Compute
Storage
For consumers
Workloads
Why Hivenet
Sign in to Hivenet
Access your files and manage your storage account.
Store
Access your files and manage your storage account.
Compute
Launch instances and manage your compute resources.
Store and Compute still use separate accounts today. We’re working on a unified sign-in.
Just need to send a file?
Use Send directlyNew to Hivenet? Get Started
Get started with Hivenet
Choose what you want to use first.
Store
Back up photos and files across your devices.
Compute
Launch self-serve GPU compute for demanding work.
Store and Compute still use separate accounts today. We’re working on a unified account experience.
Need Hivenet for business?
Talk to salesNeed to send a file?
Open SendAlready have an account? Sign in
Contact Hivenet
Choose the path that fits your question.
General contact
Questions, support, partnerships, press, or anything else not tied to a sales conversation.
Talk to sales
Ask about Compute, storage, private AI, or a larger business deployment.
Need help with setup, billing, or troubleshooting? Start with Support for the fastest answer.
Compute - GPU and CPU rental
GPU and CPU power that performs exactly as promised, every single run.
Launch RTX 4090, RTX 5090, RTX 6000-series, or vCPU instances for AI inference, open-source and open-weight models, fine-tuning experiments, notebooks, rendering, batch jobs, APIs, and development environments. Get fixed GPU rates, per-second billing, regional deployment paths, and infrastructure built for quality at the right price.
RTX 4090 from €-/hr
RTX 5090 from €-/hr
RTX 6000-series for enterprise
Predictable pricing
Per-second billing
Templates and OS images
SSH access
Team organizations
Public Compute API
France, UAE, and USA deployment paths
Workload guidance
Tell us what you want to run.
We'll point you to the right instance.
Choose the right path before you spend a euro:
General compute
Testing and research
Specialized AI throughput
Enterprise workloads
Managed API endpoint
Teams already running compute on Hivenet
Researchers, AI teams, studios, and industry groups run their GPU and CPU workloads on Compute with Hivenet because performance holds, pricing stays predictable, and they keep control of the environment.
Hivenet makes our work much easier. With Jupyter notebooks, fast access to GPUs, and reliable infrastructure across regions, we’ve been able to speed up our research on green enzymes for industrial use. It feels like a real step forward in compute platforms.
Jupyter notebooks, fast GPU access, green enzymes research.
The possibility of easily obtaining instances with graphics cards for a really cheap price. I use it a few times a month and it's really perfect.
What we like about Hivenet is that it matches how we think about AI: sustainable, efficient, and grounded in Europe. The distributed model gives us security, the option to choose European models strengthens our sovereignty message, and small details like pausing instances or fair pricing just make it practical for us to use day to day.
Sustainable, European, sovereignty, fair pricing.
We've moved over from AWS and GCP to hive. It's a great way to save on costs.
Hivenet’s GPUs have been key to scaling our AI work. They let us run advanced models smoothly, so our interactions with students stay fast and responsive. That reliability has made a real difference for us.
GPU reliability for scaling AI work with students.
Super quick setup. No BS/straightforward pricing. Fast cold starts. Reliable machines.
A strong home for open-source AI workloads.
Open-source models give teams a way off the closed-API treadmill. Hivenet gives those models the infrastructure they deserve: the right GPU for their class, full control over the serving stack, and pricing that makes running your own model the obvious call. This is the workload Compute was built for — summarization, extraction, classification, RAG, support automation, code assistance, and internal tools, on the open-source models you actually ship.
Take full control of the serving stack on Compute, or hand it off to a managed endpoint with Hivenet Inference. Either way, you run the open-source models you choose on hardware that fits them.
Qwen workloads
Test Qwen model classes against your prompts, latency target, context length, and cost-performance needs. Smaller Qwen models can fit efficient GPU workflows, while larger classes need testing before production.
DeepSeek distilled models
Run distilled DeepSeek workloads where the model size fits RTX 4090 or RTX 5090 hardware. Treat larger reasoning workloads as benchmark candidates before production use.
Llama workloads
Run Llama-class workloads for RAG, summarization, internal tools, assistants, experiments, and model-serving tests. Choose the instance based on model size, precision, concurrency, and latency target.
