🧋 Hello human and artificial friends,

According to a new ranking, Switzerland has once again been crowned the world’s most innovative nation — for the 15th year in a row. (Haha.)

China, South Korea, and Singapore barely made it into the top 10.
And they say only AI hallucinates...

P.S. We’re looking for interns! If you like our writing style and think you can match it, reply to this email with a short application. Bonus points if you’re as innovative as the Swiss.

Asia is where AI moves fastest. Every day, new tools drop. Most vanish overnight, but a few redefine how people live and work. We find those gems so you can watch, learn, and build your own.

Today’s idea

Asia is where AI moves fastest. Every day, new tools drop. Most vanish overnight, but a few redefine how people live and work. We find those gems so you can watch, learn, and build your own.

Today’s idea

What if the next big thing in imaging isn’t an app — but a device?
A new Chinese startup, Guangqi Zhijing, is building an AI-powered camera that doesn’t just capture a moment; it creates the image you wish you had taken.

The company recently raised 27 million USD from top Chinese funds, betting that the next wave of consumer AI will live in your hands, not your cloud.

Behind the lens sits a vertical AI model trained specifically for photography trends. It studies millions of real social posts to learn what people currently call “beautiful.” The result is an instant, trend-aware image that looks like it already belongs on your feed—collective intelligence at its best.

The camera removes cognitive friction. You don’t need to know aperture, color grading, or ISO. You take the shot, and the device handles the rest — from framing hints to instant stylization.

✅ Hardware with purpose: After years of “AI-in-the-app,” this is a rare return to tangible tech. By merging a dedicated camera with on-device AI, the company avoids smartphone trade-offs like heat, battery drain, or slow cloud processing.

✅ Beauty as data, not definition: Instead of hardcoding aesthetics, the camera follows what people collectively like. Its model updates based on what is trending on social media, not on what engineers decide is good. In a world where “pretty” changes monthly, that’s a durable edge.

✅ Solving the mental load (especially for IG husbands)
Users don’t need technical knowledge or endless retakes. The product removes the thinking barrier that stops people from creating.

✅ Counterintuitive timing
Launching hardware in a phone-saturated market seems crazy. That’s the point. As digital cameras decline, a stylish AI device could capture users’ craving for something different and shareable.

👉🏻 Think bigger than filters. There’s room for AI to reimagine all creative tools — from drawing tablets to audio recorders — by focusing on the thinking layer rather than just the interface. That’s where the next wave of hardware-born AI will live.

🧠 Chinese start-up MiniMax revolutionizes AI with M2 model: Alibaba-backed AI start-up MiniMax has released the M2 foundation model, which ranks as the world's leading open model, surpassing Google DeepMind's Gemini 2.5 Pro. With 200 billion parameters, the M2 model adopts the Mixture-of-Experts architecture, significantly enhancing the efficiency of Chinese AI models as it only activates a small subset of parameters for each 'forward pass' – a trend among Chinese developers to improve efficiency.

🎌 Local data as the trump card: SoftBank is co-creating one of the largest native datasets with NTT & co. ~30,000 speakers and 9,000+ hours spanning 2,600 service scenarios across 10 industries plus everyday conversations. The goal is models that reflect real Japanese interactions and that run in closed environments like finance, healthcare, and the public sector: SoftBank’s “Sarashina mini” distills its in-house large model to ~10% of the size, while NTT’s “Tsuzumi 2” runs on a single GPU and can be adapted with minimal extra data.

⚡ “Electrons are the new oil”: OpenAI warns the U.S. must massively build new power capacity to stay ahead of China in AI. In a paper to the White House, OpenAI calls for 100 gigawatts of additional energy per year, points to China adding 429 GW in 2024 versus 51 GW in the U.S., and warns of an “electron gap.” In parallel, OpenAI cites its own Stargate data centers in Texas, New Mexico, Ohio, and Wisconsin, which together will require multiple gigawatts. The goal is to provide enough affordable electricity for rising AI workloads and to safeguard U.S. tech leadership.

AI’s energy footprint is big (data centers ~1.5% of global electricity and rising), but a clear trend is emerging: using the same compute to optimize real-world systems so they waste less energy and emit less pollution.

