A weather AI company less than two years old has secured both funding and satellites. Tianxu Smart Technology (Jiaxing) Co., Ltd. recently completed a pre-A+ round of funding worth tens of millions of yuan, led by Houxue Capital. The funds will mainly be used for the construction of the "Ningkong Constellation", iteration of the energy meteorology large model, team expansion, and overseas research. This company, established in October 2024, is one of the few AI for Science players in China that simultaneously possesses vertical meteorological large models, an independent satellite observation network, and industry-based delivery capabilities. Its founder, Liu Li, is a Ph.D. in Meteorology from the University of Chinese Academy of Sciences, and has spent nearly ten years in the intersection of meteorology and energy.
In Liu Li's view, new energy is inherently dependent on the weather, with wind speed and irradiance directly determining power generation. With the proportion of new energy installed capacity in China exceeding 50%, power forecasting has created a demand of hundreds of billions of yuan, while the electricity spot market has opened up a trillion-yuan incremental opportunity. This is a golden track about to explode. Tianxu's approach is to push the large model all the way from the cloud to space.
Its meteorological large model is not a large language model, but a vertical domain model with a parameter scale of less than 10 billion. It was self-developed from the first line of code, and underwent energy-oriented modifications on mainstream architectures. It also introduced a design called the "Lorenz Framework," named after the pioneer of chaos theory and meteorologist Lorenz. The core idea is to use Transformer as a base, and modify traditional attention mechanisms to address spatial correlations and temporal drift phenomena in energy meteorology. It focuses specifically on predicting wind speed and irradiance, rather than general temperature, humidity, and precipitation forecasts.
Compared to traditional physical models that solve mathematical equations and require supercomputers to calculate for several hours, AI large models can predict the weather for more than 20 days within a minute, and in many cases achieve higher accuracy. Since 2025, Tianxu has collaborated with the Public Meteorological Service Center of the China Meteorological Administration to build workflows. In October of the same year, it jointly launched the country's first energy meteorological large model, CMA-WSP-AI, in Zhongshan, Guangdong. A step further is its vertical world model: while ordinary meteorological large models only output meteorological results, Tianxu can further simulate how meteorological changes affect fluctuations in new energy output, impact grid load, and drive price fluctuations in the spot market, achieving a transition from calculating weather to simulating energy system evolution. This provides simulation capabilities for power trading, large-scale base planning, and grid risk pre-examination. In actual testing, its high-precision 1 km wind speed prediction outperforms EC-HRES and AIFS in metrics such as RMSE and correlation coefficients. It won first place in extreme weather AI competitions, with TS scores improving by 39.6% to 351% compared to traditional baselines.
In terms of business, Tianxu has successfully implemented a three-layer model of data, platform, and capability building. At the bottom layer, it sells data, with clients including top power generation groups, meteorological departments, and power trading entities. The second layer is a SaaS-based power trading assistance decision-making and power forecasting platform, which brings continuous revenue. The third layer involves deploying data, models, and platforms into the client's internal network to meet the security and private requirements of major clients. As more stations are connected, the model becomes more accurate, forming a positive data flywheel. Liu Li pointed out the weaknesses of the old model: power generation sides transmit data to third parties, process it, and then return it. However, under energy security requirements, this has become unworkable, as data cannot leave the internal network. Therefore, Tianxu integrates the model and platform into the client's internal network, allowing data and models to operate in a closed loop internally while helping clients master this autonomous capability. Its clients now cover power generation, consumption, and grid sectors, including State Grid and Southern Power Grid, effectively securing access to the most critical customers in China's energy system.
What truly distinguishes this company from traditional meteorological service providers is that it has sent the large model into space. Liu Li believes that algorithm advantages will eventually be caught up, and data sources are the long-term barrier. Therefore, he initiated the "Ningkong Constellation" plan. On July 24, 2026, the meteorological large model was carried by a satellite into orbit and completed daily verification, making it one of the few startups in China to send a model into space. The constellation's long-term plan includes 18 satellites, establishing a global 1 km and 1-hour revisit observation network for wind, solar, temperature, and humidity, creating a rare path combining AI large models, own satellites, and on-board computing. The subsequent schedule is already set: in 2027, two pilot experimental satellites will be launched to verify inter-satellite transmission and on-board computing, addressing the shortcomings in wind and solar observations in desert areas. Then, the 18 satellites will be deployed to serve domestic and Belt and Road Initiative markets globally.
In addition to energy, Tianxu has extended its reach to low-level meteorology and financial meteorology: the former has already partnered with major clients to provide drone delivery meteorological services with a resolution of 10 meters, while the latter targets weather derivatives, insurance, and overseas forecast markets, offering a reliable high-frequency pricing and settlement data foundation. With over 80% of industries related to meteorology in the national economy, the concept of pricing and trading weather has been validated abroad, while it is just beginning domestically. When a startup welds the modified attention mechanism of Transformer, its own satellites, and the energy world model into a chain from space observation to industry delivery, it clearly aims to do more than just a forecasting tool—it wants to become the underlying pricing authority for global weather risks.