China turns APEC into an AI export channel for small business

Share
China turns APEC into an AI export channel for small business

Two very different AI stories landed on the same day. In Guangzhou, China used a 21-economy ministerial meeting to launch permanent regional AI institutions aimed at small firms — the clearest move yet to make its open-weight model stack the default for developing Asia-Pacific economies. And in a smaller but telling story, a Chinese tech personality asked why no domestic carmaker ships a full-strength LLM in the cabin, touching off a public argument about why they won't.

At the 32nd APEC Small and Medium Enterprises Ministerial Meeting in Guangzhou, China announced it is establishing an Asia-Pacific AI Empowerment Cooperation Center for SMEs in Guangzhou and an Asia-Pacific SME Cooperation Zone in Shenzhen. The meeting, chaired by MIIT Minister Li Lecheng and opened by Vice Premier Zhang Guoqing, produced a chair's statement plus a separate "Guangzhou Initiative" on SME innovation and resilience. AI ran through all three of the meeting's formal themes — technological innovation, digital and intelligent empowerment, and inclusive development — and Indonesia's MSME Minister Maman Abdurrahman used the floor to pitch digital transformation for Indonesia's more than 60 million micro, small and medium enterprises. The institutions are the concrete deliverable: a standing regional body that outlives the meeting itself.

The significance is in the framing, not the ribbon-cutting. Sun Yat-sen University economist Li Bing argued at the meeting that AI is the first general-purpose technology that lets small firms rent frontier capability cheaply, and that China's open-source model strategy gives developing economies a path that is "usable and affordable" — while warning that adoption still depends on unglamorous basics like reliable power and transport. That is a soft-power argument dressed as a development argument, and it is aimed squarely at economies the US export-control regime is simultaneously pushing away from American accelerators. This is the multilateral companion to the domestic deployment push we covered last week — China's MIIT sets hard targets for an AI service-provider army — but where that notice built an integrator bench inside China, this one builds demand and goodwill outside it. Watch whether the Guangzhou center becomes a real venue for model and tooling transfer, or just another acronym: Shenzhen hosts the APEC leaders' summit in November, which gives China a hard deadline to show something.


Luo Yonghao — the entrepreneur whose grudge against in-car software went viral — publicly asked why no Chinese EV maker ships a complete, full-strength version of a top general-purpose model like Doubao, Qwen or Kimi in its voice assistant. He says he test-drove dozens of vehicles and found none with an unabridged frontier model, and argued he would happily pay extra for it, noting drivers spend more than an hour a day in their cars. It's a consumer complaint, but it landed because it names a real pattern: automakers almost universally ship trimmed, fine-tuned or scope-limited models rather than the full product.

The reasons are unglamorous and mostly legitimate. Carmakers cite compute cost, latency, connectivity in tunnels and underground garages, content safety, and automotive data rules that make routing voice data to a third-party cloud provider a compliance headache. There's a strategic layer too: hand the cabin's conversational layer to a model vendor and you hand over the user relationship, which is why the build-your-own route persists even when the result is worse. Luo's jab is that drivers are paying a build-it-ourselves tax so carmakers can keep the interface — a fair charge, and one that lands harder now that Changan is shipping its own end-to-end driving LLM after 1.45 billion kilometers of validation, as we covered in Changan ships a self-developed end-to-end driving LLM after 1.45 billion km of validation. Expect hybrid arrangements — on-device models for vehicle functions, full cloud models for conversation — rather than a clean answer to his question.

What to watch: whether the Guangzhou center signs up a single non-Chinese model provider in its first year — the fastest tell for whether this is a genuine regional channel or a domestic one with APEC branding.

Should carmakers be allowed to ship a weaker assistant to protect their data — or does the customer deserve the full model? Tell us in the comments.

Read more

AI 101 — How do AI video generators work?

AI 101 — How do AI video generators work?

An AI video generator is a model that starts with a grid of random noise and repeatedly refines it into a moving picture that matches your prompt — the same trick image generators use, stretched across time. Nothing is filmed, and nothing is stitched together from clips the model has seen. The pixels in the finished video were invented, frame by frame, all at once. Why it matters right now Kling 4.0, Kuaishou's flagship video model, went fully live on October 15 with up to 30 seconds of vid

Houmo moves compute inside memory — and A20 enters production

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings