China Telecom leads funding for embodied-AI data startup Mifeng

State capital is quietly moving into the data layer beneath China's robot boom — this time through a six-month-old startup that sells the raw material physical AI runs on.

Mifeng (Maniformer), a Shanghai startup building one-stop data infrastructure for physical AI, has closed a new round of several hundred million yuan led by China Telecom. Zhangjiang Group followed, while existing backers Sequoia China and Yuanqi Innovation over-subscribed — the latest in a string of rounds since the company was founded about six months ago. Proceeds go to scaling production of its MEgo series of "bodyless" data-collection devices, which are deployed in real factories, logistics hubs, supermarkets, hotels and homes to capture the physical interaction data robots need, and to expanding a collect–clean–evaluate pipeline it claims is ten times more efficient than manual processing.

The bet is that embodied AI is stuck on a data problem, not just a model problem. Physical interaction data carries far lower information density and far more noise than language data, so training useful robot models demands volumes that today's hand-built collection farms can't supply. Mifeng's answer is industrializing that supply: its MEgo devices capture environment perception and hand-operation data with millimeter-level trajectory accuracy, and in some scenarios each closed-loop data cycle lifts task success rates by 5–10 percentage points. The company says it already runs innovation centers in more than 20 Chinese cities and over five overseas nodes, and provides training data for Ant's LingBot-VLA 2.0 foundation model. The timing lines up with a sector that is drowning in money but short on proof: Leiphone's industry analysis this week put Chinese embodied-AI funding at ¥93.5 billion in the first half of the year — about five times the year-earlier period — while top humanoid makers have shipped only just over 10,000 units.

The notable part isn't the check size, it's who signed it. China Telecom brings cloud-network integration, compute scheduling and security compliance to a joint "cloud-network + embodied data" push aimed at humanoid robots and smart manufacturing, and Zhangjiang Group opens its park, industrial and commercial scenarios to the company. A state telecom operator effectively placing a strategic bet on the data layer — not the models, not the hardware — is a signal that the physical-AI value chain is splitting into specialized layers, and that the data layer is where state capital sees a defensible, standardized product emerging. Mifeng still has everything to prove: the round is modest by Chinese AI standards and its customers are mostly still piloting. But when the state starts funding the plumbing, the industry usually follows.

What to watch: whether China Telecom's cloud-network muscle turns into standard data products other robot makers actually buy — and how many more state giants follow it into the data layer.

State telecoms are now funding the data layer of China's robot boom — do you think data infrastructure, not models, will decide who wins embodied AI? Tell us in the comments.

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

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

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

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

Read more

AI 101 — How do AI video generators work?

Houmo moves compute inside memory — and A20 enters production

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

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