Introducing Mithril: The AI omnicloud

Today, we’re excited to share a new chapter in our journey: Foundry is now Mithril.

As Foundry, we built fundamental mechanisms to maximize the availability, reliability, and price performance of distributed AI compute by aggregating and orchestrating idle infrastructure. This work has helped practitioners at Arc Institute, Cursor, and LG AI Research push the frontiers of AI in an era defined by GPU scarcity.

Now, as we enter a new chapter, those foundations are serving as the base for what we are building as Mithril: the AI omnicloud.

The omnicloud: aggregation + abstraction

Although global compute capacity increases daily, the economic gains from greater availability do not flow naturally to you — the practitioner building with it. Too often, teams remain stuck in legacy procurement models, forced to negotiate with multiple providers for competitive rates.

The omnicloud layer removes the market inefficiencies causing friction between growing supply and the practitioners who should benefit from it. It does so through two key functions: aggregation and abstraction.

Aggregation: a fluid market

Mithril aggregates multi‑provider cloud compute into a fluid market that you can access through a single platform with transparent, real‑time pricing. As more compute is aggregated, you benefit directly as costs adjust dynamically to reflect increased supply. You can provision capacity through instant, hourly reservations for guaranteed access, or tap into spot compute to take advantage of the most cost‑efficient resources available.

Abstraction: simple, price-performant execution

On top of this market, we’re introducing new abstractions that make it effortless to harness omnicloud compute. The first are our Batch Inference API and Flow, our batch job SDK. The Batch Inference API follows a standard, token‑based pricing construct and is purpose‑built to scale large batch jobs seamlessly across omnicloud compute. With it, you can process 500M+ tokens per day by default, with expanded rate limits for custom use cases. Flow provides tools to build similar optimizations for general‑purpose ML workloads.

Mithril omnicloud compute is generally available today for AI teams and research labs. The Mithril Batch Inference API and Flow SDK are available in beta to all Mithril users.

For instant access to Mithril, sign up with your work email.

Looking ahead

To our community: thank you for your trust and feedback. You have helped us shape every step of this journey. As Mithril, we are committed to solving the root node problems of AI infrastructure today while relentlessly pushing to make tomorrow’s infrastructure even better.

Our mission remains the same: orchestrate the world’s compute to render it universally accessible. Our work under the name Foundry was the first key step in that mission. 

Mithril represents the work ahead of us, and we can’t wait to get started.

— The Mithril team