US AI Policy Is Stuck in Training Mode

The article discusses the misalignment of U.S. AI policy with the current state of AI innovation. The U.S. policymakers have prioritized access to AI compute primarily for model training while failing to adapt to the increasing importance of inference in AI performance. Companies like Microsoft are building data centers abroad to support localized inference in AI systems. The U.S. energy policy also emphasizes support for energy production for AI training while overlooking the need for energy-efficient inference models.