In May 2004, Intel cancelled Tejas and Jayhawk because the chips couldn't be cooled, as Dennard scaling had stopped holding, leakage current was climbing exponentially, and power density was heading past what air cooling could handle. In June 2026, Gartner forecasts global data center electricity consumption at 565 TWh this year, up from 447 TWh in 2025, with power demand hitting 132 GW and projected to reach 290 GW by 2030; AI-optimized servers account for 31% of data center power consumption in 2026 and will pass conventional servers in 2027. Gartner's Linglan Wang said AI capacity is now constrained by power availability. At TSMC's Amsterdam symposium on May 28, Kevin Zhang said the improvement customers most want is energy efficiency, with TSMC targeting 30% efficiency improvement per generation while shipping 1,000W chips and looking at megawatt-class systems before the decade is out. OpenAI's Jalapeño ASIC, disclosed at Hot Chips in August, ships at 700W where NVIDIA's equivalents draw 1,200 to 1,400W. According to The 1/W Law (Chen et al., March 2026), for Llama-3.1-70B on H100-SXM5, TP=8, fp16, tokens per watt halves when context doubles: power barely moves while throughput falls linearly with context.