New CNAS Report Projects AI Models to be One Million Times More Powerful by 2030

Washington, March 13, 2024 — Today, the Center for a New American Security (CNAS) released a new report, Future-Proofing Frontier AI Regulation: Projecting Future Compute for Frontier AI Models by Paul Scharre, Executive Vice President and Director of Studies at CNAS. The report projects that, by 2030, frontier AI models will be trained on one million times more effective computing power than the most advanced models today if current trends continue.

Scharre emphasizes that policymakers must prepare for a future in which AI systems are significantly more powerful than today and structure government regulations accordingly. The amount of computing hardware used to train advanced AI models could soar by the end of this decade—up to one thousand times greater than what was used to train GPT-4. When factoring in algorithmic progress, frontier AI systems could be trained on one million times more effective computing power than GPT-4. These leaps forward are possible without government intervention (financed solely by large tech companies) and without fundamental breakthroughs in chip design.

Scharre highlights that AI’s explosive growth means policymakers must future-proof regulations to be prepared for AI systems that are significantly more computationally intensive and capable than today. Computing hardware is likely to become even more essential in the future for training the most advanced AI models. Given intense geopolitical competition with China, the United States must adopt policies that safeguard America’s advantages in chips, while ensuring AI’s benefits are widely shared.

For more information or to schedule an interview with Dr. Scharre, please contact Alexa Whaley at awhaley@cnas.org.