Access to frontier AI will soon be limited by economic and security constraints
In a post on his personal blog, Anton Leicht contends that the era of open access to cutting-edge AI is ending. He points to two primary forces: economic constraints, where training and running frontier models cost hundreds of millions of dollars, and security constraints, as governments impose export controls and licensing requirements. Leicht notes that even if models are open-sourced, the compute needed to run them remains prohibitively expensive for most. He cites examples like the US restrictions on exporting advanced chips to China and the high cost of API access to models like GPT-4. The consequence, he argues, is that only large corporations and state-backed entities will have the resources to deploy frontier AI, widening the gap between AI haves and have-nots. Developers and smaller companies may find themselves locked out of the most capable systems, forced to rely on less powerful alternatives or pay high fees.
Developers may lose access to cutting-edge AI models due to rising costs and security restrictions.