General FAQs
DataCrunch is a next generation AI cloud that gives AI builders instant access to powerful production-grade GPUs
at unbeatable prices. With self-service instances and clusters, dynamic pricing, and top-tier support,
we remove infrastructure barriers, so AI teams can focus on what matters most: building great models and products.
DataCrunch accelerates AI projects by becoming an extension of your team, focused on optimizing the performance,
reliability and costs for your AI workloads.
DataCrunch supports a broad range of use cases, including model training and inference on virtual machines
(instances and clusters), bare-metal servers, serverless containers and managed service endpoints. We also support
co-development of custom AI stacks and software integrations. Our customers and users include:
• AI-first startups and scaleups
• Applied research teams
• Infrastructure engineers deploying ML systems
• Enterprises needing inference or training at scale
1. We give AI builders instant access. DataCrunch gives you instant access to the latest GPU instances,
clusters, bare metal servers, serverless containers and managed endpoints without hassle or hurdles.
2. We're an extension of your AI team. We're always available with a team of AI infrastructure experts
to help solve performance, latency, and availability issues.
3. We optimize your AI spend, giving you the most value per GPU hour. We've optimized our AI stack to
give you the best performance and capabilities such as flexible storage management for 30-50% faster
startup times so you ultimately pay less with DataCrunch.
Dynamic pricing
can save up to 49% and spot instances provide additional cost saving opportunities.
We believe in transparency, trust, and sustainability. We're open in communication,
security practices,
privacy protections, and responsible energy usage. We're ISO 27001 certified and GDPR compliant.
We support our customers through a variety of channels including Crisp, Slack for AI teams and email,
providing access to AI experts for optimizing performance, scalability and infrastructure management.
In many cases, users are able to access instances, clusters, containers, or endpoints and begin running
workloads instantly without needing assistance. However, the DataCrunch team is always available for
assistance and wants to become an extension of your AI infrastructure team, from problem resolution
to knowledge sharing on optimization techniques.
DataCrunch makes it easy to access production-grade GPU resources, with higher availability of
AI optimized systems such as HGX servers, all at affordable prices. There are no sales hurdles or
delays to get started running AI workload, and we provide a developer-first experience through our
cloud dashboard and APIs. In many cases you'll also be able to save up to 90% compared to hyperscalers.
Compute & Infrastructure FAQs
DataCrunch provides access to a broad range of GPU models to meet the performance and cost requirements
of your specific workloads. GPUs include NVIDIA HGX B200, H200, H100, A100, L40S,
RTX 6000 ADA, RTX A6000, and V100. CPUs are primarily high-end server CPUs from the AMD EPYC family.
Please see our
instances and
clusters
pages for the latest information.
DataCrunch.io supports distributed training via dedicated multi-GPU clusters (instant or bare metal),
high-speed networking, and compatibility with industry-standard and open-source distributed training
frameworks. We support distributed frameworks such as PyTorch DDP, TensorFlow, HiveMind or advanced
frameworks like OpenDiLoCo. DataCrunch actively engages in multi-datacenter, distributed training
research and development such as the
global training of PrimeIntellect's INTELLECT-1.
Yes, you can reserve or schedule compute resources on the DataCrunch cloud. DataCrunch offers the
ability to reserve compute capacity by purchasing long-term rentals, which are paid upfront and
ensure that specific GPU instances or clusters are held exclusively for you during the contract period.
This is the primary method for guaranteeing access to high-demand resources, which can be crucial if
you need predictable, uninterrupted capacity for large projects or peak periods.
To schedule compute jobs, you can use the API combined with typical cron jobs or your own scheduling
code to ensure workloads run according to your schedule. DataCrunch instant GPU clusters come with
the Slurm job scheduling system pre-installed. Please keep in mind that scheduling is subject to available
capacity unless the capacity has been reserved.
There are many ways to deploy an AI model. The most common is a containerized model deployment using Docker,
which can be performed on a DataCrunch instance, cluster, or by using the managed Serverless Containers service.
We also provide support for the NVIDIA Triton Inference Server, vLLM, SGLang, FastAPI, Flask, and other tools and
frameworks.
Use of DataCrunch
managed inference endpoints
such as the FLUX models for image generation and editing or
Whisper model
for transcription or translation means that the model is built-in, optimized and ready to use.
The Flux.1 Kontext
managed service
endpoint is a turnkey, production-grade API for leveraging state-of-the-art FLUX models from Black
Forest Labs for next-gen image generation and editing, hosted and operated entirely as a managed
service—abstracting away the infrastructure, scaling, and performance tuning for the end user.
