Why Nvidia is worth $5 trillion: Inside a $35 billion, 1 gigawatt AI data center.

Core Insights - Nvidia is capturing a significant share of the AI spending boom, contributing to its valuation of nearly $5 trillion [1][12] - The new metric for AI data centers is gigawatts of computing capacity, with 1 gigawatt costing approximately $35 billion, representing a new economic foundation for AI [3][4] AI Data Center Costs - The largest cost component in an AI data center is GPUs, accounting for about 39% of total spending, with Nvidia's chips being the primary drivers [7][12] - Networking equipment constitutes around 13% of data center costs, benefiting companies like Arista Networks, Broadcom, and Marvell [14] - Power distribution and cooling infrastructure represent significant costs, with power distribution alone taking up nearly 10% of spending [16][18] Company Opportunities - Nvidia captures nearly 30% of total AI data center spending as profit due to its high gross profit margins of 70% [12] - TSMC, Nvidia's foundry partner, earns approximately $1.3 billion per gigawatt from manufacturing GPU components [12] - Other chipmakers like AMD and Intel are attempting to catch up, while hyperscalers such as Google, Amazon, and Microsoft are investing in custom AI accelerators [13] Infrastructure and Operational Costs - Real estate, electricity, and labor account for about 10% of upfront costs, with operational costs being relatively low [18] - The annual electricity cost to run a 1 gigawatt AI data center is about $1.3 billion, and personnel costs are minimal, with only 8 to 10 staff members typically required [18] - The bottleneck is shifting towards power availability, with companies like Siemens Energy and GE Vernova experiencing increased orders for turbines and grid infrastructure [19]