The world of artificial intelligence is changing and control is shifting to the physical infrastructure

“Artificial intelligence is written in code, but its economics are determined in the field,” Beiler says. “As the deployment grows, the bottlenecks are no longer just in software or chips, but in electricity, cooling and networks.” According to her, if in the past organizations were content with offices and small server rooms, today large facilities are being built designed to operate and cool thousands of servers under high load.

This means that the demand for AI may be born in the digital world, but the ability to deliver it depends on very tangible factors. You need land to build facilities on, you need a connection to the electricity grid, you need advanced cooling systems, and all of this competes for resources with other uses in the economy.

At the same time, the risks also look different. Projects can be delayed due to infrastructure constraints, regulation or local opposition. “You can’t just look at the demand,” Beiler says. “We also need to understand if it is really possible to build, connect and operate the systems on a large scale.”

In the end, according to Beiler, the one who will lead the field of artificial intelligence will not only be the one who develops the most advanced algorithms, but also the one who will manage to establish and operate the infrastructure behind them. As the domain grows, the advantage gradually shifts from code to the ability to manage power, cooling, and networks at scale.