AI pushes data centers to rethink power and cooling
The rapid rise of AI is surpassing what traditional data centers can handle, putting strain on everything from power supply and cooling systems to cabling infrastructure. GPU clusters now use so much electricity—often over 50 kW per rack—that standard cooling isn't enough, risking performance slowdowns. Data centers are responding with new cabling, modular designs, and higher-capacity cooling, but upgrades take time and money.
- AI GPU racks now exceed 50 kW per cabinet
- Standard cooling systems can't handle new thermal loads
- Grid power capacity is a growing bottleneck for expansion
- Data centers require modular cabling and faster deployment
- Front-end infrastructure sees less impact than power and cooling
Sources covering this
More in Enterprise
Windows 11 23H2 Enterprise support ends in 30 days
Microsoft just announced that Windows 11 version 23H2 Enterprise, Education, and IoT Enterprise will stop getting security updates in…
Developer releases open-source clones of Office apps
Brandon Thomas, creator of open-source alternatives to Adobe's creative software, has released new free clones of Microsoft Word, Excel,…
Enterprises shift focus from AI experimentation to measurable returns
A report from consultancy ISG finds that while enterprise investment in AI remains strong, companies are prioritizing clear returns and…
Salesforce details Spark resource optimization for 8M daily jobs
Salesforce's Data 360 team runs about eight million Apache Spark jobs daily and found most workloads use more computing resources than…