US Prepares for Free Chinese AI Models Distributed Like Torrents, Weighs Policy Shift on Open-Source Systems

The Trump administration is preparing for the emergence of powerful Chinese AI models that any user will be able to download for free from the internet, similar to torrent distribution. This scenario is prompting Washington to reconsider existing rules governing American open-source AI systems.

Officials are currently consulting with industry representatives to establish unified capability requirements for such models. Chinese systems already available in open form are being used as the reference point. American developers may be permitted to release models of comparable performance, while more advanced developments would likely face additional restrictions.

Companies anticipate that Chinese models in the Mythos class will become publicly available within the next 6–12 months. Users will then be able to run these systems on their own hardware, modify them, and embed them into other applications without any control from the original developer.

For the United States this creates a difficult policy challenge. Harsh prohibitions could slow domestic companies without stopping the spread of foreign technologies. China is already actively developing open neural networks and distributing them freely to strengthen the international position of its developers. Some Chinese systems have gained rapid popularity after restrictions were placed on American models, largely because they remain accessible and cost less.

Simultaneously, technology companies are seeking ways to reduce electricity consumption in data centers. These facilities currently account for approximately 1.5% of global electricity usage, and the International Energy Agency expects demand to continue rising quickly.

The Washington Post observes that long-term forecasts may prove overstated. Companies are testing superconducting cables capable of transmitting more energy with lower losses, as well as optical connections that move data using light instead of electrical signals. These technologies could significantly cut the energy required to transfer information between computing devices.

Because the new solutions have not yet been deployed at scale, real-world savings remain difficult to quantify. Nevertheless, progress in superconductors and optical systems indicates that growth in computing power will not necessarily produce an equally sharp increase in electricity consumption.