Mistral, Falcon, Gemma, Phi, and smaller models
Use right-sized GPU capacity for smaller and mid-sized model families where throughput, latency, and cost control matter more than maximum model size.
Anyone can quote a GPU price. We deliver the whole outcome.
Compute with Hivenet is built for teams that need predictable performance, the right workload fit, and real operational control. If all you want is the rock-bottom marketplace rate, a spot-style GPU provider may look cheaper until your job gets preempted, your costs swing, and your engineers lose a day. If you need predictable pricing, regional deployment, repeatable environments, and a clean path from experiment to production, this is the decision Hivenet was built for.
Quality at the right price
Fixed, published GPU rates mean you know the spend before you run. Per-second billing keeps short jobs honest, charged for the time they actually use. Predictable performance at the right price, not the cheapest number on a marketplace.
The right instance, before you waste a run
Match the model, batch size, latency target, and operating model to the right hardware, instead of gambling on a GPU name and paying for the mistake.
Repeatable workloads
Templates, OS images, SSH workflows, and your own stack take you from notebooks and tests to repeatable production workloads without rebuilding every time.
Sovereign and compliant
Deploy suitable workloads across available regions, including France, the UAE, and the USA, on enterprise-grade infrastructure operated by Hivenet end-to-end, with the control and exit path to keep the workload yours.
Real help on serious workloads
Talk to Hivenet when the workload needs a human: instance choice, model fit, production setup, or migration. No ticket-queue limbo.
Ready for your whole team
Organizations, shared billing, role-based access, and the Public Compute API for when the workload is bigger than one person clicking through a console.
Run any workload, big or small.
Whatever the size, there's a right path for it. Start with vCPU for general-purpose compute, step up to RTX 4090 for testing and research, RTX 5090 for specialized AI throughput, and RTX 6000-series for enterprise-scale work. Small experiment or production deployment, the platform fits the job.
Workload
Recommended path
Why
Web app, API, dev database, or background service
vCPU
General-purpose workloads usually do not need GPU acceleration
Batch scripts or preprocessing
vCPU or GPU
Start with vCPU unless the job uses CUDA or parallel acceleration
Jupyter, PyTorch, or model experiments
RTX 4090 or RTX 5090
GPU acceleration helps with ML workflows and iterative testing
ComfyUI, Stable Diffusion, Flux, or rendering
RTX 4090 or RTX 5090
Image and rendering workloads benefit directly from GPU acceleration
Sub-13B model serving
RTX 4090
Strong cost-performance for smaller open-weight models
Sub-30B model serving
RTX 5090
More VRAM and memory bandwidth for stronger single-GPU throughput
70B-class model testing
8× RTX 5090 host
Possible with tensor parallelism, but latency and throughput should be tested
Enterprise-scale production workloads
RTX 6000-series
Enterprise headroom for larger model classes and sustained serving
Managed OpenAI-compatible endpoint
Hivenet Inference
Compute gives you an instance; Inference gives you a managed API endpoint
Swipe left to see more
Right-size the instance. Maximize compute per euro.
Compute with Hivenet hands you enterprise-grade infrastructure with full control of the instance, the environment, and the stack. Each GPU has a job.
RTX 4090 · The testing and research GPU
The cost-efficient workhorse for testing, research, and dev.
Reach for RTX 4090 when the workload is smaller, cost-sensitive, or still in development. It is the GPU your team prototypes on, runs research experiments on, and develops against before scaling up. It works best on Hivenet because per-second billing and predictable rates make iterative testing cheap to repeat.
Specs
24 GB GDDR6X VRAM
About 1 TB/s memory bandwidth
Ada Lovelace Tensor Cores
From €0.40/hr
Per-second billing
Best for
Sub-13B inference
Llama 3.1 8B
Mistral 7B
Qwen 7B/14B
Phi-4
Fine-tuned 7B-class models
ComfyUI and image generation
Cost-efficient GPU development
RTX 5090 · The specialized AI GPU
Specialized single-GPU throughput for demanding AI.
