At NVIDIA’s GTC 2026 conference, CEO Jensen Huang once again laid out the “five-layer cake” blueprint for the AI industry—ranging from the foundational layers of energy, chips, and infrastructure, to the model and application layers—aiming to establish ecosystem influence at every level. Around the same time, the inflection point in computing power consumption brought about by OpenClaw propelled Chinese models to capture the top three spots in global token consumption, accounting for over 60% of the total. Meanwhile, Elon Musk predicted that China’s electricity generation could reach three times that of the United States by 2026.
These three pieces of news converge on a crucial but often overlooked theme: a paradigm shift is underway in energy trade within the AI era—electricity no longer needs to cross borders to be “exported.” By “packaging” electricity into tokens through domestic large models (such as Minimax, Kimi, GLM, etc.), China is transforming its energy advantages into computing power service advantages in an unprecedented way, bypassing the traditional constraints of geography, hardware, and regulations that have long restricted electricity exports. Tokens are becoming the new form of electricity export.
I. The AI “Electricity-Guzzling” Era: Electricity Becomes the New Oil Jensen Huang’s “five-layer cake” framework places energy at the very bottom, which is no accident. The electricity demand of AI data centers is reshaping the global energy landscape. According to the International Energy Agency (IEA), global data center and AI-related electricity consumption could rise from 460 TWh in 2022 to 620–1050 TWh by 2026, with a four-year compound annual growth rate reaching as high as 22.9%. By 2030, data centers alone could account for 8%–12% of total U.S. electricity consumption.
In a January 2026 podcast, Musk stated bluntly: “China will produce more electricity than any other country, and it will likely also have more chips. Based on current trends, China will far surpass the rest of the world in AI computing.” He estimated that by 2026, China’s electricity generation could be three times that of the United States—a decisive advantage in the AI race.
A Goldman Sachs report in November 2025 noted that electricity shortages could slow the U.S. pace in the AI race. In contrast, China has been steadily expanding its energy capacity, expected to possess about 400 GW of spare power capacity by 2030—more than three times the total electricity demand of global data centers.
II. A New Strategic Path: The Token Economy—The Optimal Carrier for China’s Electricity Export Traditional electricity exports face three inherent barriers: geographically, electricity cannot be transmitted over long distances across borders; hardware-wise, grid interconnection requires massive infrastructure investment and geopolitical maneuvering; regulatory-wise, electricity trade is subject to strict national energy security policies.
However, the emergence of the token economy has fundamentally reshaped the logic of electricity export. When domestic large models (Minimax, Kimi, GLM, etc.) deploy their computing centers within China, using the abundant domestic electricity resources to complete inference computations and then exporting token services globally via APIs, electricity has already achieved “value export”—what users consume are tokens, but what they essentially utilize is Chinese electricity.