The AI Infrastructure Arms Race: Can Big Tech's $750 Billion Spending Be Sustainable?

The massive capital expenditure (capex) plans of major technology companies have emerged as one of the most significant stories of this year. Google, Meta, Amazon, and Microsoft are all investing heavily to secure leadership positions in the race to build the infrastructure that will drive the artificial intelligence (AI) revolution. The total capital spending by these four tech giants is expected to reach $750 billion this year, equivalent to approximately half of the United Kingdom's annual government expenditure. This figure far exceeds the budgets that the quartet of high-tech companies had previously anticipated, and is projected to increase even further next year.

Capital Spending Outpaces Profit Growth

Shareholders support this plan, to a certain extent. Since 2023, the average stock price of the four companies has doubled. However, this growth has not kept pace with the average quarterly capex budget, which has increased fourfold during the same period. This raises questions about the sustainability of such spending levels.

Physical and Financial Barriers

Trillion-dollar companies cannot escape limitations when expanding their computational capabilities. First, due to physical barriers such as chip supply and energy/water infrastructure, with water resources beginning to face practical constraints in some developed regions. Second, due to enormous construction costs, as most AI projects have yet to reach break-even points, and there isn't sufficient cash flow from other sources to offset these expenses.

Google's parent company, Alphabet, has raised $85 billion individually through debt in the past year. They plan to raise an additional $80 billion through equity in the coming months—an unprecedented funding round that is not something they can do indefinitely.

Rapid Obsolescence - Overlooked Maintenance Costs

Most attention has focused on building data centers. But there's another significant factor, and one at risk of being overlooked—the maintenance costs. The expense of keeping AI operational after infrastructure is established will be substantial.

Data center servers typically have a lifespan of 3 to 6 years before requiring replacement. With the pace of innovation and computational intensity required for AI, this figure is likely to trend toward the lower end for hyperscale providers. Equipment inside AI data centers accounts for up to two-thirds of construction costs. When adding replacement costs to capex projections for the coming years, everything begins to look frighteningly expensive.

Data Center Asset Lifespan Projections

Last year, Amazon reduced the estimated useful life of its data center assets from 6 years to 5 years, a move the company said was "due to the increasing pace of technological development, particularly in the fields of artificial intelligence and machine learning."

To date, Meta, Microsoft, and Alphabet have not followed suit, still maintaining 6 years, but it seems only a matter of time before they too must concede and reduce this number further, pushing depreciation costs even higher.

Financial Overview

Future Challenges and Considerations

Something must change—sooner or later. Companies are facing a trade-off between maintaining leadership in the AI race and ensuring financial sustainability. Accelerating equipment replacement and reducing asset lifespans will hasten depreciation rates, affecting financial reports and long-term profitability.

Meanwhile, growing energy and water demands for AI data centers pose infrastructure challenges, particularly in regions already facing resource shortages.

The big question is whether companies can sustain this level of investment while waiting for AI projects to generate profits, or if they must slow down to rebalance their balance sheets. The answer will shape not only the future of these companies but the entire AI technology industry.

The truth is, the AI race is becoming more expensive than anyone might have predicted, and the real costs may have only just begun.