Nscale

What Nscale does

Nscale is building the engine of superintelligence, providing full-stack AI infrastructure that powers the world’s most powerful systems, from ground to cloud. They offer a complete AI cloud platform designed for scale, resilience, and speed, enabling the deployment of AI. Their services include AI Services (inference endpoints, fine-tuning workflows, prompt engineering workbench), Infrastructure Services (high-throughput, low-latency backbone for AI and HPC workloads), Fleet Operations (automated system-wide telemetry, configuration control, health monitoring), and Platform Services (VMs, bare metal nodes, Nscale Kubernetes Service, Slurm clusters). Nscale aims to reduce time-to-market for AI solutions, deliver differentiated models, shorten R&D cycles, lower operational risk, and maximize GPU utilization while reducing costs. They emphasize an AI-native approach, unifying the AI stack and providing precise control over token economics.

Products

• AI ServicesInference endpoints, fine-tuning workflows, and a unified workbench for prompt engineering to reduce time-to-market, deliver differentiated models, and shorten R&D cycles.
• Infrastructure ServicesHigh-throughput, low-latency backbone engineered for AI and High Performance Computing (HPC) workloads, delivered as raw bare-metal nodes on NVIDIA GPUs with AI-optimized storage tiers and RDMA/InfiniBand/NVLink fabrics.
• Fleet OperationsAutomated system-wide telemetry, configuration control, and health monitoring to maximize GPU utilization at scale, lower run-rate costs, drive cost accountability, and reduce financial risk.
• Platform ServicesInstances available as virtual machines (VMs) or bare metal nodes, with orchestration options using Nscale Kubernetes Service (NKS) or Slurm clusters, including Nvidia’s Slinky for HPC-grade batch scheduling.
• AI MarketplaceA platform offering turnkey AI development and deployment, providing access to ready-to-go AI & ML tools and resources, an extensive model library, and tailored resources for efficient and scalable model development and deployment.
• Inference EndpointsFully managed endpoints for deploying and scaling production inference, offering low-latency and high throughput with strict customer isolation.
• Fine-TuningService to customize foundation models with enterprise data, offering streamlined workflows to lower cost and complexity, and move tuned models into production.
• Prompt WorkbenchA tool to make prompt engineering reproducible, collaborative, and production-ready, reducing trial-and-error costs and accelerating time-to-prototype.
• Managed SlurmAn HPC-grade Slurm batch scheduler that runs on Kubernetes to manage large GPU training runs, simplify mixed workloads, and provide predictable R&D timelines.
• Nscale Kubernetes Service (NKS)A service for running production-ready Kubernetes and lightweight Kubernetes clusters for various workloads and experiments, offering fast spin-up, multitenancy isolation, and failure recovery.
• InstancesCompute flexibility with managed bare-metal nodes and Virtual Machines for experimental workloads, offering data and network sovereignty with VPC isolation.
• Control CenterA unified lifecycle manager to maximize GPU efficiency, automate routine tasks, track node health, and operate complex multi-cluster environments.
• ObservabilityPlatform-grade telemetry across compute, storage, and networking to reduce downtime, control costs, track compliance, and provide predictable budgeting.
• Radar APIAn API exposing real-time fleet signals for planning and procurement, providing GPU availability, repair metrics, resource stats, and maintenance notices.
• GPU NodesBare-metal infrastructure accelerated by NVIDIA DGX Vera Rubin NVL72, GB300 NVL72, H100, H200, and GB200 GPUs for high-performance AI, ML, and HPC workloads.
• AI Compute for Training LLMsA highly scalable, performance-optimized architecture with industry-leading GPUs to significantly reduce training times and boost productivity for AI models.
• Platform InfrastructureTools to provision and manage compute, networking, and storage resources directly in the Nscale Console, including Instances, VPC Networks, Filesystem, Managed Kubernetes, Security Groups, and Terraform Provider.

