In the Spring Festival of 2026, while the world was still marveling at OpenAI's latest model parameters, China used a Spring Festival Gala to showcase another facet of AI to the world—the physical implementation of embodied intelligence.
Looking at the program list of the 2026 CCTV Spring Festival Gala, we witnessed an unprecedented "AI military parade." This was no longer the simple mechanical dance displays of a few years ago but a concentrated explosion of China's robotics industry featuring "multiple enterprises, multiple models, and full-scene applications."
- Magic Atom's full-stack cluster made robots the best "atmosphere group," accompanying Chen Xiaochun and Yi Yangqianxi in "Intelligent Manufacturing Future" with coordination so seamless it was hard to distinguish from humans.
- Unitree's G1 and H2 robots demonstrated astonishing motion control capabilities in "Wǔ BOT"—without real-time remote control, relying entirely on edge-side computing power for autonomous balance. When the H2, draped in a red robe, performed a sword dance, it proved that the motion cerebellum of Chinese robots had matured.
- Songyan Power, in the skit "Grandma's Favorite," assigned robots the comedic role of delivering and responding to punchlines, completing the transition from "prop" to "actor."
- Galaxy General's Galbot G1 performed "walnut rolling" in a microfilm—a seemingly simple action that showcased the pinnacle of dexterous hand and tactile feedback technology.
The Gala sent a clear signal: China's AI is not just confined to servers; it has grown limbs and entered reality.
However, just as we were cheering for the robots, Wall Street was plunged into silent panic. They discovered that the "lifeblood" powering these systems—electricity—was running dry. Shifting our gaze from the Gala stage to Silicon Valley's data centers, we encounter the elephant in the room—power.
By early 2026, U.S. residential electricity prices had surged by 36%, reaching $0.18 per kWh. But this was just the surface; the core crisis lay in the collapse of the supply side. Training a GPT-4-level model consumes as much electricity as 100,000 households use in a year. By 2028, the annual electricity consumption of U.S. data centers is projected to skyrocket to 600,000 GWh.
The U.S. power grid is facing a dual crisis of "heart disease" and "vascular blockage." Five percent of its electricity relies on aging fossil fuels and nuclear power, and these units are facing a retirement wave. The U.S. grid is fragmented into three major islands—East, West, and Texas—with poor interconnectivity. Approving an interstate transmission line can take up to 15 years, preventing wind power from the Midwest from reaching East Coast data centers.
As Sam Altman said, "Energy is currency." In Silicon Valley today, CEOs are no longer troubled by chip quotas but by—where can they find enough electricity to run these chips?
If computing power is the engine of AI, then electricity is its fuel. In this energy game, China, with a decade of超前布局 (forward-looking planning), has built a strategic moat that the U.S. cannot easily replicate. If computing power is the engine of AI, then electricity is its fuel. In this energy game, China, with a decade of超前布局 (forward-looking planning), has built a strategic moat that the U.S. cannot easily replicate.
By 2025, China had completed 45 ultra-high voltage (UHV) projects, with the total length of UHV DC transmission lines exceeding 40,000 kilometers. These "electricity superhighways" can transmit abundant clean energy from the west to data centers in the east at millisecond speeds, or directly support the "East Data West Computing" hubs. China possesses 35 of the world's 37 largest high-voltage direct current cable systems—an infrastructure gap that is an insurmountable chasm for the U.S. in the short term.
The high energy consumption of AI inherently demands clean energy. In 2025, the proportion of renewable energy installed capacity in China historically surpassed 60%, with new wind and solar installations exceeding 430 million kilowatts. Of the total electricity consumption, nearly 4 out of every 10 kWh came from green power. While the U.S. is still grappling with nuclear power plant delays, China has achieved grid parity for photovoltaics and wind power, providing a cheap and green energy solution for high-consumption AI data centers.
China is the global manufacturing center for transformers, accounting for over 60% of global production. The biggest pain point for U.S. grid upgrades is transformer shortages, with delivery times now stretching to 3-4 years. Whether through transshipment via Mexico or direct procurement, the maintenance of the U.S. grid is highly dependent on Chinese manufacturing. While U.S. data centers face shutdowns due to transformer shortages, China's power equipment companies are operating at full capacity, supporting the rapid expansion of domestic computing infrastructure.
The 2026 Spring Festival Gala was not just a carnival of robots but also a profile of China's industrial strength.
When we watch Unitree's robotic dogs tumble and Galaxy General's robots perform tasks on screen, we must not forget: behind every agile movement, there is not only advanced algorithms but also a stable current transmitted thousands of kilometers through UHV lines, supported by a robust power grid.
In the second half of this AI revolution, the marginal cost of computing power growth will no longer depend on the nanometer size of chips but on the cost of acquiring joules. The U.S. possesses top-tier algorithm design, but China has the most powerful energy conversion and delivery system.
For investors, the logic is clear: in this gold rush, if NVIDIA is selling shovels, then China's infrastructure builders (UHV, power equipment, green energy) control the water source.
