US Implements New AI Chip Regulations to Control Advanced Technology Exports

The US government has announced a new regulation to manage the export of advanced artificial intelligence (AI) chips and related technologies. This rule aims to limit global access to US-designed GPUs (graphics processing units) and other critical components essential for training and running sophisticated AI applications.

GPUs, initially developed for graphics rendering, have gained prominence for their ability to process large datasets simultaneously, making them integral to advanced AI systems like OpenAI’s ChatGPT. The regulation sets country-specific caps on GPU compute power, measured in Total Processing Performance (TPP), restricting exports to a total of 790 million TPP through 2027 for most nations. This is roughly equivalent to nearly 50,000 Nvidia H100 GPUs, a high-powered GPU model critical for cutting-edge AI research and applications.

However, exceptions exist. Companies with “Universal Verified End User” or national authorizations can access up to 320,000 GPUs over two years. Small-scale GPU orders and gaming GPUs are also exempt. Furthermore, countries such as the US, Japan, and most of Europe remain unrestricted in acquiring these technologies.

The rule also addresses “model weights,” a critical element in training AI models, by setting security measures to safeguard sensitive parameters in advanced non-public models. These restrictions aim to ensure that the development of advanced AI remains within secure and trusted environments.