Cheaper intelligence was supposed to end the AI trade. The bear case rested on compute staying scarce. DeepSeek cracked that assumption by matching frontier output at a fraction of the cost. Cheap intelligence meant the money pouring into capacity was about to be wasted. However, the opposite happened. The reason starts with scaling. More compute makes it possible to run more capable models. So when cost falls the same spend buys capability the old models never had. Jevons paradox does the rest. Each decline in cost makes new use cases viable, so total spending rises even as the cost per unit falls. That accounts for demand, but the bubble case rests on supply. AI compute doesn't flood the market the way telecom did in the 1990s, because the constraints are physical: fab capacity, advanced memory, power, and the EUV behind them. Each takes years to expand. Supply stays scarce while demand climbs. If that holds, the market oversupply that people are expecting may not form. What looks like a bubble could just be demand meeting a hard limit on supply. (Original content provided by @PonderingDurian at Delphi Ventures)