How they use Nebius

Mithril uses Nebius Token Factory for real-time and batch inference. Its testimonial describes running and optimizing large workloads while retaining the flexibility and transparency of open models. Mithril also integrates Nebius H100 and H200 capacity into its compute marketplace. Foundry rebranded as Mithril in August 2025.

Nebius listed Foundry as a marketplace ecosystem partner in its May 2025 earnings presentation. A July 2025 company post announced access to Nebius H100, H200 and Blackwell infrastructure through Foundry Cloud Platform. Nebius’s July digest describes access to Nebius AI Cloud regions through Mithril Console for customer experimentation, large-scale training and production inference. Mithril’s announcement also describes access through its compute API. Nebius’s launch post names eu-north1, eu-central1 and eu-west1 as the regions offering H100 and H200 instances through custom reservations. Nebius’s partner catalogue describes self-serve access across three European regions, with H100 and H200 capacity or dedicated GB200 NVL72 and HGX B200 clusters available for training and inference. In March 2026, founder Jared Quincy Davis described tools that let Nebius customers turn standard reservations into flexible reservations. Customers can pause reservations to earn when their nodes serve Mithril’s preemptible pools and reclaim them when needed. In a GTC 2026 interview published in April, Davis described a joint product being launched with Nebius that would also allow early termination, upgrades to new compute types and rebates when others use idle instances.

SemiAnalysis reported in September 2026 that Mithril’s marketplace covered three Nebius availability zones with H100 or H200 capacity. It reported a scale-out networking option only for H200. Mithril’s engineering recruitment describes intended work to bring Nebius resources under management through the Mithril API.