“AI for efficiency”—deploying models inside buildings, grids, industry, and cities to cut demand, time usage to cleaner power, and eliminate unnecessary emissions.

This is the number of AI cameras watching the bustling streets of Hanoi starting from December this year.

🚦 Action plan: Hanoi is deploying the cameras this year as part of a plan to install more than 40,000 by 2030. The system aims to improve traffic flow, urban safety, and real-time monitoring across the city.

📸 How it works: Each camera runs on-device AI to detect violations, assess risks, and adjust traffic lights in real time. With 360° rotation and 700 m range, the system operates reliably in any weather and syncs data with central monitoring hubs. Together, they create a live digital map of Hanoi’s traffic flow.

👉 Watch: Hanoi follows a regional trend — across Asia, cities like Singapore, Seoul, and Jakarta are wiring infrastructure with intelligence.

📢 Opencut is a free, open-source video editor aiming to be a simple, privacy-first CapCut alternative, with a live early beta you can use in the browser. It runs as a web app and can also be run locally with Docker/Node via GitHub. Edge: no subscriptions or watermarks and local/on-device processing.

🤖 Genspark AI is an AI “super agent”. It performs deep research via an AI browser, generates Docs/Slides/Sheets, manages Gmail/Calendar, and places calls. Example: To reschedule your dentist appointment, it finds the initially scheduled time in your email/calendar, calls the clinic, asks for the next available slot that fits your calendar, confirms it, updates your Calendar, and sends you a short summary.

🍿Migaku turns your favorite Anime or Netflix drama into study material. The tool from Japan with an Anki-style flashcard edge can be used as a Chrome extension. Click on a word to see the meaning and pronunciation. It auto-creates a sentence flashcard with audio, which you can review and study later.

🤖 LibLib AI raises US$130m to push creative AI frontiers: The Chinese startup developed an AI image-generation platform for multi-scenario and multi-dimensional content creation. It has now secured US$130 million in new funding co-led by HongShan and CMC Capital. The platform enables interactive, generative tools for artists, marketers, and developers to produce visual content across formats and media.

🤖 Shengshu AI launches Vidu Q2: Challenging OpenAI’s SORA, the video creation model turns up to seven reference images into eight-second AI videos with synced dialogue and sound. Backed by Bohua Capital, the Beijing-based startup claims to have over 30 million users and US$20 million in annual revenue.

🗣️ Wiz.AI raises millions to give chatbots a human voice: The Singapore-based startup has raised US$12.1 million in a Series B round. The startup builds realistic voice agents for call centers. The service focuses on Southeast Asian languages, enabling enterprises in banking, healthcare, and e-commerce to automate inbound and outbound calls while keeping the tone personal.

✅ Check out Deloitte’s survey about GenAI in the APAC region
✅ Keep your laptop alive with this Power Bank.*
✅ Generate a YouTube video summary.
✅ Get a glimpse of how China is already living in the future — and read this travel guide first. Enjoy 15% off with code asiabitsfriends. **
✅Drink more Vietnamese coffee.*

_{*Yes, this is an affiliate link. No, you won’t pay a cent more. But it helps support asiabits. Thanks :)
** A message from our sponsor.}

🌏 NVIDIA: 2026 Internships - Autonomous Vehicles and Robotics, China

🌏 Ekimetrics: Junior Consultant in Strategy & Data Science, China

🌏 LINE Plus: AI Product Manager, South Korea

🌏 Persly: AI Engineer (LLM, RAG), South Korea

🌏 Daangn: Data Analytics Engineer, South Korea

🌏 Sixertech: STEM Experts - Global Resources (Live), APAC

🌏 Geniee (JAPAN AI): AI Consultant, Japan

🌏 DataAnnotation: Bilingual AI Trainer, Hong Kong

🌏 Cummins: AI Intern (LLM), China

🌏 Hyphenconnect: AI - Leveraged Product Development Generalist, China

Reckless Robo Run

Maybe robots are not yet capable of outperforming humans in their natural abilities. In the race, though, that does not matter…as long as they can somehow manage to literally take down the human opponent.