United States
United Kingdom
Canada
Australia
New Zealand
IrelandAI's cooling problem doesn't end at the chip
CNBC explains why recycling water inside a data center can still leave a thirsty, power-hungry job outside.
CNBC
Story of Why this story? 940,382 views so far, and here is what is inside it.
The Truth About How Data Centers Use Water And Power To Keep Cool
Things we can learn from the video
Press to jump
- 2:22Closed-loop liquid cooling recirculates coolant, but the facility must still release the collected heat outdoors.
- 4:41Equinix's Lindsay Schulz shows the plumbing that delivers chilled water to the racks.
- 7:11Dry, evaporative and hybrid cooling illustrate the trade-off between water consumption and electricity demand.
- 8:50Memphis and xAI's Colossus illustrate community concerns about data centers' water requirements.
- 9:55DeepX demonstrates cooler-running hardware with a chip that operates without melting butter.
- 10:39SK Hynix presents memory technology designed with integrated cooling elements to help remove heat.
Inside an Equinix data center in San Jose, Katie Tarasov follows the plumbing that keeps racks of AI chips cool. Moving heat away from the processors is only the first part of the job. It still has to leave the building.
That distinction is the heart of CNBC's "The Truth About How Data Centers Use Water And Power To Keep Cool," which has drawn around 940,000 views in four days. A closed-loop system can circulate cooling liquid repeatedly rather than constantly drawing fresh water. But calling it closed-loop doesn't answer how the facility gets rid of all that collected heat.
Tarasov separates the process into its parts: liquid cooling at the chips, chilled-water infrastructure serving the racks, and equipment that releases heat outdoors. The report's animated comparison explains the trade-off. Dry cooling avoids evaporating water but can require more electricity, particularly in hot conditions. Evaporative cooling uses water to carry heat away, losing that water into the air. Hybrid systems combine approaches.
It is a practical distinction for communities questioning what new AI facilities will mean for local water supplies and utility bills. The report examines those concerns in Memphis, home to xAI's Colossus data center, rather than treating the cooling system inside a rack as the whole resource story.
The final section turns to hardware that produces less heat or makes it easier to remove. DeepX demonstrates a chip operating without melting butter, while SK Hynix presents memory technology with integrated cooling elements. After the pipes and cooling towers, the butter supplies a much smaller-scale demonstration of the same problem.
Published on Trari
Where next?
Is this your channel?For the channel's owners and team: a ready picture and text for your audience
Share the story with your audience. Everything is ready:
For an Instagram story or a WhatsApp status
For a community post or your channel's description
- Channel
- CNBC
- Subscribers
- 4,180,000
- Channel views, last 7 days
- 3,173,614
- Channel views, pace per day
- 678,915 (on a normal day: 459,331)
- Video published
- November 5, 2026
- Video views so far
- 940,382 (4 days online)
- Video views, pace per day
- 361,472 (on a normal day: 195,488)
The numbers
CNBC's cooling report has approached a million views since publication. Here is how its audience and the channel's wider activity compare.
💡 Test your knowledge
What does GPU stand for?
GPU stands for graphics processing unit, a type of processor also widely used for AI workloads.