Neocloud Firms: The Infrastructure Layer of the AI Boom

The companies building data centres for AI training are becoming as strategically important as the AI companies themselves.

Published: 29 July 2026 Category: AI Infrastructure Sources: ChosunBiz, industry analysis

The Layer Below

The AI revolution runs on data centres. Not ordinary data centres — specialised facilities with massive power requirements, liquid cooling systems, and tens of thousands of interconnected GPUs. Building these facilities requires billions of dollars, years of planning, and favourable regulatory environments. The companies that do this building — the "neocloud" firms — are becoming as strategically important as the AI labs that train the models.

The term "neocloud" distinguishes these companies from traditional cloud providers like AWS, Azure, and Google Cloud. Neocloud firms — CoreWeave, Lambda Labs, Crusoe, and others — specialise in GPU-intensive workloads. They do not offer general-purpose cloud services. They offer AI compute, and they are building data centres specifically designed for AI training at scale.

The Economics

The economics are extraordinary. A single data centre optimised for AI training can cost $2-5 billion to build. The power requirements — hundreds of megawatts — strain local electrical grids. The cooling requirements challenge conventional designs. And the chips inside depreciate rapidly as newer, more efficient GPUs become available.

Despite these costs, demand exceeds supply. Anthropic's reported $10 billion resource lease with Meta is partly directed at neocloud capacity. OpenAI's training runs consume everything available. Google, Microsoft, and Amazon are building their own facilities but still depend on neocloud partners for overflow capacity.

The Strategic Position

Neocloud firms occupy a crucial position in the AI value chain. They are not model builders, but model builders cannot exist without them. They are not chip manufacturers, but chip manufacturers depend on them for demand. They are the railroads of the AI age — unglamorous, capital-intensive, and absolutely essential.

The strategic implications are significant. Neocloud capacity is concentrated in the US, with smaller clusters in Europe and the Middle East. China is building domestic capacity but faces chip access limitations. Countries that want AI capability — which is increasingly every country — must secure neocloud partnerships or build their own facilities.

The Verdict

Neocloud firms are the most underreported story in AI. They receive less attention than model releases and less funding than frontier labs, but their capacity constraints determine what AI research is possible. A lab with a brilliant architecture cannot train it without compute. A company with a compelling product cannot scale it without inference capacity.

The AI boom will continue as long as neocloud capacity expands. The boom will slow — or prices will rise dramatically — if expansion cannot keep pace with demand. The companies building the infrastructure are not the headline names, but they may be the most important.

When historians write about the AI era, they will spend chapters on the models and the companies that built them. They should also spend chapters on the data centres, the power grids, and the cooling systems. The intelligence revolution is built on concrete, steel, and electricity. The neocloud firms are the ones pouring the foundations.

Read next: The AI Circus: July 2026 Roundup