⚡︎ Independent, reader-supported & ad-free · Follow the megawatts in all 50 states ⚡︎ Support Us
Watch · Our Take

Some data centers are submerging their servers in liquid πŸ‘€πŸ’§ #trendingshorts #ai #datacenter #tech

The Rundown AI · 11 months ago

The short version: Rowan Cheung's short on boiling-fluid immersion cooling is a slick tech demo, but the real story is what 5,000-watt chips mean for local grids and water systems.

Our Take

Credit to Rowan Cheung for a clean explainer on two-phase immersion cooling β€” dunking whole servers in dielectric fluid that boils at 43Β°C is genuinely wild engineering, and it's a real efficiency gain over brute-force chillers. But every one of these cooling breakthroughs exists because chip wattage is climbing so fast that air alone can't keep up anymore. That's not a footnote, it's the headline: the industry is racing to cool a heat problem it created by packing ever more power-hungry silicon into the same footprint, and someone downstream β€” ratepayers, aquifer users, taxpayers offering the incentives β€” is footing the bill for that arms race.

Exotic fluids and ocean-submerged pods make for great B-roll, but they don't change the fundamental math: these facilities still pull enormous amounts of electricity off local grids, and in plenty of cases still lean on municipal water for evaporative cooling loops even when the chip-level tech goes 'dry.' The efficiency gains rarely get passed back to the communities hosting these buildings in the form of lower bills β€” they mostly get reinvested into building the next, bigger cluster. Slick cooling tech is a marketing win for hyperscalers; it's not the same as public accountability for what a gigawatt-scale campus does to a regional grid.

If there's a data center proposed or under construction near you, don't wait for a press release to tell you it's efficient β€” check the map for what's actually been approved in your area, and see what levers residents have pulled elsewhere on take-action.

This is GridWatch the USA’s original commentary. The video above is the work of The Rundown AI, published on YouTube — full credit to the creator.