# AI Infrastructure Spending Is Modeled at $7.6 Trillion Through 2031

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/24280
Published: 2026-09-25T22:14:19.000Z
Updated: 2026-09-25T22:14:19.000Z
Section: Technology

> Annual spending is estimated at $765 billion in 2026 and $1.6 trillion in 2031, with chip replacement cycles driving the capital needs.

The artificial-intelligence infrastructure boom is entering a more demanding phase as investors assess whether future revenue can support the enormous cost of chips, data centers and power systems.

A scenario analysis estimates that the AI infrastructure build-out could require about $7.6 trillion in cumulative capital expenditure from 2026 through 2031. The estimate includes compute, data centers and power, with annual spending modeled at approximately $765 billion in 2026 and $1.6 trillion in 2031.

The figures are scenario estimates showing how infrastructure assumptions affect the amount of capital required, not a forecast.

The most influential variable is the useful life of AI chips, estimated at four to six years, compared with roughly 20 years for data centers and at least 25 years for much power infrastructure. A faster replacement cycle for chips would sharply increase the amount of capital needed across the industry.

The analysis also assumes NVIDIA accounts for 75% of total compute spending. It models data-center construction costs of $15 million per megawatt, new-power costs of $2,500 per kilowatt and a power-usage-effectiveness ratio of 1.2.

That spending is arriving before the full economic return from AI is clear. The largest hyperscalers spent $106 billion on capital expenditure in the third quarter of 2025, including AI and non-AI spending, up 75% from a year earlier. The consensus estimate for hyperscaler capital expenditure in 2026 was $527 billion.

The scale is approaching previous technology investment cycles. AI capital expenditure recently equaled about 0.8% of gross domestic product, while peaks during earlier major technology investment cycles reached at least 1.5%. Hyperscaler capital expenditure would need to reach approximately $700 billion in 2026 to match the intensity of the late-1990s telecommunications build-out.

The market’s focus is therefore moving beyond spending totals. Companies will need to show that AI infrastructure is generating durable revenue, while investors will have to judge whether those returns arrive before hardware becomes obsolete or power and construction constraints delay deployment.

An eventual slowdown in capital-expenditure growth could pose a risk to the valuations of companies benefiting from the infrastructure cycle. The total investment requirement depends on assumptions that could change as chip designs, data-center architecture, power availability and AI demand evolve.