Neocloud for AI Inference
"These units are expected to utilize 40,000 GPUs, providing a staggering 3.6 exaFLOPS of compute power. By 2030, the planned network is aimed to amass over 400,000 GPUs, equating to 36 exaFLOPS —comparable to five traditional hyperscale data centers.
"The company’s footprint will span multiple countries, with each Policloud unit constructed to be modular and easily deployable in less than five months, a far cry from the typical 24-month build time seen in standard facilities. This architecture allows Antimatter to respond swiftly to market demands while maintaining a lower overhead."
Comments
Post a Comment
Empathy recommended