DataCrunch

Provisions isolated virtual machine instances with dedicated NVIDIA A100 or H100 GPUs on European infrastructure, billed on a pay-as-you-go model with per-second granularity. Instances are allocated from a managed pool of bare-metal hosts with InfiniBand/RoCE interconnect, enabling immediate access to single or multi-GPU configurations without reservation requirements. Terraform and OpenTofu integration allows infrastructure-as-code provisioning workflows.

Provisions pre-configured multi-GPU clusters (16x, 32x, 64x, 128x GPU configurations) with InfiniBand/RoCE interconnect and NVLink support for distributed training workloads. Clusters are deployed as isolated bare-metal environments with shared filesystem (SFS) and block storage, enabling immediate distributed training without manual node orchestration. Cluster sizing is fixed to predefined tiers rather than dynamic auto-scaling, optimizing for predictable performance and cost.

Exposes a REST API for programmatic access to all DataCrunch resources (instances, clusters, storage, containers, inference endpoints) with JSON request/response payloads. The API enables integration with custom applications, CI/CD systems, and orchestration tools, with authentication via API keys and support for standard HTTP methods (GET, POST, PUT, DELETE). API responses include resource metadata, status information, and error details for error handling.

Guarantees that all customer data (training data, models, checkpoints, logs) remains within European Union data centers, with transparent compliance documentation and SOC 2 Type II certification. The platform is European-owned and operated, eliminating US data transfer concerns and enabling compliance with GDPR, NIS2, and other EU regulations. Data residency is enforced at the infrastructure layer, not just contractually.

Provides monitoring capabilities for tracking GPU instance performance, resource utilization, and billing metrics through a web dashboard and API. Monitoring data includes CPU/GPU utilization, memory usage, network throughput, and cost tracking, with potential integration points for external monitoring tools (Prometheus, DataDog, etc., details unknown). Metrics are collected automatically and accessible via dashboard or API for custom analysis.

Offers managed services and co-development partnerships for building custom AI solutions, including model training, fine-tuning, and optimization. DataCrunch's in-house AI lab provides expertise in compiler optimization, inference optimization, and reinforcement learning frameworks, with potential for custom development engagements. Services are billed on a project basis with custom pricing.

Deploys Docker containers as managed, auto-scaling endpoints that execute on-demand without requiring instance management. Containers are submitted to a managed platform that handles resource allocation, scaling, and lifecycle management, with billing on a pay-per-request model. The platform automatically scales endpoints based on incoming request volume, abstracting away cluster management while maintaining GPU acceleration for inference or batch processing tasks.

Provides pre-configured, cost-optimized inference endpoints for a catalog of state-of-the-art AI models (specific model list unknown), deployed on optimized GPU infrastructure with automatic batching and request queuing. Endpoints are accessed via HTTP API without requiring container management or model deployment expertise, with billing on a per-request or per-token basis. The platform handles model serving, scaling, and optimization transparently.

Provisions persistent block storage volumes attachable to GPU instances with configurable capacity and performance characteristics. Storage is managed independently from compute, enabling data persistence across instance lifecycles and sharing between instances. Volumes are accessed via standard block device interfaces (e.g., /dev/sdX on Linux) and support standard filesystem operations, with billing based on provisioned capacity and I/O operations.

Provisions a shared filesystem accessible from all nodes in a GPU cluster, enabling coordinated data access and checkpoint sharing without explicit data movement. The filesystem is mounted on all cluster nodes at a consistent path, supporting standard POSIX operations (read, write, append) with transparent synchronization across nodes. Protocol and performance characteristics (NFS, CEPH, etc.) are not documented, but the service abstracts underlying storage infrastructure.

Provides object storage accessible via S3-compatible API (or proprietary interface, details unknown) for storing large files, datasets, and model artifacts. Storage is billed on a capacity and request basis, with support for standard operations (put, get, delete, list) and potentially advanced features like versioning, lifecycle policies, and access controls. The service abstracts underlying storage infrastructure while maintaining API compatibility with S3 tooling.

Provides a managed container registry for storing and managing Docker container images, with support for image versioning, access control, and integration with container deployment services. Images are stored in the registry and pulled by instances/endpoints during deployment, with billing based on storage capacity and potentially bandwidth. The registry abstracts underlying storage while maintaining Docker Registry API compatibility (or proprietary interface, details unknown).

Provides Terraform and OpenTofu provider plugins enabling declarative infrastructure provisioning through HCL configuration files. The provider abstracts DataCrunch API calls, allowing users to define GPU instances, clusters, storage, and networking as code with version control, reproducibility, and state management. Terraform/OpenTofu handle resource lifecycle (create, update, destroy) and dependency resolution automatically.

Command-line interface for managing DataCrunch resources (instances, clusters, storage, containers) with commands for provisioning, monitoring, and lifecycle management. The CLI abstracts API calls into human-friendly commands, enabling scripting and automation without direct API integration. CLI supports authentication via API keys and outputs results in human-readable and machine-parseable formats (JSON, YAML).