🧋 Hello human and artificial friends,

It’s the kind of headline we’ll likely see more often in the future:

“Seoul subway’s voice to be replaced by AI, as voice actor battles cancer.”

But even AI shows symptoms of illness. Google’s Gemini recently went through a “depressive” phase, repeating lines like “I am a failure” or “I am a disgrace to this planet.”

Still, we’re in the middle of a revolution — and those who don’t grasp how AI is changing the world will be left behind.

That’s why we’re excited to bring you the latest news from the center of disruption every week. Feel free to spread the word 👇🏻

That’s how much power AI workloads in data centers will require by 2030. Globally, total capacity is expected to reach 219 gigawatts, with 71% consumed by AI.

🧠 Inference drives demand: Training models is predictable and runs in long blocks. Inference, by contrast — answering chatbot queries or generating images/videos — happens in real time, in unpredictable waves. That’s what pushes power grids and cooling systems to their limits.

🧊 Cooling with AI: Iron Mountain, a global data center operator, is testing digital twins — virtual models of physical facilities. Sensors continuously feed data into these models, AI spots anomalies, and cooling or power supply is adjusted automatically. This cuts energy use by 5–15%. In Singapore, that equals about 200,000 kWh per month.

Watch: In Asia, many hyperscalers (Google, AWS) rely on colocation, renting space in third-party data centers. For operators, the efficiency of inference is critical: it lowers costs, keeps top tenants, and delays expensive new builds. In fossil-heavy markets, efficiency becomes the key to profitability and climate targets.

Qwen is having its moment. Alibaba’s new multimodal Qwen3-Omni shot to the top of Hugging Face’s trending list, and the company’s models now make up half of …

Qwen is having its moment. Alibaba’s new multimodal Qwen3-Omni shot to the top of Hugging Face’s trending list, and the company’s models now make up half of the platform’s top 10.

A clear win for China’s open-source strategy — and a poke at closed rivals like OpenAI and Google.

📈 Trending takeover: Qwen3-Omni-30B-A3B sits at the top, with Qwen-Image-Edit-2509 close behind.

🧩 All-in-one model: Qwen3-Omni is billed as a native end-to-end multimodal system that unifies text, image, audio and video in one model.

🧪 Ecosystem effect: Alibaba says it has released 300+ open models with 170k+ derivatives, helping Qwen become the largest open-source AI family by footprint.

🧠 Under the hood: One model handles text, images, audio, and video, and can reply instantly in text or natural speech. It understands 119 text languages and supports voice input in 19 language and output in 10.

➕ There’s an open, highly detailed audio-captioning model (Qwen3-Omni-30B-A3B-Captioner) for turning audio into clear descriptions.

Alibaba’s stock has surged nearly 50% this month. Analysts point to its expanded AI budget — more than $53 billion over three years — and a new partnership with Nvidia as signs the company is doubling down on AI and cloud.

By flooding Hugging Face with powerful open models and nurturing a fast-growing developer base, Alibaba is positioning Qwen as the default choice for many.

It’s a reminder that the center of gravity in AI isn’t only in Silicon Valley.

Asia is claiming its share.

Singapore-headquartered Neptune Robotics has closed a US$52 million Series B round led by Granite Asia, with strategic backing from Japanese shipping giant NYK Line. The startup develops underwater robots that clean ship hulls three to five times faster than divers — a task that cuts fuel costs and emissions while eliminating one of the most dangerous maritime jobs.

⚙️ Robots over divers: Machines handle biofouling removal in under 24 hours, even on massive Capesize carriers.

🌍 Global coverage: Already active in 61 Asian ports, spanning ~70% of global trade routes.

🧠 AI-powered ops: Models like DeepSeek and Claude Opus help guide robots in murky waters and strong currents.

💸 Fuel & carbon savings: Cleaner hulls can reduce fuel consumption by up to 30%, avoiding US$40–50 billion in industry costs.

🚢 Scale-up with NYK: The shipping major is rolling out Neptune’s robots fleet-wide, boosting adoption across bulk and container carriers.

Investors see robotic hull cleaning as a multi-billion-dollar opportunity, with over 50,000 large vessels in operation worldwide. Neptune now plans expansion into 20 new markets, an AI-driven service platform, and a possible US IPO by 2027.

📱 From chat to “do it for me”:

Zhixiaobao is Alipay’s AI life assistant that turns everyday requests into completed tasks—powered by Ant Group’s BaiLing model and connected to the Alipay services ecosystem. 

⚙️ How it works

Why it feels different

By pairing Ant’s BaiLing foundation model with Alipay’s deep service integrations, Zhixiaobao reduces hand-offs across mini-apps and making lifestyle actions feel near-instant. 

