Elon Musk warns: American chips will face dark times, while China will perfectly avoid this disaster!

At the early 2026 Davos Forum in Switzerland, Elon Musk warned that the American artificial intelligence chip industry is on the brink of a crisis of 'oversupply but unusable'; meanwhile, China may bypass this storm unscathed due to a completely different energy infrastructure path.

Musk's core argument is quite simple: no matter how many AI chips there are, they are useless without power. He pointed out that the global production of AI chips is currently skyrocketing—especially NVIDIA's H100 and B100 series GPUs, which have astonishing performance and overflowing orders. However, the problem is that once these chips are deployed in data centers, each one 'devours electricity like a cow.' A 10,000-watt AI training cluster easily exceeds 100 megawatts in power consumption, equivalent to the electricity consumption of a medium-sized city.

The U.S. is precisely stuck at this 'last mile': aging power grids, lengthy approval processes, and insufficient transmission and distribution capacity. In states like California, Texas, and Arizona, which are crowded with tech companies building data centers, power supply has increasingly tightened in recent years. Insiders in the energy industry have revealed that NVIDIA's two ultra-large-scale AI data centers planned in Santa Clara may be forced to delay their launch for several years due to the local grid's inability to provide stable high-load power supply. This is not a technical issue; it is a hard constraint of infrastructure.

What’s more troublesome is that the American power grid system is highly decentralized, operated by hundreds of private and local utility companies, and upgrading capacity requires layers of approval, often taking five to ten years. Meanwhile, residential electricity prices are rising—because grid companies pass on upgrade costs to users. This creates a vicious cycle: the hotter AI gets, the more electricity it demands, the grid can’t keep up, electricity prices rise, ultimately slowing down the entire pace of AI deployment.

In contrast, by the end of 2025, China’s solar power installed capacity will reach 1,118,000 megawatts, nearly 4.7 times that of the United States at 238,000 megawatts. More importantly, China is massively building 'wind and solar power bases' in the western regions—concentrated photovoltaic and wind farms, paired with ultra-high voltage transmission lines, directly sending green electricity to eastern data center clusters.

According to the International Energy Agency, by 2030, global data center electricity consumption will account for more than 4% of total electricity demand, with AI loads exceeding 60%. Whoever can ensure power supply while controlling costs will hold the ticket to enter the AI era. Currently, the U.S. still leads in chip design and manufacturing, but in the more practical issue of 'whether chips can actually run,' it is being dragged into a 'dark moment' by its own infrastructure shortcomings. Meanwhile, China, leveraging the large-scale deployment of renewable energy and the advancement of a national computing power network, is likely to not only keep pace in this round of AI competition but also accelerate its progress.

Brother Dao believes that Musk's warning is less of a warning and more of a calm analysis of the gap in reality.