AI Infrastructure's Biggest Bottleneck Isn't Hardware — It's Execution

Nebius CTO Danila Shtan outlined the growing list of constraints slowing AI infrastructure expansion during a discussion with Jacob Effron on Unsupervised Learning.
The shortages span nearly every part of the AI buildout: GPUs, CPUs, NVMe drives, network transceivers, data center facilities, power, and even transformers. Some equipment and infrastructure projects face lead times measured in years.
When asked to rank the biggest bottlenecks, Shtan argued that ranking them misses the point. The central constraint, he said, is the ability to execute.
He added that experienced project managers are among the most sought-after hires at AI cloud companies, describing demand for them as the highest he has seen throughout his career.
As AI infrastructure scales, coordinating people, equipment, facilities, and power may prove just as critical as securing the hardware itself.