AI Infrastructure: Physical Barriers to a Digital World
In this paper, we explore the physical barriers that could have important implications for inflation, corporate margins and infrastructure completion.
The debate around AI rages on. Who will be able to monetize AI mass adoption? Are hyperscaler and semiconductor company valuations justified? While these are crucial questions, we believe another concern is just as pressing, yet often overlooked: Can the infrastructure required to support mass AI adoption be delivered on time and in a cost-effective manner?
Power Infrastructure: A Significant Component of AI Economics
The popular narrative around artificial intelligence (AI) is that it will require vast amounts of electricity, possibly outpacing energy production. While that remains a possibility, a more pressing question is whether the electric grid can deliver energy to where data centers are being built. In our view, the primary bottleneck is more likely to be transmission infrastructure, or the grid, than electricity generation itself.
Transmission projects typically require seven to ten years to permit and construct, whereas modern hyperscale data centers can be completed in as little as 18 to 24 months. As a result, AI infrastructure is being deployed roughly five times faster than the electricity network needed to support it.1
1. Source: Belfer Center, Harvard Kennedy School. As of 2026.
26-5809682