It Ain't AGI Until H100s Go for $30 an Hour (corbt.com)

Greg Brockman recently announced that GPT-6 Astra signifies the dawn of the "AGI era," although true AGI—capable of outperforming humans in economically valuable tasks—remains elusive until AI compute costs rise significantly. Currently, the price for renting an H100 GPU stands at approximately $3.52 an hour, with predictions suggesting that to reflect the value of an AGI capable of replacing white-collar jobs, prices would need to surge to approximately $30 per hour. This dramatic increase would align the high demand for AGI with the limited supply of compute resources, potentially rationing availability or pushing prices even higher. The implications of these cost dynamics are profound for the AI and machine learning community. Should compute costs reach the predicted thresholds, and if AGI can effectively match or exceed the capabilities of human workers, it could revolutionize the job market and redefine economic structures. With the most advanced AI systems currently costing around $8.21 per hour, there are concerns that as demand escalates, prices could climb to as much as $75 per hour based on optimistic market conditions. This scenario highlights not just the technological progression towards AGI but also the need for strategic resource management in AI compute to ensure equitable access and advancement in the field.