This report is further proof that physical infrastructure is the primary bottleneck of the neural network market. The growth in the number of multi-gigawatt clusters requires massive volumes of transformers, uninterruptible power supplies, and liquid cooling systems. While software startups try to prove their profitability, heavy electrical equipment manufacturers like Schneider Electric record maximum margins, serving as the essential foundation without which no NVIDIA or Huawei supercomputer can physically plug into the grid.
Source: Schneider Electric / Reuters
HardwareSchneider ElectricEnergyData CentersMacroeconomics