Goldman Sachs just put a price tag on the AI buildout, and it reads like a defense budget. The bank’s May 2026 report, titled “Tracking Trillions,” estimates that annual AI-related capital expenditure will climb from $765 billion this year to $1.64 trillion by 2031.
Add it all up and the cumulative tab comes to $7.6 trillion over six years. For perspective, that figure exceeds the GDP of every country on earth except the US and China.
Where the money goes
Goldman’s breakdown splits the spending into three buckets: compute, data centers, and power. Compute infrastructure, meaning the chips and servers that actually run AI workloads, is projected to absorb roughly $5.1 trillion of the total. That’s about two-thirds of every dollar spent.
Data centers, the physical buildings that house all that silicon, are expected to attract $2.1 trillion in investment. Power infrastructure, the grids and generation capacity needed to keep the lights on inside those buildings, rounds out the picture at $358 billion.
The compute number carries an important asterisk. Goldman assumes NVIDIA will capture 75% of that market. If that estimate holds, Jensen Huang’s company would be on the receiving end of roughly $3.8 trillion in revenue over the forecast window.
The hyperscaler arms race
The four largest cloud companies, Meta, Microsoft, Amazon, and Alphabet, are expected to collectively spend around $5.3 trillion in capital expenditure from fiscal years 2025 through 2030. Goldman notes that this figure alone surpasses the GDP of Japan, which has the fourth-largest economy in the world.
Global AI-related investment, including private companies and non-US spending, is estimated to exceed $1 trillion for 2026 alone.
The bottleneck nobody wants to talk about
Goldman’s analysts flagged two risks that could complicate this spending trajectory. The first is power. The $358 billion earmarked for power infrastructure might sound like a lot, but it represents less than 5% of the total AI capex forecast.
The second risk is more fundamental: whether these investments will generate justified returns in the near term. The AI infrastructure buildout is, at its core, a bet that demand for AI services will scale proportionally with the supply of compute.