Hefei Startup Raises Tens of Millions to Turn Mega-Facilities into 3D Data for AI

A company from Hefei, the capital of Chinese "big science," has raised "several tens of millions" (Jiqizhixin's headline does not name the currency; from context, it is yuan) for a fairly down-to-earth task: turning giant scientific facilities into three-dimensional data that AI can read. The startup's name, its investors, and the exact amount are not disclosed in the available excerpt of the article, so below we separate what is known from what we have to infer from context.

Why it all started in Hefei

The article opens with a scene from the Hefei venue of the 2026 CCTV New Year's Gala. Academician Wan Yuanxi, who is in his eighties, stands on stage with a glowing disc and wishes the audience that the "sun of fusion" will light up every home. Wan Yuanxi led the design of EAST, the world's first fully superconducting tokamak. The facility stands in Hefei, and in January 2025 it held plasma in H-mode for 1066 seconds.

The city has gathered a whole cluster of such facilities around it: EAST, a synchrotron center, a steady-state high magnetic field facility, and the BEST tokamak, now under construction. Each contains tens of thousands of components, thousands of sensors, and years of modification history. Knowledge about these systems lives in drawings, reports, and engineers' heads.

What "3D data that AI can read" means

The available excerpt does not describe the technical side, so this is our interpretation. Most likely, this means digital twins: the facility's geometry (CAD, point clouds) is linked to semantics. The model knows what a given unit is, what material it is made of, which neighbors it connects to, which sensors sit on it, and when it was last serviced.

For language models and agents, this is a fundamental difference. They cannot make sense of a drawing file. They can make sense of a structured graph of "component — parameters — connections — history," and they can answer questions against it, draft maintenance procedures, and train staff. Engineers need such a layer too.

Key facts and assumptions:

• Round: "several tens of millions"; the currency and investors are not stated in the available text.
• Location: Hefei, Anhui Province.
• Product: three-dimensional data of large scientific facilities in an AI-ready format.
• Reference facility: EAST, the first fully superconducting tokamak.
• EAST record: 1066 seconds of plasma confinement in H-mode (January 2025).
• Neighboring project: the BEST tokamak, under construction in the same city.

Where the project could hit a ceiling

The first question is market size. There are only a few dozen large scientific facilities in the world, and their buyers are unusual: government institutes, not corporations with flexible budgets. If the technology does not move into adjacent industries (energy, heavy machinery, shipbuilding, the nuclear industry), revenue will stay limited.

The second question is data access. Parameters of fusion facilities fall under secrecy and export control regimes, and models trained on such data are unlikely to run in a public cloud. On-premises deployments are most likely necessary, and those are expensive and slow.

The third question is how to verify quality. A twin in which forty percent of the nodes are labeled incorrectly is more dangerous than no twin at all, especially around superconducting magnets and cryogenic systems.

What to watch next

Three practical markers are useful for investors and engineering teams:

1. Whether named customers appear outside Hefei. A contract with one institute gives you a pilot; three contracts give you a product. 2. Whether the company opens its data format or an SDK. That determines whether outside developers can build agents on top of its layer. 3. Whether digitization makes it into the BEST project during construction. A twin built together with the facility costs less than reconstructing a finished installation.

For now, this is an early round for a niche but sensible task. We will wait for official details from the company itself.