Data Center Infrastructure Podcast

Steve Wood ran 2 MW of Bitcoin miners in the basement of a closed paper mill and piped their heat into a 600,000 square foot building. The mill stopped buying 200,000 gallons of propane a winter. Now he’s building AI data centers on the same idea.
Eric Sonner and Lance Teel talk with the Idaho farm kid turned Keystone founder about auditing at Coopers & Lybrand, 13 years building plants for Pyrotek, and the 2016 Bitcoin pitch his paper mill turned down. Then they get into industrial symbiosis. Liquid cooling captures about 75 percent of a data center’s heat today, so a 100 MW site has roughly 70 MW to give to greenhouses, fish hatcheries or a town pool. Steve explains why you ask the community what it wants before you build, and what he’s planning on 200 MW at Gigawatt Land with Gordon Baier.
Plus karaoke in Japan, career advice for the next generation of builders, and Steve’s charity pick, Tip of the Spear.

Before a data center runs a single AI workload, somebody has to try to break it. Mitch Dec is that somebody.
Eric and Lance sit down with Mitch Dec, VP of Commercial Services at Data Airflow, who leads commissioning on some of the largest AI data centers in America. Mitch walks through commissioning from Level 0 to Level 5, from factory testing to cutting power on purpose to prove the backup paths work. He also explains how to test a direct-to-chip AI hall before any servers arrive.
The water discussion covers the energy and water tradeoff, how new refrigerants let closed loop systems match open loop efficiency, why the glycol in these systems is food grade and not a forever chemical, and the payback that convinces investors. Plus how Mitch and Eric went from a phone call to a signed proposal in under 36 hours, a 500 person town hall in Appalachia, and why your power bill may have gone up years before the data center arrived.
Mitch’s advice closes the show. Every degree has a place in this industry, and so do the trades.
If you use your phone, you use a data center. Get to know what’s good and what’s bad.
Watch on YouTube. https://youtu.be/N7zhj4MhEOg
More at https://dcipodcast.com

Training AI happens in giant campuses. Inference, the next wave, needs compute close to the people using it, and many states don’t have the network to support it.
Recorded in Austin, Eric and Lance sit down with Bryan Lubin of Moonshot Energy, a former private banker and chief credit officer who now underwrites power, land, and compute. Bryan explains the infranet. Moonshot and Connected Nation Internet Exchange Points are deploying modular AI pods next to internet exchange points at university campuses, starting at 25 sites and scaling to 125. In states without a local exchange, traffic travels hundreds of miles before it’s answered. For self-driving trucks and robotics, that delay is a serious problem.
Plus what modular should mean (precast concrete pods, not metal boxes), closed loop cooling built with Data Airflow, sites under five megawatts, a six to eight month timeline from dirt to data, and why OEMs now want proof a site is ready before they ship servers.
Bryan’s advice closes the show. Be who you are, stay focused, and expect to fail. That’s how you learn.
If you use your phone, you use a data center. Get to know what’s good and what’s bad.
Watch on YouTube. https://youtu.be/zArAps8I-y4
More at https://dcipodcast.com

AI data centers swing their power draw by tens of megawatts in seconds, and the grid wasn't built for it. David Holmes, Global Industries CTO for Energy at Dell Technologies, sits down with Eric Sonner and Lance Teel in Austin to talk about what it takes to fix that.
David covers why AI loads are so hard on generators and transformers, how long-duration flow batteries buffer a data center from the grid, and where stranded power is hiding at substations and in half-empty buildings. He lays out his nine-box framework for deciding where AI workloads should run, explains sovereign AI, and makes the case that PUE is only half the scorecard. Tokens per watt is the other half.
Then he gets into Concept Astro, Dell's proof of concept for a grid-aware AI factory, and the bigger idea behind it, a grid that's aware of its data centers. The conversation closes on ERCOT's Batch Zero process, what makes a good grid citizen, and why physical AI is the next shift to watch.

Concept Astro white paper https://www.delltechnologies.com/asset/en-us/solutions/business-solutions/industry-market/concept-astro-whitepaper.pdf
David Holmes on LinkedIn https://www.linkedin.com/in/energygm/

Hosted by Eric Sonner, CEO of Data Airflow, and Lance Teel of Atlas Energy.

A normal data center battery buys you 10 minutes. This one runs for 10 hours, can't catch fire, and can turn a data center into a power plant for its neighbors.
Recording in Dallas, Eric and Lance sit down with Jeff Goldberg, Chief Development Officer and co-founder of TerraFlow Energy, maker of vanadium flow batteries that work as a power hub. Every source a site has flows through the battery, the data center draws from the battery, and the chemistry absorbs AI load volatility in milliseconds while protecting equipment on both sides.
Jeff breaks down the numbers. 250 megawatt hours on under an acre, an electrolyte that's 75 percent water with no lithium fire risk, a 20 to 30 year life filled once, and simultaneous charge and discharge. With 10 plus hours of storage, a data center can carry itself through peak hours and push power back to the grid. In Jeff's words, you turn a data center into a peaker plant. The vampire becomes a blood donor.
Plus the origin story, a real estate developer, his metal fabricator tenant Jon Parrella, and two words that started it all. I'm in. Five founders, 59 employees in two years, and a factory opening in Katy, Texas. Jeff's advice closes the show. Get an internship. The industry needs you.
If you use your phone, you use a data center. Get to know what's good and what's bad.
Watch on YouTube. ⁠⁠https://youtu.be/DV-diphbD98⁠⁠
More at ⁠⁠https://dcipodcast.com