Nebius is increasingly positioning itself as more than a provider of GPU capacity, building software and managed services around the hardware required for artificial intelligence. Aether, Soperator and Token Factory create layers above the underlying compute infrastructure that can help customers deploy, train and serve AI models more efficiently, potentially allowing Nebius to capture more value from each unit of infrastructure it operates.

Aether Expands the Value of AI Cloud Infrastructure

Nebius AI Cloud’s Aether platform is designed to support organizations running AI workloads at production scale. The platform combines compute, storage, networking, security, observability and governance, with newer releases focused on making infrastructure easier to operate and giving enterprise customers greater control.

The significance of Aether is that customers are not simply renting individual GPUs. They are using a broader environment designed to reduce the complexity of deploying and managing AI workloads. Serverless capabilities, storage improvements and infrastructure-management tools can reduce the operational burden on developers and allow companies to spend more time building applications rather than configuring cloud environments.

Soperator Targets Higher GPU Utilization

Soperator, Nebius’ Kubernetes operator for Slurm, is aimed at large-scale AI training and high-performance computing. The managed service handles cluster provisioning, dependencies, health checks and recovery, while smart scheduling and topology-aware job placement are designed to improve the efficiency of large GPU clusters.

That matters because the economics of AI infrastructure depend on utilization as much as installed capacity. A GPU that remains idle produces little economic value, while better scheduling can allow customers to extract more productive compute from the same hardware. Automatic recovery can also reduce the impact of hardware or node failures during long-running training jobs.

Token Factory Moves Nebius Further Into AI Inference

Token Factory adds another layer by focusing on production inference rather than simply supplying the infrastructure underneath it. The platform supports leading open models and provides managed endpoints, autoscaling, observability, function calling, structured outputs and fine-tuning capabilities. Nebius is also integrating newer inference hardware into the platform, including NVIDIA Groq 3 LPX, which is designed to accelerate token generation for latency-sensitive and agentic AI workloads.

The broader opportunity is therefore tied to how much useful AI work customers can perform on Nebius infrastructure. Token Factory can help turn compute capacity into production workloads, while Aether and Soperator can make that capacity easier and more efficient to use. Nebius’ second-quarter revenue reached $582.3 million, up 454% year over year, while annualized run-rate revenue reached $3 billion at the end of June.

Going forward, the key question for NBIS is whether these software and managed-service layers can continue increasing utilization, customer retention and revenue faster than the cost of expanding physical infrastructure. If Nebius can capture value across compute, orchestration and inference rather than relying primarily on GPU rentals, its addressable market and potential revenue model could become broader as enterprise AI workloads move from experimentation into large-scale production.