Key capabilities

• Full-stack AI infrastructure
• AI cloud platform
• Inference endpoints
• Fine-tuning workflows
• Unified workbench for prompt engineering
• Autoscaling inference layer
• Nscale-managed GPU clusters
• Serverless, API-driven fine-tuning pipelines
• Browser-based workbench with versioned prompts and real-time model feedback
• High-throughput, low-latency backbone for AI and HPC
• Raw bare-metal nodes on latest-generation NVIDIA GPUs
• Parallel, AI-optimised storage tiers with GPU-tuned distributed file systems
• RDMA/InfiniBand/NVLink fabrics, multi-rack topology, low-latency interconnects
• Automated system-wide telemetry, configuration control, health monitoring
• Unified lifecycle manager for provisioning, scaling, patching, node health tracking
• End-to-end visibility into workloads with built-in dashboards, alerts, reporting
• Real-time GPU resource governance and repair visibility via Radar API
• Instances as virtual machines (VMs) or bare metal nodes
• Nscale Kubernetes Service (NKS)
• Slurm clusters
• Nvidia’s Slinky (HPC-grade batch scheduling service)
• Isolated Kubernetes environments for rapid testing
• GPU-aware scheduling, seamless autoscaling, enterprise-grade security
• Nscale-managed lifecycle controllers, prebuilt AI images, optional VPC isolation
• Access to top AI/ML frameworks (PyTorch, TensorFlow)
• Extensive open-source model library with proprietary optimisations
• Preconfigured templates and customizable tools
• Quick deployment of applications and resources
• 80% lower cost compared to hyperscalers
• Integrated software and hardware tailored for AI
• Up to 30% faster time to value for AI projects
• Serverless training and inference
• GPU nodes
• Nscale's Data centers powered by renewable energy
• LLM Library
• Pre-configured Software and Infrastructure
• Job Management and Scheduling
• Container Orchestration
• Optimised Libraries, Compilers and Tools, Runtime
• Purpose-built infrastructure for distributed GPU workloads
• Topology-aware scheduling
• Single control plane for mixed training
• Full-stack observability
• Granular visibility into model costs
• Flexible capacity
• On-demand access to frontier GPU clusters
• Developer-first APIs and tooling
• Cost-per-experiment visibility
• Sovereign by default (in-country deployment)
• Modular, sovereign, sustainably-powered data centers
• Control Center for unified lifecycle automation
• Radar API for real-time fleet signals
• NVIDIA DGX Vera Rubin NVL72, GB300 NVL72, H100, H200, GB200 GPUs
• 3,600 PFLOPS NVFP4 inference performance
• 2,520 PFLOPS NVFP4 training performance
• 75 TB total fast memory
• 72 Rubin GPUs unified by sixth-generation NVLink
• Simplified Scheduling and Orchestration with Slurm and Kubernetes
• Fastest GPU nodes available for bare metal
• VPC Networks for isolated private networks
• Filesystem for shared persistent NFS storage
• Security Groups for firewall rules
• Terraform Provider for infrastructure as code

Use cases

• Launch inference in minutes
• Customize frontier models to specific domains
• Accelerate path from POC to production
• Experiment and optimize prompts quickly
• Maximize GPU efficiency and utilization for experiments
• Ensure predictable throughput for training and inference at scale
• Scale training from dozens to thousands of GPUs
• Automate provisioning, scaling, patching, and remediation workflows
• Gain end-to-end visibility into workloads for cost accountability and compliance
• Real-time GPU resource governance and capacity planning
• Manage mixed workloads with predictable queue times
• Provision isolated Kubernetes environments for rapid testing
• Production-ready training with GPU-aware scheduling and autoscaling
• Deploy a wide range of software and hardware resources
• Build using top AI/ML frameworks like PyTorch and TensorFlow
• Access open-source models with proprietary optimisations
• Optimize models for production
• Run large-scale inference
• Distributed GPU workloads training
• Production inference and serving
• Research and experimentation
• Deliver AI services for Telcos
• Optimize 5G networks
• Support advanced AI workflows
• Drive next-generation solutions for Telcos
• Enhance model development for AI-native companies
• Support critical operations for AI-native companies
• Drive innovation in tech solutions for AI-native companies
• Deploy and scale production inference with fully managed endpoints
• Fine-tune models with own data to align behavior, accuracy, and outputs
• Make prompt engineering reproducible and collaborative
• Accelerate prompt iteration and tuning
• Run serverless, autoscaling inference on Nscale-managed GPUs
• Ship reliable AI with reproducible prompts and fine-tuning
• Manage large GPU training runs with HPC-grade Slurm
• Run production-ready Kubernetes and lightweight Kubernetes clusters
• Maximize performance for intensive workloads with bare-metal nodes
• Experimental workloads with Virtual Machines
• Data and network sovereignty with VPC isolation
• Optimize AI fleet onboarding and lifecycle management
• Reduce operational overhead by automating routine tasks
• Improve efficiency with node-level health tracking
• Operate complex multi-cluster environments
• Spot issues early with telemetry across compute, storage, and networking
• Predictable budgeting with integrated cost reporting and dashboards
• Meet compliance and audit requirements with detailed operational metrics
• Eliminate uncertainty around GPU availability for planning and procurement
• Accelerate large-scale AI inference workloads
• Frontier-scale model training and post-training
• Reasoning, simulation, and large AI workloads
• AI training, rendering, and scientific computing
• Graphics rendering and simulation for gaming and entertainment
• Medical imaging and analysis for healthcare
• Quantitative analysis and risk modeling for finance
• Autonomous driving and vehicle simulation for automotive
• Simulation and modelling for aerospace and engineering
• Evaluate model performance
• Run popular LLMs effortlessly with API
• Provision and manage compute, networking, and storage resources
• Create GPU and CPU virtual machines with SSH access
• Manage Nscale infrastructure as code with Terraform