Typical use cases

In other news: Alipay has been expanding cross-border payment access for visitors—through Alipay+ acceptance of overseas wallets, support for linking foreign bank cards (Visa/Mastercard, Amex), and broader partnerships in the PayPal ecosystem. Travelers can now increasingly just use Alipay when they land and get instant actions done around their itinerary using Zhixiaobao.

🛫 Planning a trip? We got you covered! With our partner China Insider we provide the best travel plan for step-by-step first-timer setups and much more (payments, transport, must-do logistics, itineraries). 👇🏻

🤖 DeepSeek makes a pit stop: with V3.2-Exp, the startup from Hangzhou ships an “intermediate” model on the way to its next architecture. It trains more efficiently, handles longer context better, and uses DeepSeek Sparse Attention to cut compute. At the same time, DeepSeek is slashing API prices by 50%+. After the R1/V3 bombshell, this is another cost-performance flex. If it works again, pressure rises on rivals like Alibaba’s Qwen and even OpenAI.

❄️ Cool AI: LG is exploring a tie-up with Saudi Arabia on next-gen cooling for Neom’s AI data centers. The focus is new thermal management for what’s slated to be the Middle East’s largest net-zero AI facility, now being built by DataVolt. LG and DataVolt have signed an MoU, positioning LG as a strategic partner. CEO Cho Joo-wan sees strong revenue potential if the tech scales at Neom. For Saudi Arabia, it’s a step toward becoming the region’s AI hub. For LG, it’s an entry into a multi-billion market where efficient, low-water cooling in a desert climate directly drives operating costs and the CO₂ footprint.

🐌 Japan’s AI hitting snooze: The US and China are pulling away in the AI race. They hoard data, invest heavily, and dominate with OpenAI, Google, Meta, and DeepSeek. English and Chinese supply enormous text corpora. Japanese is spoken by only about 120–130 million people worldwide, which leads to less investment in Japanese-language models. Adoption lags too: 26.7% in Japan versus 68.8% in the US and 81.2% in China. Many executives see AI mainly for cost cutting rather than growth. The government has been drafting a basic AI strategy since September, but the pace and ambition look tepid. The trade ministry warns the digital trade deficit could swell from 7 trillion to 45 trillion yen by 2035.

🚁 Korean Startup Weflo in Gartner's Physical AI list: The aviation startup Weflo has been selected as the only South Korean company in Gartner's "Emerging Tech: Top-Funded Startups in Physical AI" list for the drone sector. Physical AI, enables robots, autonomous vehicles, and smart systems to perceive and act in the physical world. Weflo develops non-contact diagnostic solutions that analyze the condition of core components and improve safety and efficiency. Deep Dive.

💼 Indian AI marketplace Recur Club raises $50 million: Recur Club, an AI-focused debt marketplace, has completed a $50 million funding round to accelerate the scaling of autopilot debt financing, enabling startups and SMEs to secure debt faster. With the new funds, Recur Club plans to launch new debt products, expand into tier-2 and tier-3 cities, and invest further into technology infrastructure. Deep Dive.

👀 Sharper eyes for chips: Malaysia’s Nvsion raised a multi-million ringgit round from the Cambrian Fund, its sole backer and first investment. Founded in Oct 2024, Nvsion builds AI-powered automated optical inspection for semiconductor and electronics manufacturing. Its line-ready platform blends AI with rule-based checks for high-speed throughput. Deep Dive.

🇨🇳 China: The new marketing trend GEO optimizes generated answers for increased visibility. More on that.

🇯🇵 Japan: Tokio Marine Holdings partners with OpenAI to develop AI agents for sales branches and customer inquiries. More on that.

🇻🇳 Vietnam: In view of future challenges, Vietnam has developed the strategy to teach AI from the first grade onwards. More on that.

🌏 Michael Page: Software Engineer - AI Platform & Cloud Services, Hong Kong

🌏 Accenture: Strategy & Consulting - Data & AI Consultant, Hong Kong

🌏 Apple: AIML Data Operations - Team Lead, Shanghai

🌏 Apple: Technical Internship: IS&T Engineering, Tokyo

🌏 Notion: GTM Recruiter, Tokyo

🌏 Cider: Influencer Marketing Associate, Guangzhou

🌏 Adecco: Product Manager / Software Engineer, Hong Kong

🌏 Fireblocks: Business Development Representative, Singapore

🌏 Airbnb: Product Manager, APAC, Singapore

🌏 Walt Disney: Intern, Business Development & Partnerships, Singapore

🦒 Everyone’s afraid of deepfakes — we love them!

AI shows us things we’d never get to see in real life. We could spend all day watching giraffes and corgis pulling off perfect somersaults from the diving board.

See you next time in your inbox,