Data centers are quietly rewriting the rules of construction, but not everything you've heard about them is true.

In this episode of The Punch List, hosts Jon Wright, TK, and Kanav Hasija bust the biggest myth about data center water use, break down why smaller AI models trained on proprietary data might beat Claude and OpenAI at real-world tasks, and unpack what HUD's new AI grant means for the future of plan review.

Along the way, the crew talks CapEx forecasts, the historical parallel between AI and the steam engine, and why AI might be the first technology that actually meets construction professionals where they already are, instead of forcing them to change how they work.

Highlights:
(00:00) Introduction
(01:15) AI spend shifts from proprietary to open source models
(02:45) The Bridgewater and Thinking Machines experiments
(06:53) Vertical intelligence vs general intelligence
(11:14) Data center CapEx forecasts and construction demand
(14:23) AI as a general purpose technology, lessons from the steam engine
(18:49) Busting the data center water use myth
(27:43) HUD's grant for AI-powered plan review
(31:36) Why AI fits construction better than SaaS ever did

Connect with us:

• Connect with Tanmaya Kala on LinkedIn: https://www.linkedin.com/in/tanmaya-kala-pe-3324a13/
• Connect with Jon Wright on LinkedIn: https://www.linkedin.com/in/jonpwright/
• Connect with Kanav Hasija on LinkedIn: https://www.linkedin.com/in/kanavhasija/

Resources:

• Coinbase CEO on AI spending cut by half at his org: https://x.com/brian_armstrong/status/2070670644577280109
• Custom trained models in enterprise brings more accuracy at way lower costs: https://thinkingmachines.ai/news/learning-to-replicate-expert-judgment-in-financial-tasks/