AI approach

Nscale provides full-stack AI infrastructure, offering GPU cloud computing solutions for training, fine-tuning, and inference of AI models. They provide access to leading AI/ML frameworks, an extensive model library, and preconfigured hardware options. They optimize open-source models with proprietary software on NVIDIA GPUs and offer serverless, API-driven fine-tuning pipelines. Their infrastructure is designed for high-performance computing (HPC) workloads, with features like autoscaling inference layers, GPU-tuned distributed file systems, and RDMA/InfiniBand/NVLink fabrics.

Tech named: PyTorch, TensorFlow, NVIDIA GPUs, NVIDIA A100, NVIDIA H100, NVIDIA GB200, NVIDIA H200, NVIDIA V100, NVIDIA DGX Vera Rubin NVL72, GB300 NVL72, NVLink, RDMA, InfiniBand, ONNX Runtime, Kubernetes, Slurm, Nscale Kubernetes Service (NKS), Nvidia Slinky, GPT OSS 120B, GPT OSS 20B, Qwen 3 4B Instruct, Llama 4 Scout, Qwen3 4B Thinking, Qwen3 8B, Stable Diffusion XL Base 1.0, Mistral 8x22B Instruct, FLUX.1 [schnell], Kimi K2.5

Industries served

• Technology
• Information
• Internet
• Telco
• Artificial Intelligence (AI) and machine learning research and development
• Gaming and entertainment
• Healthcare
• Finance
• Automotive
• Aerospace and engineering

What it says sets it apart

• Full-stack AI infrastructure from ground to cloud
• Purpose-built for AI with scale, resilience, and speed
• Reduces time-to-market for AI solutions
• Enables customization of frontier models
• Shortens R&D cycles and lowers operational risk
• Maximizes GPU efficiency and utilization
• Predictable throughput for training and inference at scale
• No network bottlenecks with advanced fabrics
• Automated system-wide configuration control and health monitoring
• Unified lifecycle manager for operational overhead reduction
• End-to-end visibility for cost accountability and compliance
• Real-time GPU resource governance and repair visibility
• Flexible deployment with VMs, bare metal, Kubernetes, Slurm
• HPC-grade batch scheduling with Nvidia’s Slinky
• Isolated Kubernetes environments for rapid testing
• GPU-aware scheduling and enterprise-grade security
• Access to leading AI/ML frameworks and extensive model library
• Proprietary optimisations on NVIDIA GPUs
• 80% cost-saving compared to hyperscalers
• Up to 30% faster time to value for AI projects
• Integrated software and hardware tailored for AI
• AI-native approach unifying the AI stack
• Precise control of token economics
• Purpose-built infrastructure for distributed GPU workloads
• Single control plane for mixed training
• Full-stack observability
• Granular visibility into model costs
• Flexible capacity that matches demand
• On-demand access to frontier GPU clusters
• Developer-first APIs and tooling
• Sovereign by default for regulated workloads
• Modular, sovereign, sustainably-powered data centers
• Control Center for unified lifecycle automation
• Radar API for real-time fleet signals
• High-performance NVIDIA DGX Vera Rubin NVL72, GB300 NVL72, H100, H200, GB200 GPUs
• Unmatched scalability, energy efficiency, and fully integrated enterprise solutions
• Raw performance of bare metal GPUs without bloated infrastructure
• Simplified Scheduling and Orchestration with Slurm on Kubernetes (SLONK)
• Fastest available GPU-accelerated bare metal nodes
• Comprehensive support including performance tuning and model optimisation techniques
• Robust security measures including encryption, access controls, network security protocols, and compliance with GDPR/HIPAA
• Multi-tenant environments ensuring resource isolation and data privacy

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.