Maybe, but I would suggest that larger factors arise from utilization and overhead. These are going to be the competitive advantages that AI operators need to win the arms race.

TPUs are more efficient at LLMs because they do more TOPS/KWh.

"OpenTPU: Open-Source Reimplementation of Google Tensor Processing Unit (TPU)" (2025) https://news.ycombinator.com/item?id=44111452

"A Comparison of the Cerebras Wafer-Scale Integration Technology with Nvidia GPU-based Systems for Artificial Intelligence" (2025) https://arxiv.org/html/2503.11698v1

From https://news.ycombinator.com/item?id=44648575 :

> "Next-generation datacenters consume zero water for cooling" (2024) https://news.ycombinator.com/item?id=42376406

>> this design will avoid the need for more than 125 million liters of water per year per datacenter

"Microsoft’s Datacenter Community Pledge: To build and operate digital infrastructure that addresses societal challenges and creates benefits for communities" (2024-06) https://blogs.microsoft.com/blog/2024/06/02/microsofts-datac... :

> We will design and operate our datacenters to support society’s climate goals and become carbon negative, water positive and zero waste before 2030. [...]

> By 2025, we will procure 100% renewable energy on a global scale, both significantly expanding and decarbonizing local electricity grids.

> Our datacenter designs are more water efficient than traditional enterprise datacenters, and our plan by 2030 is to replenish more water than we consume locally.

Here's this about CNT cooling:

"Cyberpower begins selling desktop PCs with carbon nanotube CPU cooling" (2025) https://news.ycombinator.com/item?id=44899495

"A carbon-nanotube-based tensor processing unit" (2024) https://news.ycombinator.com/item?id=41322070

Graphene semiconductors should be at least 10X more energy efficient; but how much less water would graphene-based chips waste?