How AI Became More Expensive Than the Workers It Replaced [video]

- The video argues that the current wave of AI (especially large language models) is so costly to develop and run — in terms of GPUs, energy, data-center infrastructure, and constant human oversight — that it often ends up being more expensive than the human workers it was supposed to replace. - Key reference point: the "Jevons paradox" — the idea that as a technology becomes more efficient, demand for it grows so much that total resource use actually increases. Applied to AI: cheaper inference costs didn't reduce spending; they led to bigger models, more users, and vastly higher total costs. - The video draws on widely reported figures: training a single frontier model can cost $100 million+; running inference at scale requires huge server fleets; and companies like OpenAI, Google, and Microsoft are investing hundreds of billions in data centers with uncertain returns. - For context: much of the current AI hype cycle assumes that automation will bring down costs. This video challenges that assumption, suggesting that AI's high operational cost may limit its economic viability as a replacement for labor in many use cases.