How A.I. Is Changing the Way the World Builds Computers (Published 2025)

data-centersgpunvidiaai-infrastructureenergycooling

Abstraction: AI is fundamentally rebuilding computing hardware, data centers, and energy systems

Key points:

Amazon, Meta, Microsoft, and Alphabet combined capital spending on data centers could top $320 billion in 2025 — more than twice what they spent two years prior; OpenAI's Stargate plan alone targets $500 billion.
GPUs (originally for game graphics) now power AI's parallel math operations; a single GPU draws up to 1,000 watts vs. 250–500W for a CPU; converting a 5MW traditional data center to AI GPUs reduces capacity to 8–10 server rows.
Data centers consumed 4.4% of total US electricity in 2023; could triple by 2028 per DOE; companies are pursuing nuclear (Three Mile Island restart by Microsoft, Amazon/Google reactor deals) and gas turbines (xAI Memphis).
Google's data centers consumed 6.1 billion gallons of water in 2023 (up 17% YoY) for liquid cooling; AI chips require direct water-pipe contact with chips, not just room cooling.
DeepSeek built a top-performing AI using only 2,000 H800 chips vs. OpenAI's 10,000+ H100s, challenging the bigger-is-better spending thesis — but US giants remain unfazed in their AGI pursuit.
Meta incurred a $4.2 billion restructuring charge in 2023 to redesign future data centers for AI; its newest Louisiana facility would cover most of lower Manhattan.

Connections: Nvidia · Openai · Google · Meta · Microsoft · GPU Computing · AI Infrastructure