Goldman Sachs Just Mapped Where $7.6 Trillion Goes — And It Confirms the Map of AI

Infrastructure Analysis — Goldman Sachs just released the most precise projection of AI capital expenditure ever published. $7.6 trillion over six years. The breakdown maps directly onto the first three layers of the Map of AI.

The Numbers

Goldman projects $765 billion in AI capex in 2026 alone, growing to $1.6 trillion annually by 2031. Cumulative 2026-2031:

Map of AI: Where the Money Flows

Goldman’s breakdown maps directly to Layers 1-3 of the Map of AI:

The Data Center Evolution

Goldman’s data center specification chart shows a 100x power density increase in four generations:

The $1.76 Trillion Swing Factor

Goldman identifies silicon useful life as the single biggest variable in the entire model. The difference between a 3-year and 7-year GPU replacement cycle is $1.76 trillion in depreciation:

The Product Overhang Read

This is the Product Overhang Doctrine at infrastructure scale. Goldman’s model is not a prediction of whether AI spending happens — it is a model of the minimum physical capital required to deploy infrastructure that has already been contracted, already announced, and is already under construction.

The overhang is not theoretical. It is $7.6 trillion of committed capital flowing through three layers over six years. The companies standing in the path of that capital — Nvidia (compute), Vertiv (cooling), and nuclear operators like Vistra (power) — are not making bets on AI. They are collecting tolls on infrastructure that is already being built.

In the Map of AI, Layers 1-3 are the foundation that everything else sits on. Goldman just quantified the foundation: $7.6 trillion. The question is not whether it gets spent. The question is who captures margin at each layer.

At 75% gross margin on data center GPUs, Nvidia is collecting the largest infrastructure toll in the history of technology. $3.8 trillion through one company’s products over six years. That is not a business. It is a tax on the future.

Frameworks:
The Map of AI — 9 Layers of the AI Economy
Product Overhang Doctrine
Beyond the Nvidia Tax

Source: Goldman Sachs Global Investment Research, NVIDIA projections (March 2026). Data assumes straight-line depreciation and no terminal value for GPUs.