Not like conventional public clouds, these approaches are sometimes constructed from the bottom as much as deal with the distinctive calls for of recent AI infrastructure. This implies high-density GPU configurations, liquid cooling methods, and energy-efficient designs. Extra importantly, they allow corporations to shift to possession or shared useful resource fashions that scale back prices in the long run.
Betting on the unsuitable enterprise mannequin
Public cloud suppliers are positioning themselves because the pure house for constructing and deploying AI workloads. Naturally, the deal with AWS re:Invent 2024 It was once more about generative AI and the way the AWS cloud helps generative AI options. Early-stage AI experimentation and pilots have pushed a near-term surge in cloud income as organizations flock to hyperscalers to coach complicated fashions and rapidly check new use instances.
Coaching AI fashions on public cloud infrastructure is one factor; implementing these methods at scale is one other. When betting on AI, public cloud suppliers rely closely on consumption-based pricing fashions. Sure, it is easy to construct sources within the cloud, however the cracks on this mannequin are more and more troublesome to disregard. As corporations transfer from experimentation to manufacturing, long-term GPU-heavy AI workloads don’t translate into price efficiencies.
