Mithril

What it does

Mithril is an omnicloud platform that aggregates and orchestrates multi-cloud GPUs, CPUs, and storage into a single interface. It provides transparent pricing, short-term reservations, spot bidding, and a batch SDK/API for asynchronous workloads. The platform supports training, fine-tuning, inference, and batch jobs on NVIDIA A100s, H100s, and H200s with InfiniBand interconnect and high-performance WEKA storage.

Who it is for

Mithril targets AI-native startups, research labs, public companies, and academic institutions that need flexible, cost-effective compute for machine learning workloads. It is designed for teams that want to avoid overprovisioning and negotiate with multiple cloud providers.

Why it matters

Mithril addresses the friction of procuring and managing GPU compute across clouds. By offering algorithmic pricing based on supply and demand, extendable short-term reservations, and spot bidding, it helps users get the best prices without sales negotiations. The platform enables scaling to thousands of GPUs and supports batch inference via a dedicated API, reducing operational overhead.

Launch signal

Mithril launched its omnicloud platform as announced on Hacker News and its blog. The company was previously known as Foundry Technologies. It lists customers including Arc Institute (developers of EVO, a genomic foundation model) and claims trust from AI-native startups, research labs, public companies, and academic institutions.

Brand and naming

The name "Mithril" evokes the fictional precious metal from Tolkien's legendarium, known for being lightweight yet incredibly strong. This aligns with the platform's promise of powerful, flexible compute that is easy to access. The brand positions itself as a "compute sanctuary" and "omnicloud," emphasizing simplicity and multi-cloud aggregation.