Big Tech's AI Datacenter Spend Rivals Nations' GDPs
The sheer scale of investment pouring into new data center infrastructure by the world’s leading cloud operators now rivals the economic output of entire nations. According to figures compiled by Omdia, cloud giant Amazon’s annual capital expenditure on data centers alone has surpassed $100 billion. This staggering sum is roughly comparable to the entire Gross Domestic Product (GDP) of Costa Rica and significantly exceeds the economic output of Luxembourg or Lithuania.
Other major cloud players are also investing at an unprecedented pace. Google’s capital expenditure for data centers stands at $82 billion, a figure larger than Slovenia’s national economy. Microsoft’s $75 billion investment surpasses Uganda’s GDP, while Meta’s $69 billion outstrips the total economic output of Bahrain. This massive spending spree is not isolated; Omdia estimates that global data center capital expenditure will reach an astounding $657 billion for the entirety of 2025. This represents a near doubling of investment since 2023, when the total stood at $330 billion.
A primary driver behind this explosive growth in infrastructure spending is the insatiable demand for ever-increasing compute power, particularly to fuel the rapid development of artificial intelligence. Companies are betting heavily that these colossal investments in AI infrastructure will eventually yield substantial returns. However, this optimism is tempered by a recent report from McKinsey & Company, which revealed that many corporate executives harbor skepticism about whether the vast sums currently being allocated to AI infrastructure will generate measurable returns on investment in the near future. Adding to this nuanced picture, Meta disclosed in its second-quarter earnings that its profits are primarily driven by conventional machine learning models powering its recommender systems, rather than its highly publicized generative AI projects.
Despite these reservations, data center operators continue to expand rapidly, acting as the “shovel sellers” in the ongoing AI gold rush. Clients persistently demand more AI-capable infrastructure, and providers are eager to oblige. Omdia’s latest Cloud and Data Center Market Snapshot indicates that, for the foreseeable future, orders for AI compute resources continue to outstrip supply. The development of newer and larger AI models, such as the recently introduced GPT-5, is driving intense demand for training capacity. Concurrently, the broader adoption of AI across various sectors is fueling the need for inferencing capabilities. Omdia highlights that a significant portion of the population in developed economies now uses AI in some form, with ChatGPT alone boasting over 700 million users and more than 120 million daily visits.
Delving deeper into this investment, Omdia predicts that IT equipment will remain the largest component of data center costs in the coming years. However, spending on physical infrastructure is projected to grow at an even faster rate. This accelerated growth is necessitated by the urgent need for innovative solutions in power generation, distribution, and thermal management to cool and power the increasingly dense and energy-hungry AI servers. The exponential growth in compute density within data hall racks demands continuous innovation in cooling and power delivery systems. The immense energy requirements of this IT infrastructure are also pushing operators to invest in on-site power generation equipment and explore “microgrid-as-a-service” models, where specialized companies provide on-site electricity generation for data center facilities.
Looking ahead, the pace of data center construction is accelerating dramatically, and the capacity of the largest sites is expanding to unprecedented levels. Multi-gigawatt facilities, some equivalent to the entire current power capacity of a country like Canada, are already in the pipeline. Meta, for instance, has signaled plans for several multi-gigawatt campuses scheduled to come online from 2026. A recent Deloitte Insights report further claims that some sites currently in early planning stages could eventually top an astonishing 5 gigawatts, underscoring the monumental scale of the digital infrastructure being built to power the future of AI.