HomeArticle

An AI large meteorological model company has received tens of millions of yuan in investment from Houxue, and has built its proprietary satellite constellation | 36Kr Exclusive

欧雪2026-10-13 10:12
Construct the Energy Meteorology Large Model CMA-WSP-AI

Source / Enterprise

This article contains approximately 3300 words, with an estimated reading time of 7 minutes

Author | OU Xue

Editor | YUAN Silai

Hard Krypton learned that Tianxu Intelligent Technology (Jiaxing) Co., Ltd. (hereinafter referred to as "Tianxu Intelligence"), a meteorological AI company, has recently completed a Pre-A+ round of financing of tens of millions of yuan, with exclusive investment from Houxue Capital, and Yirong Capital acting as the long-term exclusive financial advisor. The funds from this round of financing will be mainly used for the construction of the "Ningkong Constellation", the iteration of CMAWSPAI and vertical domain world model, team expansion, key customer development and pre-research of overseas business.

Founded in October 2024, Tianxu Intelligence is one of the few global hard-tech AI for Science companies that simultaneously master vertical meteorological large models, independent and controllable satellite observation networks, and industry base delivery capabilities.

Founder LIU Li holds a doctorate in meteorology from the University of Chinese Academy of Sciences. He has been deeply engaged in meteorology throughout his undergraduate, master and doctoral studies, and has long studied the intersection of meteorology, artificial intelligence and energy.

"We have been deeply engaged in this industry for nearly 10 years, tried the integration of meteorology with energy, transportation, agriculture and other fields, and finally found that energy is the golden track driven by policies, with the largest market space, the strongest willingness to pay, and the most scarce high-precision meteorological base around the world.

Tianxu Intelligence is one of the very few domestic startups that have completed the "large model on satellite" layout, committed to building an independent and controllable new type of energy meteorology infrastructure, and deploying energy vertical world models and industrial-level physical simulation.

The company believes that new energy "depends on the weather", and wind speed and irradiance directly determine the power generation. At present, the proportion of domestic new energy installed capacity has exceeded 50%, the market-oriented reform of electric power is accelerating, power forecasting has spawned a 100-billion-level rigid demand, and the spot electricity trading has opened up a 1-trillion-level increment, and the track is on the eve of an outbreak.

Energy is only the first stop of the weather economy. Meteorology is the largest variable of power supply and demand, and also one of the underlying pricing variables of the economy and society. Tianxu Intelligence's goal is to base on China's energy scenarios, start from power trading, become the underlying pricer of global weather risks, and build the "most powerful brain" in the energy vertical field.

From the perspective of industry development, traditional weather forecasting relies on physical models, which is essentially a process of solving mathematical and physical equations, requiring a large amount of computing power.

LIU Li gave Hard Krypton an intuitive example: in the past, forecasting the weather for the next 10 days might take more than ten hours with a supercomputer; now, with the AI large model, the weather for the next 20+ days can be calculated in one minute, and the forecasting accuracy of many meteorological elements is even higher.

Tianxu Intelligence's meteorological large model is not a large language model, but a vertical domain model with a parameter scale within 10B. The company has independently developed from the first line of code, carried out vertical transformation in the energy field on the basis of the mainstream architecture, and innovatively proposed the "Lorentz Framework" — this name pays tribute to Lorentz, the meteorologist who proposed the chaos theory.

"Our core is to take Transformer as the underlying framework, and modify the traditional attention mechanism for the phenomena such as upstream and downstream spatial correlation and time drift in energy meteorological forecasting. Our goal has been focused on the forecasting of wind speed and irradiance from the very beginning, which is different from general meteorological large models", LIU Li introduced.

Since 2025, Tianxu Intelligence has worked with the Public Meteorological Service Center of China Meteorological Administration to build a unique workflow from large model data processing to vertical model training, and jointly released CMA-WSP-AI, the first domestic energy meteorological large model, in Zhongshan, Guangdong in October 2025.

Different from traditional meteorological large models that "only output meteorological results such as temperature, humidity and precipitation", Tianxu Intelligence's energy-oriented vertical world model can further deduce: how meteorological changes transmit to new energy output fluctuations, how to affect the load of the whole network, how to drive the rise and fall of spot electricity prices, realizing the leap from "calculating weather" to "calculating the evolution of energy system", and providing simulation and deduction capabilities for power trading decision-making, large base planning and power grid risk rehearsal.

In terms of model effect, Tianxu Intelligence's 1km high-precision wind speed prediction outperforms EC-HRES and AIFS in indicators such as RMSE and correlation coefficient; it ranks first in the extreme weather AI competition (including four items: hail, thunderstorm gale, thunderstorm and short-time heavy precipitation), and the TS score is increased by 39.6% to 351% compared with the traditional baseline method.

In terms of commercialization, Tianxu Intelligence has run through the three-layer business model of "data - platform - capacity building". The bottom layer is data sales, and its customers include leading power generation groups, meteorological departments and multiple types of entities participating in power trading; the second layer is the SaaS-based auxiliary decision-making platform for power trading and power forecasting platform, which is subscribed by account and generates continuous revenue; the third layer is capacity building, which deploys data, models and platforms to the customer's intranet as a whole to meet the needs of leading customers for energy data security, digital transformation and private deployment. The three layers of business progress step by step: data is the entry, the platform brings repurchase, and private deployment locks in leading customers; the more stations and scenarios are accessed, the more accurate the model is, forming a positive data flywheel.

"The traditional model is that the power generation side transmits data to a third party, and the third party processes it into products and then transmits it back. However, with the improvement of energy security requirements, this model no longer works, and data cannot leave the intranet", LIU Li explained, "We build models and platforms in the customer's intranet, so that data and models run in a closed loop in the intranet, and help customers master this independent capability at the same time."

