot itself. In principle, that allows the resulting data to generalize across different robot morphologies rather than remaining tied to a single machine.
At a media briefing earlier this year, Lumos founder and CEO Yu Chao compared UMI’s efficiency and cost with teleoperation.
“For a task like folding clothes, teleoperation data collection takes about 50 seconds and costs RMB 3–5 (USD 0.42–0.70),” Yu said. “Using FastUMI Pro, it takes just ten seconds, and costs less than RMB 0.6 (USD 0.08). That significantly improves efficiency while lowering costs.”
Yu previously led embodied robotics at <a href="/company/dreame">Dreame</a>, where he oversaw the development and mass production of Xiaomi’s CyberDog, delivering more than 1,000 units, according to the company. Co-CTO Ding Yan, who worked on UMI in its early research phase, was among the first in China to push the framework toward industrial applications.
In 2025, Lumos built a dedicated data collection center and said it achieved an annualized production capacity of 100,000 hours of data. Yu estimates that by 2026, leading embodied AI models will require at least one million hours of training data.
Lumos’ stated goal for 2026 is to achieve an annual UMI data production capacity of one million hours. Reaching that scale would require moving beyond centralized facilities toward distributed collection in everyday settings.
<blockquote>“Robot training data should not be this expensive or scarce. Humans generate data constantly while working in the physical world. It’s everywhere. It just hasn’t been properly collected.”</blockquote>
The FastUMI Pro is designed around that premise. Worn as a backpack, it functions as a portable, standardized workstation that converts real-world operations into structured training data. Historically, embodied data collection has relied on laboratories or tightly controlled environments. That approach often produces repetitive datasets, with robots performing a narrow range of actions in fixed contexts, limiting a model’s ability to generalize.
By miniaturizing its toolkit, Lumos is attempting to lower the barrier to capturing more diverse operational data.
The company plans to target six broad scenarios, industrial, residential, hospitality, food and beverage, retail, and office environments, subdivided into 30 task categories. It describes the objective as building a structured, multidimensional operational dataset.
At the heart of Lumos’ system is what it calls an integrated loop connecting collection, training, and deployment. The new initiative relies on that closed cycle.
Using FastUMI Pro, Lumos said its dual-arm robot Mos completed a factory quality inspection workflow, including data collection, policy training, and model inference, within five hours. After on-site deployment in Hefei, the company said the system completed the same sequence in seven hours in a real-world setting.
In parallel, Lumos has introduced what it calls a “data supermarket.” It has standardized portions of its datasets into products that customers can purchase through its website.
For now, Lumos’ strategy centers on data infrastructure rather than model architecture.
“I come from a model training background,” Ding said in interviews with several media outlets earlier this year. “I used to focus on training models. But we ran into a major issue. To train a really strong model, you need a solid data pipeline including data production, evaluation, and filtering. Building that pipeline takes time.”
After evaluating its priorities, the company decided to focus first on data.
“In the end, competition in model architecture only goes so far,” Ding said. “What truly differentiates models is the quality of their data.”
Whether that view proves durable as embodied AI evolves remains uncertain. What is clear is that the upper bound of embodied intelligence is constrained by the scale and diversity of real-world operational data. If standardized datasets can be sourced as readily as hardware components, the threshold for training industry-grade models may decline.
Lumos is betting that shifting from centralized labs to thousands of backpack-mounted devices will expand that supply. If operational data becomes more abundant and standardized, embodied systems may begin to generalize beyond controlled demonstrations and into routine, repeatable work.
KrASIA features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3680210722254473">article</a> was written by Qiu Xiaofen for 36Kr.

This company has built a backpack-style system for robotics data collection

Robotics

Embodied Intelligence