How they use Nebius AI Cloud

CentML runs its inference software on Nebius GPU infrastructure. In a customer interview, CEO Gennady Pekhimenko described using Nebius bare-metal servers and virtual machines beneath CentML's software stack. The companies also announced a combined offering for developers and enterprises deploying AI models.

CentML optimizes model execution, while Nebius provides compute. The accounts describe inference, support for training jobs and flexible capacity reservations. Pekhimenko described three- and six-month reservations alongside on-demand pricing. CentML's partnership announcement describes dedicated engineering support for complex multi-GPU implementations and model optimization across infrastructure layers. CentML also says the combined offering supports seamless deployment across cloud environments.

Nebius reports flexible scaling, streamlined deployments and improved hardware utilization for CentML's inference platform. Its customer collection reports improved reliability and scalability, enhanced EU-based compute capabilities, one week to bring a cluster online and fivefold lower costs compared with other major providers. Pekhimenko also described building frameworks for enterprise AI agents and anticipated increased compute demand from those workloads. In Nebius's Q1 2025 presentation, CTO and co-founder Shang Wang credited Nebius's capacity allocation, rapid onboarding, flexible pricing and reliable availability with supporting CentML's growth.