This seems less about frontier models and restriction and more just lack of compute capacity to meet demand. This has always been an issue for large clients running on cloud, though not to this extent.