Compounding a 4x per-year growth in model and compute size with continuously growing usage demand means demand for millions of GPUs is not speculative—it is a near-certainty based on the mathematics of scaling.