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

As the world of artificial intelligence transitions into an industrial phase, Nvidia's valuation has soared to an astonishing $5 trillion. This valuation is deeply interconnected with the burgeoning demand for AI data centers, which are now evaluated by their computing capacity in gigawatts rather than traditional metrics like square footage or server counts. Recent insights reveal that one gigawatt of AI data center capacity is estimated to cost around $35 billion. This figure might seem staggering, yet it underscores the new economic landscape of AI technology. Analysts from TD Cowen have drawn parallels between this measure of capacity and the output of a nuclear reactor, highlighting the immense energy demands of next-generation AI facilities such as xAI's Colossus 2 and Meta's Prometheus. A significant portion of the costs associated with these data centers is attributed to GPUs, which account for approximately 39% of total spending. Nvidia's cutting-edge chips, including the GB200 series, are at the forefront of this expenditure. With gross profit margins exceeding 70%, Nvidia captures nearly a third of the total AI data center spending as profit, solidifying its position as a dominant player in the market. Furthermore, each gigawatt translates to over a million GPU dies, with Nvidia's foundry partner, TSMC, earning around $1.3 billion per gigawatt for manufacturing the essential components. Competing chipmakers like AMD and Intel are striving to keep pace, while major cloud service providers such as Google and Microsoft are investing heavily in custom AI chips to enhance their systems. In addition to GPUs, networking infrastructure plays a crucial role in the overall cost structure of these facilities. Analysts estimate that about 13% of the expenses are allocated to networking equipment, benefiting companies like Arista Networks and Broadcom. The physical infrastructure surrounding the compute units, including power systems and cooling facilities, constitutes another significant expense. Power distribution alone accounts for nearly 10% of the total costs, while thermal management—both air and liquid cooling—represents around 4%. Despite the hefty initial investments, the operational costs of running a gigawatt AI data center are relatively low, with annual electricity costs around $1.3 billion. Interestingly, personnel expenses are minimal, with just a handful of staff managing these vast facilities. As the demand for reliable and scalable power surges, companies like Siemens Energy and GE Vernova are experiencing increased orders for turbines and grid infrastructure, indicating a race among hyperscalers to secure the energy necessary for future growth.