About $700 billion has been poured into the physical infrastructure that runs modern artificial intelligence, a long-form explainer finds. The spending is concentrated in five layers the author maps as Energy, Chips, Cloud, Models and Applications, and much of the capital has gone into facilities and equipment rather than consumer-facing products. In plain terms, the money is buying power plants, specialised processors, vast warehouses of compute and the data and training work that make models, with new data-centre campuses under construction in places such as Texas, Iowa and Hyderabad. The author says follow-up sections will attach numbers to quantify energy use, capex and other inputs.

Those are large sums and they alter where value sits inside the tech economy. The long-form explainer argues that roughly $700 billion has been deployed into the lower parts of an AI stack made up of five layers: Energy, Chips, Cloud, Models and Applications. In plain terms, the money is buying power plants, specialised processors, vast warehouses of compute, the datasets and training work that make models, and finally the apps that ordinary users see.

Where the capital is actually going

The account stresses that much of the capital has flowed into facilities and equipment rather than end-user products. At the chips layer, specialised processors are described as the scarce input doing the heavy compute work, and they're distinct from the general-purpose laptop chips most people know. That scarcity helps explain why capital has concentrated on chip fabrication, advanced packaging and sourcing of power at scale.

Practical examples are listed. Chips and new packaging technologies are being manufactured, large-scale cooling and power infrastructure is being installed, and new data-centre campuses are under construction in places including Texas, Iowa and Hyderabad. The explainer paints the cloud layer as "massive warehouses" of interconnected hardware and networking that host models and serve applications at scale. Those warehouses aren't light investments. They require years of engineering, steady power, and unusually large capital outlays.

Where money is captured depends on the layer. Cloud providers, chipmakers and energy suppliers each grab different slices of value, and the flow of funds can move both up and down the stack. That means a company supplying power or specialised packaging can capture rents just as a model owner or an app maker can.

Models, apps and the misalignment of attention

The models layer is presented as the intellectual property and software trained on large datasets. The work here is less visible, but crucial. Model training consumes both specialised chips and a great deal of electricity. Above that sits the applications layer, where consumers and businesses interact with AI through chatbots, search, fraud detection and other services.

Mastheads

That's the layer most people encounter, and it draws headlines and user attention.

The through-line of the explainer is a misalignment of visibility and value. Consumer-facing products attract attention while the real economic rents and bottlenecks sit in commodity-like infrastructure that requires heavy, long-lived capital investment. The piece points to investments that look more like utilities than software: power plants, specialised manufacturing lines and sprawling data campuses.

The author reports spending a year researching the supply chain and promises to attach numbers across each claim in the full explainer. That signals an intent to quantify energy use, capex and other inputs in subsequent sections. For now, the narrative provides a map of where capital has gone and why the industry’s economic centre of gravity has shifted away from visible apps toward invisible hardware and facilities.

There are concrete tensions embedded in that map. Specialised processors aren't interchangeable with general-purpose silicon, and the need for scale makes power availability a strategic bottleneck.

Building packaging lines and data campuses takes time. The result is a form of industrial concentration: whoever controls manufacturing, advanced packaging or reliable power gains leverage over the rest of the stack.

That concentration also shapes where profits and constraints appear. An app developer may be constrained by chip shortages or by the cost of colocating on the right cloud provider. At the same time, chipmakers and energy suppliers can extract value from their position in the chain. The explainer frames those dynamics without assigning market-share numbers or energy statistics, because the author says those figures will follow in the quantifying sections.

Readers should note the limits of the material available. All of the core claims in the piece come from a single long-form source. The $700 billion total, the list of construction sites, and the Davos quote are single-sourced in the material supplied. The account doesn't cite independent central-bank figures, industry statistics, or third-party measures of energy demand and capex within the bundle provided for this report.

Even so, the piece adds clarity to a debate often framed entirely around consumer apps. By mapping the five layers it highlights the heavy, industrial nature of the buildout beneath the chat windows and app icons.

Related Articles

The narrative is anchored to that five-layer map, and the author says follow-up sections will attach numbers to quantify each layer.

This article was created with AI assistance.