London, UK, May 20th 2026, ZEX PR WIRE — Durantic today emerged publicly as the managed infrastructure layer for AI workloads on heterogeneous GPU fleets. Founded by infrastructure engineers from Meta and Hudson River Trading, the company operates AI infrastructure for customers who have compute but lack the operational depth to run it efficiently — across customer-owned, colocated, leased, and hybrid environments.
The launch comes as AI infrastructure increasingly fragments. As workloads move across hyperscalers, GPU clouds, colocated bare metal, and sovereign clusters, the operational layer underneath has become the bottleneck: mixed GPU generations, Kubernetes and Slurm running side by side, custom network fabrics, hardware acceptance and burn-in, RMA evidence collection. Frontier labs are visibly hiring hyperscaler infrastructure operators. Enterprises deploying private AI factories discover their IT teams lack bare-metal depth. The bottleneck in AI infrastructure is no longer hardware availability — it is operationalization.
“AI companies are being forced to become infrastructure operators, and most don’t want to be,” said Ivan Diachenko, Durantic’s founder and CEO. “We write the software that operates AI infrastructure, and we operate it for customers directly. That combination is what produces operational leverage at scale. Software vendors can’t compress operational complexity because they don’t run the infrastructure. Services companies can’t because they scale with people. Doing both is the position we’re building.”
Durantic’s engagement model is productized: a 30-day pilot on a defined fleet slice, followed by recurring monthly platform-and-operations fees that scale with GPU and server count. Customers bring the compute; Durantic operates the infrastructure layer.
Underneath the managed service sits a bare-metal-native control plane written by the same team that operates it. An agent runs on every machine, handling provisioning, networking, storage coordination, telemetry collection, and the full infrastructure lifecycle from hardware acceptance through RMA. Every fleet operated by Durantic sharpens the software that operates the next one.
The company’s thesis is that the AI infrastructure market is bifurcating into two structurally different layers. The capital-and-capacity layer — neoclouds like CoreWeave, Fluidstack, Crusoe, and the sovereign AI buildouts — is increasingly defined by GPU procurement, financing, anchor tenant contracts, and capacity deployment. The operating layer underneath has no incumbent. Provisioning, networking, orchestration, observability, automation, and hardware lifecycle operations across heterogeneous compute are difficult to build internally because they require systems software, platform engineering, and infrastructure operations discipline in a single organization.
“The pattern rhymes with VMware over commodity servers, Datadog over cloud visibility, Red Hat over Linux,” said Diachenko. “Operating layers above commoditizing capacity are historically the most durable infrastructure businesses. AI infrastructure is producing the same dynamic now.”
The founding team brings hyperscaler operational depth to the AI infrastructure problem. Diachenko was a founding member of Meta’s Traffic Disaster Recovery team, where he built network disaster recovery for over three billion users. He subsequently spent four years at Hudson River Trading building low-latency bare metal infrastructure for global trading operations, running directly on physical hardware without cloud abstractions. Co-founder and CTO Dmitrii Skokov was Infrastructure Platform Lead at Replika, building MLOps and GPU provisioning for AI services running at scale.
The company is headquartered in London. Durantic is currently building its team and pilot pipeline across the US, UK, and European markets.
About Durantic
Durantic is the operating layer for fragmented AI infrastructure. The company writes the software that operates AI infrastructure and operates it for customers directly — managed Kubernetes, Slurm, inference serving, bare metal provisioning, network fabric, GPU fleet monitoring, and hardware acceptance — across customer-owned, colocated, leased, and hybrid environments.
Durantic is headquartered in London. More information at durantic.io.
The Post Durantic launches as the operating layer for fragmented AI infrastructure first appeared on ZEX PR Wire
