China’s Envision Group is setting its sights on what it calls the next frontier in artificial intelligence: building a “physical AI” energy foundation model that Zhang Lei, the company’s chairman, says could outpace American efforts in both scale and sophistication within three years.
Speaking at a closed-door Envision technology conference on October 19, themed “Artificial Intelligence and Future Energy Systems,” Zhang detailed his vision for integrating AI with physical laws, system boundaries, and knowledge graphs. Unlike traditional large language models, which largely identify correlations, physical AI aims to understand causality—leveraging principles like the conservation of energy, aerodynamic equations, and power flow calculations to generate reliable real-world outputs.
“Traditional AI can only recognize relationships; it cannot construct causality,” Zhang said. “The future lies in physical AI, which will allow us not just to understand the world but to change it.”
Envision, founded two decades ago as a wind turbine manufacturer, has expanded into energy storage, power batteries, green hydrogen, ammonia, and zero-carbon industrial parks. Today, the company describes itself as an energy systems provider.
Building a large energy model, Zhang said, is central to Envision’s strategy. “This is an area the United States can’t handle,” he asserted. “American AI still leans heavily toward consumer applications. But when it comes to physical AI and large energy models, they lack both industrial scenarios and hands-on experience in new energy manufacturing.”
China, by contrast, Zhang argued, has abundant application scenarios and industrial data, giving it a potential edge in global energy AI leadership.
Zhang framed his vision against China’s broader energy transformation. In March 2021, China’s Central Financial and Economic Affairs Commission emphasized the creation of a renewable-energy-centered power system. While wind and solar power can drastically cut carbon emissions, their intermittency complicates grid management and electricity market operations, creating anxiety among operators.
“This anxiety is an opportunity for AI,” Zhang said. “Energy foundation models can process massive amounts of data in milliseconds, uncover hidden patterns, and make optimal decisions. They enable wind turbines to generate higher returns, energy storage to participate more effectively in markets, and the grid to accommodate more renewable energy safely.”
However, building AI with such “super capabilities” is no small feat. Zhang emphasized that isolated experience in a single energy domain—wind turbines, solar panels, or storage—is insufficient. True energy AI requires understanding the entire electricity market: fluctuations in wind, grid operations, and load changes, supported by extensive underlying data.
Envision’s projects, such as the Chifeng Zero-Carbon Hydrogen Industrial Park, provide ideal training grounds. The park, home to the world’s largest green hydrogen-ammonia project and an independent renewable energy grid, forms a closed-loop system of devices generating rich datasets for AI model training.
At the conference, Envision unveiled the “Tianshu” Energy Foundation Model, which integrates massive amounts of weather, device, grid, and market data. Using advanced algorithms—including graph neural networks, spatiotemporal models, and multimodal Transformers—the system enables real-time control through cloud-edge-device collaboration.
The company also revealed AI-powered products, including the Envision Galileo AI Wind Turbine and Galileo AI Energy Storage. While specific performance metrics are still limited, internal tests reportedly showed AI wind turbines achieving 20.9% higher returns than non-AI turbines at the same site.
Beyond technical innovation, Zhang suggested that physical AI could help solve a chronic problem in China’s new energy sector: “internal competition.” Overcapacity, price wars, and homogenized offerings have plagued photovoltaics, wind power, and battery industries, pushing many companies into financial strain. National initiatives have recently sought to curb these practices through industry self-discipline agreements and anti-competition campaigns.
“The recurring internal competition stems from obsession with size and muscle,” Zhang said. “Physical AI-driven energy foundation models shift the focus from material assets to intelligent assets. This enables companies to compete on efficiency and performance rather than scale alone.”
He added that intelligent energy systems could usher in a new era of rational prosperity, balancing high returns with high efficiency, while also supporting China’s dual-carbon transition goals.
Zhang highlighted that as the energy sector matures, decision-making complexity will increase. “In a more market-oriented power system, with financialization of energy, it’s no longer enough to just ‘build muscle’; more attention must be paid to ‘growing brains,’” he said.
Large energy models, he said, are crucial enablers of this shift. They provide the analytical foundation for intelligent operations, risk management, and optimized decision-making across complex energy systems.
Physical AI and large energy models are still emerging. Many current solutions face limitations in data quality, computing power, safety, and verification, and the industry lacks standardized evaluation metrics. Furthermore, how companies recover R&D investments and generate commercial returns remains uncertain.
Nonetheless, Zhang expressed confidence in near-term progress. He predicted that within one to two years, the impact of Envision’s large energy model will become visible, and in roughly three years, it will reach a maturity comparable to Level 3 autonomous driving systems in terms of operational autonomy.
Looking further ahead, Zhang envisions a future energy ecosystem composed of millions of intelligent agents capable of continuous evolution—akin to a coral reef. This system would optimize green electricity integration, lower costs, and enhance operational robustness and security.
“We are committed to driving the development of large energy models to empower the entire energy industry ecosystem, facilitating its transformation from equipment-based to an ‘intelligent agent’ ecosystem,” Zhang said.
If successful, Envision’s approach could redefine the global competitive landscape in energy AI. While U.S. tech firms dominate consumer-focused AI applications, they lack comparable industrial-scale datasets and hands-on experience in renewable energy operations. China’s wealth of energy scenarios, combined with deep integration of AI and physical laws, could give companies like Envision a decisive advantage.
Analysts note that the push toward physical AI is consistent with broader trends in energy digitalization, grid optimization, and industrial intelligence. Yet they caution that commercialization, regulatory hurdles, and integration challenges will test even the most advanced models.
For now, Zhang Lei’s vision signals China’s ambition to lead not just in renewable energy deployment but in the AI systems that will govern and optimize it. In a sector where efficiency, intelligence, and reliability are increasingly valuable, Envision’s Tianshu model could mark a transformative step from understanding energy systems to reshaping them.
