The author reflects on the rapid advancements in AI, particularly through insights gained from a podcast episode about Google as an AI leader. They highlight a quote from Greg Corrado of Google Brain, stating that nature operates in the most energy-efficient manner. The author emphasizes the vast energy disparity between human brains and current AI technology. A human brain consumes about 20 watts, while powerful GPUs, like the RTX 5090, can consume upwards of 800 watts. Even with advanced models such as GPT 5.2 or Claude Opus 4.5, there's a significant gap between AI capabilities and human efficiency. The article traces the evolution of computing from the IBM 7090, a large and power-hungry machine from 1959, to today's AI landscape dominated by giants like Anthropic, OpenAI, and Google. The author notes the anomaly of Nvidia's growth, driven by the demand for AI processing, which has shifted from gaming to data centers. They express skepticism about the sustainability of the current AI boom, particularly regarding the heavy investment in power and real estate for data centers. The high costs associated with GPU hosting are likely unsustainable as model architectures evolve. Looking ahead, the author predicts a shift where AI computing will be integrated into everyday devices rather than relying solely on powerful data centers. They argue that cloud hosting will remain important, but on-device AI capabilities will grow, leading to lower costs and greater accessibility. As model architectures improve in efficiency, the author warns of potential declines in the market capitalizations of hardware vendors and real estate investors tied to the current AI infrastructure, shifting the landscape of AI capabilities.