At present, Tianxu Intelligence's customers have covered the power generation side, power consumption side and power grid side: on the power generation side, it has reached cooperation with many leading central power generation enterprises; on the power grid side, both State Grid and China Southern Power Grid are its customers. This means that Tianxu has obtained the "admission ticket" of the most core customers in China's energy system. LIU Li believes that the transformation from "result procurement" to "capacity building" is the core difference between Tianxu and traditional meteorological service providers.

The large model is the "brain", and the satellite is the "eye". LIU Li believes that the algorithm advantage will eventually be caught up, and the data source is the long-term barrier. Tianxu Intelligence launched the "Ningkong Constellation" plan. On July 24, 2026, the meteorological large model was carried on a satellite into orbit and completed daily verification, making it one of the earlier domestic startups to send a meteorological large model "into space". The long-term plan of the constellation is 18 satellites, which will build a global 1km/1h observation network for wind, light, temperature, pressure and humidity, take the rare route of "AI large model + own satellite + on-board computing", and serve the domestic, Belt and Road Initiative countries and the global market with a global vision.

In addition to energy, the company has also deployed low-altitude meteorology and financial meteorology: the former has carried out meteorological services for UAV distribution with leading customers, with a resolution of 10-meter level; the latter provides a credible and high-frequency pricing and settlement data base for weather derivatives, insurance and overseas forecasting markets.

The following is an excerpt of the conversation between Hard Krypton and LIU Li, founder of Tianxu Intelligence (edited):

Hard Krypton: From "data consumption" to "data production", how to understand this statement? Why does Tianxu make its own satellites?

LIU Li: The so-called data producer is to rely on the meteorological large model to generate meteorological data and master the data source. Therefore, we will gradually shift our competitiveness from algorithm advantage to data advantage.

The overall plan is divided into three steps: in July 2026, the large model has been carried on a satellite into space, and the on-orbit operation verification has been completed. In 2027, two pilot test satellites are planned to be launched to verify the capabilities of inter-satellite transmission and on-board computing, serve new energy stations, make up for the shortage of wind and solar observation data in the "sand, gobi and desert" areas, and lay a solid foundation for resource assessment and energy prediction. In the long term, 18 satellites are planned to be deployed to realize the detection of wind, light, temperature, pressure and humidity elements with 1-hour revisit and 1km resolution across the globe.

This kind of observation data is a scarce resource in new energy scenarios. In addition to serving the domestic market, it can also support overseas expansion, especially serving countries along the Belt and Road Initiative. Satellite data capability is one of the most core barriers for meteorological service companies in the future.

Hard Krypton: Compared with traditional enterprises in the industry, what are the differences of Tianxu?

LIU Li: There are two main points. First, different from traditional power forecasting manufacturers, we are the producer of underlying meteorological data ourselves, and we develop business applications from core meteorological data, with different technical starting points. Relying on the meteorological large model, we have obvious advantages in prediction accuracy in some regions.

Second, different from scientific research teams, we are a team growing up from industry scenarios. More than 80% of the company's employees have master's degrees or above, and a large number of core personnel have been deeply engaged in the energy field for more than 10 years. We are good at technical innovation and commercial implementation in cross-cutting fields.

Hard Krypton: What challenges is the industry currently facing? Will the evolution of technical routes affect your subsequent R&D direction?

LIU Li: First, we will closely follow the cutting-edge trends at the technical level. The large model is transitioning from pure data-driven to the two-wheel driven route of physical constraint plus data-driven. We will combine the energy scenario to dig deep into the underlying logic, continuously evaluate and test cutting-edge technologies, and find a differentiated implementation path.

Second, we are not afraid of industry competition in commercial promotion. Our team has an open and cooperative gene. What really needs to be considered is the fundamental of the long-term development of the new energy industry, and the general direction of China's vigorous development of new energy will not change. Previous regional conflicts pushed up international oil prices, while China was relatively less affected. The root cause lies in the transformation of the energy structure, and the proportion of new energy installed capacity has exceeded 50%. Looking back, the top-level energy strategy has a long-term vision.

Third, the market space of the weather economy is far from being opened. About 80% of the industries in the national economy are related to meteorology, and new retail, finance, agriculture, shipping and other industries can achieve refined operation and risk management with the help of meteorology. Overseas weather derivatives and forecasting markets have proved that weather can be priced and traded; this market in China has just started, which is not only a challenge for us, but also the biggest opportunity.

Investor View:

Houxue Capital said: As China accelerates the construction of a new power system and the global carbon neutralization process continues to advance, energy meteorology has been upgraded from an auxiliary tool to a key digital new infrastructure related to energy security and global green transformation. Tianxu Intelligence independently develops vertical meteorological large models, pioneered the layout of the "Ningkong Constellation" independent satellite observation network, and opened up the whole chain of "space observation - meteorological large model - energy vertical world model simulation - industry base delivery". Its technical system spans meteorological science, artificial intelligence, cutting-edge world models, aerospace engineering and power industry. It not only deeply empowers the construction of domestic energy system, but also has great potential to export China's original meteorological technology base to the world, with broad imagination space in energy, low altitude, agriculture, finance and other national economy and people's livelihood tracks. The founding team has solid meteorological scientific research skills, full-stack engineering implementation capabilities, industrial commercial operation experience and international vision. We are optimistic that the team will seize the major era opportunities of scientific and technological innovation and global energy industry upgrading in the "15th Five-Year Plan" period, build a globally leading independent and controllable energy meteorology technology system, and help national energy security and empower the high-quality development of the global real economy with the integrated intelligent infrastructure solution of software and hardware.