This article dissects the AI super‑cycle economics by breaking down the full‑stack cost of AI factories, revealing that GPUs account for only half of expenses while power infrastructure, labor, and cooling dominate, and examines how token value, bottlenecks, and competitive strategies shape the market.
Goldman Sachs’ analysis models AI capital expenditure, showing a $7.6 trillion cumulative investment from 2026‑2031 and highlighting four key variables—chip lifespan, data‑center costs, architecture mix, and construction delays—that determine whether AI infrastructure spending will expand or contract.
Alphabet’s announced $1.75‑$1.85 trillion AI‑focused capex for 2026, dwarfing past spending and outpacing rivals, reveals a pivotal change in the AI competition, with 75% earmarked for GPU clusters and data‑center power, driving higher cloud costs, tighter supply, and strategic shifts for data teams.
The article explains the leaf‑spine network architecture introduced by Facebook, details its three layers (cabinet, Leaf, Spine), quantifies the required servers, switches and optical modules per cabinet, and analyzes the associated CAPEX for ultra‑large cloud data centers.
China's Internet Data Center (IDC) industry, valued at 1600 billion yuan in 2019, is transitioning from basic services to high‑performance, cloud‑driven solutions, with CAPEX and OPEX dominated by infrastructure and power costs, while 5G and cloud growth fuel further market expansion.
