Durantic has officially launched, positioning itself as a managed infrastructure layer for AI workloads across diverse GPU fleets. Founded by engineers with backgrounds at Meta and Hudson River Trading, the company provides solutions for businesses that have computing resources but face operational complexities in colocated, leased, and hybrid environments.
As AI infrastructure becomes more fragmented, the operational layer is increasingly seen as a bottleneck. Current challenges include managing mixed GPU generations and running Kubernetes alongside Slurm, complicating workload management across hyperscalers and GPU clouds. Ivan Diachenko, founder and CEO of Durantic, pointed out, "AI companies are being forced to become infrastructure operators, and most don't want to be." This highlights the urgent need for solutions that simplify operational demands.
Durantic's strategy merges software development with infrastructure management, aiming to provide operational efficiency at scale. The company initiates its model with a 30-day pilot on a defined fleet slice, transitioning into monthly fees based on the number of GPUs and servers used. This approach allows clients to retain ownership of their computing resources while Durantic handles the underlying infrastructure.
Central to Durantic's service is a bare-metal-native control plane, developed by the same team that manages it. An agent installed on each machine oversees essential tasks such as provisioning, networking, storage coordination, telemetry collection, and the entire lifecycle of the infrastructure, from hardware acceptance to return merchandise authorization (RMA).
Each fleet managed by Durantic operates independently while also enhancing the software capabilities for future fleets, fostering a cycle of continuous improvement. As AI infrastructure divides into two distinct layers, Durantic aims to lead in the operational layer, where traditional solutions are limited. This separation consists of a capital-and-capacity layer with neoclouds like CoreWeave and Fluidstack, focused on GPU procurement and capacity deployment, contrasted with the operational layer that faces challenges in provisioning, networking, and orchestration.
The need for effective management of AI infrastructures has become clear as enterprises increasingly implement private AI factories. Many companies find their IT teams unprepared for the demands of bare-metal operations, which underscores the importance of Durantic's offerings. By reducing the operational burden, Durantic aims to enable AI companies to focus on their core competencies instead of getting caught up in infrastructure management.
As the market for AI infrastructure evolves, the demand for solutions that address operational gaps is likely to grow. With its unique positioning and experienced leadership team, Durantic is well-equipped to establish a niche in this competitive field. Successful execution could redefine the management of AI workloads, potentially leading to more efficient and scalable AI operations across various sectors.
The stories that move AI & crypto markets — before the market reacts.
Free. 7am ET. Five stories. 62,400 readers.
