Foundry's Cloud Platform: A New Addition to GPU Cloud Arena
Introducing Foundry's Cloud Platform
AI infrastructure startup Foundry steps into the competitive GPU cloud market with the launch of its Foundry Cloud Platform (FCP). Founded by former DeepMind researcher, Jared Quincy Davis, this Palo Alto-based AI upstart aims to break through the industry's standards by providing more than just rent-a-GPU cluster services.
A Unique Approach to AI Infrastructure
Stepping away from traditional large clusters used for training, Foundry's strategy involves maintaining a pool of reserved nodes that function as supplementary support for when inevitable technical issues arise. Unlike other competitors who allow these reserved nodes to sit idle, Foundry plans to rent them out to customers as preemptible spot instances at prices lower than those offered by other cloud providers.
Offering Tailored Services
These available instances, along with hosted Kubernetes clusters, disk state saving, and auto-mounting persistent storage, will make it possible to run properly configured inference workloads continuously, even if underlying nodes need to be re-purposed. While Foundry's compute resources are limited compared to competitors like CoreWeave or Lambda, it seems the company's unique orchestration-layer might allow them to potentially offer more with less, by maximizing the usage of their present kit.
Flexible Reservation Options
A notable departure from the usual long-term contracts associated with the sphere, Foundry offers customers the option to reserve compute resources for as little as three hours at a time, marking a significant shift in user-friendly GPU services.