“In the current exponential scale-up phase of compute, an AI model trained for $1 billion might generate $4 billion in revenue and cost $1 billion to inference, resulting in $2 billion profit for that model (75% gross margins), but the company still loses money by spending $10 billion to train the next model.”