The Sequence Opinion- Issue 918: The Energy Scaling Laws of AI

AI processing requires significant physical infrastructure, including substations, cooling systems, and power plants, rather than solely existing in the cloud. The advancement of AI scaling is increasingly dependent on the efficient conversion of energy, such as photons and atoms, into usable intelligence. The physical reality behind modern AI applications involves substantial energy consumption and heat generation, highlighting the critical role of power infrastructure.

• AI computation happens in physical infrastructure like substations, cooling loops, transmission networks, and power plants.
• The next phase of AI scaling depends on efficiently converting energy (photons, atoms) into intelligence.
• Modern AI applications, despite a seemingly weightless interface, rely on extensive hardware for processing, memory, cooling, and power conversion.
• Nearly all energy used in AI clusters is eventually released as heat.
• The cloud’s operation is fundamentally tied to its power source.
  https://bender.layer3.press/articles/dc4f07ea-cff3-480d-ac5a-874497bf8747