Tianxu Intelligent raises funding to put its weather AI model in orbit
Tianxu Intelligent Technology (Jiaxing) has closed a Pre-A+ round of tens of millions of yuan, solely funded by Houxue Capital, with proceeds going mainly to its Ningkong satellite constellation, energy-weather model updates, hiring and overseas research. The company says its self-developed vertical weather model, under 10 billion parameters, uses a "Lorentz framework" that adapts Transformer attention for wind-speed and irradiance forecasting, and claims its 1-km wind predictions beat EC-HRES and AIFS on RMSE and correlation metrics. It adds that its weather model reached orbit aboard a satellite on July 24, 2026 and passed daily validation, with a long-term plan for 18 satellites and two pilot satellites in 2027.
The pitch shows how energy-weather AI in China is being bundled with proprietary satellite data and on-premise delivery to serve power forecasting and electricity trading.
一家成立不到两年的气象AI公司，把融资和卫星同时揣进了口袋。天序智能科技（嘉兴）有限公司近期完成数千万元Pre-A+轮融资，由厚雪资本独家投资，资金主要砸向凝空星座建设、能源气象大模型迭代、团队扩张与海外预研。这家2024年10月才诞生的公司，是国内极少数同时握着垂类气象大模型、自主卫星观测网和行业底座化交付能力的AI for Science玩家，创始人刘栗是中国科学院大学气象学博士，近十年都泡在气象与能源交叉的地带。
在刘栗的判断里，新能源天生靠天吃饭，风速和辐照度直接决定发电量，而国内新能源装机占比已经突破50%，功率预测催生百亿级刚需，电力现货交易又打开千亿级增量，这是条正处爆发前夜的黄金赛道。天序的打法，是把大模型从云端一直推到太空。
它家的气象大模型并不是大语言模型，而是一个参数规模在100亿以内的垂直领域模型，从第一行代码自研，在主流架构上做了能源向改造，并抛出了一套名为洛伦兹框架的设计——名字致敬混沌理论提出者、气象学家洛伦兹。核心思路是以Transformer为底座，针对能源气象里上下游空间关联、时间飘移等现象改造传统注意力机制，从一开始就把目标锁死在风速和辐照度的预报上，而不是泛泛地报温度湿度降水。
#Tianxu Intelligent#Houxue Capital#Ningkong Constellation#CMA-WSP-AI#China Meteorological Administration#Lorentz framework#State Grid#China Southern Power Grid#EC-HRES#AIFS