China's Tianxu raises tens of millions for energy weather AI, puts model in orbit
What's the deal? Tianxu Intelligent TechnologyDealroom has a profile for this one. Try Dealroom →, a Chinese weather AI startup, has raised tens of millions of yuan in a Pre-A+ round. Houxue CapitalDealroom has a profile for this one. Try Dealroom → was the sole investor, with Yirong CapitalDealroom has a profile for this one. Try Dealroom → acting as long-term financial adviser. The money will fund satellite constellation buildout, model iteration, hiring, customer expansion, and early overseas work.
What does it do? Founded in October 2024, Tianxu builds vertical weather models for the energy sector. Founder Liu Li, a meteorology PhD from the University of Chinese Academy of Sciences, has spent nearly a decade on the intersection of weather, AI, and energy. Its model runs under 10B parameters — not a large language model, but a domain-specific one built on a Transformer base.
Why now? New energy generation is weather-dependent, with wind speed and irradiance directly setting output. Renewables now exceed 50% of China's installed capacity, and power-market reforms are accelerating demand for accurate forecasting. Tianxu's customers already span generators, grid operators, and power traders, including State GridDealroom has a profile for this one. Try Dealroom → and China Southern Power GridDealroom has a profile for this one. Try Dealroom →.
The product: In October 2025, Tianxu and the China Meteorological AdministrationDealroom has a profile for this one. Try Dealroom →'s public service centre jointly released CMA-WSP-AI, which it calls the country's first energy weather large model. Unlike standard models that output temperature, humidity, and rainfall, Tianxu's version simulates how weather shifts feed through to renewable output, grid load, and spot electricity prices. It says its 1km wind-speed forecasts beat EC-HRES and AIFS on key accuracy metrics.
What's the endgame? Liu argues algorithms eventually get matched, so the lasting moat is owning the data source. Tianxu launched its "Ningkong" constellation plan and, on July 24, 2026, put a weather model into orbit — among the first Chinese startups to do so. The long-term plan is 18 satellites delivering global 1km, hourly observations of wind, solar, temperature, pressure, and humidity.
By the numbers: In an extreme-weather AI competition covering hail, thunderstorm winds, thunderstorms, and short-duration heavy rain, Tianxu placed first, improving threat scores over traditional baselines by 39.6% to 351%. More than 80% of its staff hold master's degrees or higher.
Tianxu runs a three-layer model: selling data, SaaS subscriptions for trading and forecasting tools, and on-premises deployment inside client networks. "We put the model and platform inside the client's internal network, so data and model run in a closed loop," Liu said (translated from Chinese), citing tightening energy-security rules.
The signal: As China builds out its new power system, weather forecasting is shifting from a supporting tool to energy-security infrastructure. Houxue called it "key digital new infrastructure." Tianxu is betting that combining a vertical model with its own satellites turns a crowded forecasting market into a data-ownership play — one it hopes to export along Belt and Road routes.
Read more: 36kr.com
Image credit: Generated with Gemini