DataCrunch plans to release additional managed endpoints, such as FLUX.1 on the Krea platform.
Yes, DataCrunch.io is designed to let customers deploy and run their own machine learning models—containerized
or otherwise—on-demand, with minimal friction and support for a broad spectrum of deployment scenarios. You can
upload weights or mount them from local files, Hugging Face Hub, or cloud object storage (e.g. S3, GCS).
Developer Tools and APIs
DataCrunch focuses on a great developer experience, with fast access, easy onboarding, and an API-first approach.
DataCrunch provides a comprehensive set of APIs and client libraries for interacting with its GPU cloud resources,
running workloads, managing infrastructure, and deploying inference endpoints. These include a
public REST API and a
Python SDK.
DataCrunch provides the following developer tools, in addition to the published API:
• CLI: Deploy, monitor, and scale workloads
• Python SDK: Automate pipelines and experiment tracking
• Web Dashboard: Manage endpoints, track usage, inspect logs
• Monitoring: View latency, throughput, memory, and GPU metrics in real time for serverless containers
Yes, DataCrunch supports OpenAI style APIs for deploying and serving language models. This is possible
via integrations with popular open-source LLM frameworks such as SGLang and vLLM, both of which can be
deployed on DataCrunch in configurations that expose endpoints compatible with the OpenAI API protocol.
Billing & Pricing
DataCrunch bills for GPU compute using a flexible and transparent pricing structure designed to align
with market demand.
The main billing options are:
• Pay As You Go: Users are billed in 10-minute increments deducted from their account balance.
You can choose between two pricing models for these instances.
• Fixed pricing: The price per unit of GPU time remains constant throughout your usage period.
• Dynamic pricing: The rate fluctuates twice a day according to market demand, and savings can
be up to 49% of the fixed price. When demand is low, dynamic prices may be significantly lower
than fixed prices, but if demand surges, they can go up—though there is a maximum cap at 1.5x
the fixed price at the time you deploy the instance.
• Long-term Rentals: For users needing resources over an extended period, DataCrunch offers
discounted contracts paid up-front. This option is available for GPU Instances and Instant Clusters
• Spot Instances: These are 50% of the current dynamic pricing, and can only be accessed with Pay
As You Go pricing and may be terminated at any time, reflecting their variable and often lower costs
For current price lists, please see the product pages for instances, clusters, and serverless containers.
Payments can be made through the payment card on file or by bank transfer.
Yes — startup and research teams may qualify for free compute credits. In addition, qualified new accounts
may be granted limited credits to conduct a trial or proof of concept. Reach out at
[email protected].
DataCrunch.io does not impose traditional hard quota limits on GPU usage for general customers. Instead,
access is governed by real-time availability and elastic scaling. For special cases or very large clusters,
users can reserve capacity in advance. For constrained resources such as recently released new GPU models
in high-demand, users can request additional availability by contacting
[email protected].
Dynamic Pricing adjusts GPU rates based on real-time supply and demand — ensuring cost-efficiency
for users while maximizing utilization.
• When GPU demand is low, prices automatically decrease, helping you save on inference or batch training jobs.
• When demand spikes, prices may increase modestly to prioritize reserved and latency-sensitive workloads.
This model ensures you always pay a fair, market-aligned rate while giving you the option to reserve GPUs
in advance at stable prices for critical workloads. Read more about Dynamic pricing
here.
Security, Compliance & Privacy
We take privacy and security compliance seriously. We prioritize top-notch data security and safeguard your
intellectual property. DataCrunch is ISO 27001 certified and adheres to GDPR requirements.
DataCrunch is dedicated to upholding full European Union data sovereignty in all our data centers and cloud
services. We ensure that our infrastructure, operations, and contractual commitments rigorously adhere to the
core requirements for EU digital autonomy, regulatory compliance, and data protection.
Please see additional details about security controls and compliance in the DataCrunch
Trust Center and in the separate security and compliance
FAQs.
All of our data centers are
located
in the EU (Finland) or European Economic Area (Iceland). As such, they are under the GDPR. Please see our
docs for additional details.
Our datacenters adhere to the highest standards for physical and environmental security. These include 24/7 monitoring,
biometric access controls, and on-site security teams. All systems are protected by cooling, redundant power,
and fire suppression systems to maintain the highest possible availability of our services.
Our first line of defense is CloudFlare, which performs filtering of incoming traffic. Once traffic reaches
our services, additional protections are in place to further filter traffic to allow in only appropriate traffic.
Several automated systems continuously monitor our servers' and networks' behavior.