Durantic Emerges as Key Player in AI Infrastructure Management

In a market where AI infrastructure is increasingly fragmented, Durantic has officially launched its managed services, marking a shift in how AI workloads operate. Founded by experts from Meta and Hudson River Trading, Durantic targets clients with computational resources but lacking the capacity to manage them effectively.

Addressing Operational Bottlenecks

The AI infrastructure market is undergoing a major transition as workloads migrate across various platforms, including hyperscalers, GPU clouds, and colocated bare metal servers. This fragmentation has created operational challenges that many companies struggle to navigate. According to Ivan Diachenko, CEO of Durantic, “AI companies are being forced to become infrastructure operators, and most don’t want to be.” The company offers a solution by merging software development with direct operational management, helping clients manage the complexities of AI infrastructure.

Durantic employs a productized engagement model, providing a 30-day pilot for clients to trial its services on a specific portion of their GPU fleet. Following the pilot, clients pay monthly fees based on the number of GPUs and servers used. This model allows clients to retain control over their computing resources while delegating the complexities of infrastructure management to Durantic.

A Unique Technology Approach

Central to Durantic’s services is a bare-metal-native control plane, designed by the engineers who will also operate it. Each machine in a client’s fleet includes an agent that oversees provisioning, networking, storage coordination, and telemetry collection. This infrastructure lifecycle management, from hardware acceptance to RMA, boosts Durantic’s operational efficiency with each new fleet deployed.

Diachenko elaborates on the company’s perspective regarding the division in the AI infrastructure market. He identifies two layers: the capital-and-capacity layer and the operational layer, which lacks a dominant player. “The pattern rhymes with VMware over commodity servers, Datadog over cloud visibility, Red Hat over Linux,” he noted, stressing that historically, operating layers above commoditized capacities have proven to be the most sustainable infrastructure businesses.

Experienced Founders with a Vision

The founding team of Durantic brings substantial operational expertise. Diachenko has a background with Meta, where he contributed to the Traffic Disaster Recovery team, developing solutions for over three billion users. His co-founder, Dmitrii Skokov, previously led infrastructure platforms at Replika, focusing on MLOps and GPU provisioning for large-scale AI services. Their combined experience equips Durantic to address the operational challenges many enterprises encounter as they implement private AI systems.

Looking Ahead

Based in London, Durantic is expanding its team and developing a pipeline of pilot projects across the US, UK, and European markets. The company’s dual emphasis on software development and infrastructure management positions it uniquely to meet the urgent operational needs of AI companies. As the AI infrastructure market evolves, Durantic aims to simplify operations, allowing businesses to concentrate on their core competencies without being overwhelmed by infrastructure management complexities.