Reach for RTX 5090 when you need more single-GPU headroom, stronger throughput, and support for larger practical model classes. It is the specialist for production inference and high-concurrency serving. Best on Hivenet because you get this class of GPU at predictable rates and can benchmark against bare metal.
Specs
32 GB GDDR7 VRAM
1.79 TB/s memory bandwidth
5th-gen Tensor Cores
PCIe 5.0
From €0.75/hr
Per-second billing
Best for
Sub-30B inference
High-concurrency small and medium models
Qwen, Llama, Mistral, Gemma, Phi, and distilled DeepSeek workloads
Rendering and creative pipelines
CUDA-heavy experiments
HPC-style pipelines that fit the hardware profile
RTX 6000-series · The enterprise GPU
Enterprise-grade headroom for production-scale workloads.
Reach for the RTX 6000-series when production scale, larger model classes, or enterprise deployments need more headroom than a single consumer-class GPU. It is the enterprise tier for teams running serious, sustained workloads. Best on Hivenet because you get enterprise capacity with the same predictable pricing and end-to-end control.
Specs
32 GB GDDR7 VRAM
1.79 TB/s memory bandwidth
5th-gen Tensor Cores
PCIe 5.0
From €0.75/hr
Per-second billing
Best for
Enterprise production deployments
Larger model classes
Sustained high-utilization serving
Demanding multi-tenant or regulated workloads
vCPU · General-purpose compute
General-purpose compute, no GPU premium.
Reach for vCPU when your workload does not need GPU acceleration.
Specs
CPU-only instances
Flexible vCPU options
Fixed RAM, disk, and bandwidth options
Per-second billing
Simple setup for everyday compute work
Configuration-based pricing
Best for
Web apps
APIs
Development environments
Testing databases
Automation
CI/CD
Preprocessing
Background services
Match the open-source model to the right GPU.
Use the fit table to match common open-source model classes to RTX 4090, RTX 5090, or RTX 6000-series capacity before you launch.
Model or workload class
Fit on RTX 4090
Fit on RTX 5090
Notes
Llama 3.1 8B
Strong fit
Strong fit
Good for efficient inference and development
Mistral 7B
Strong fit
Strong fit
Good for low-cost inference and experiments
Qwen 7B/14B
Strong fit
Strong fit
Good fit for smaller open-weight workloads
Phi-4
Strong fit
Strong fit
Good for smaller model workflows
Qwen 32B-class workloads
Test fit
Strong fit
RTX 5090 gives more headroom
Mistral Small 3 24B
Test fit
Stronger fit
Better suited to RTX 5090 for practical serving
DeepSeek-R1 distilled 7B/8B/14B
Strong fit
Strong fit
Good distilled-model candidates
DeepSeek-R1 distilled 32B
Test fit
Stronger fit
Benchmark before production use
70B-class models
Not single-GPU fit
Multi-GPU only
Requires tensor parallelism and latency testing
Full DeepSeek V3
Not a fit
Not a fit
Requires larger frontier-class infrastructure
Kimi K2 production serving
Not a fit
Not a fit
Requires larger frontier-class infrastructure
Swipe left to see more
Clear rates. Billed by the second. Priced for the outcome.
Some jobs run for minutes. Some run overnight. Some sit idle while a test takes longer than expected. Per-second billing helps you pay for what you actually use, with predictable performance at the right price.
RTX 4090
Starting price
Per-second billing
Sub-13B inference, fine-tuning, image generation, cost-efficient GPU work
RTX 5090
Starting price
Per-second billing
Sub-30B inference, demanding GPU jobs, strongest single-GPU performance on Compute
vCPU
Starting price
Per-second billing
Web APIs, dev databases, CI/CD, background services, general-purpose compute
No long commitment required. Use prepaid credits, stop instances when you do not need active compute, and terminate instances when you are done and no longer need local storage.
Need an instance or a managed endpoint?
Use Compute with Hivenet when you want GPU or CPU infrastructure and control over the stack. Use Hivenet Inference when you want an OpenAI-compatible managed endpoint without having to operate the inference layer yourself.
I want root-level control over an instance
I want to run vLLM, TGI, SGLang, llama.cpp, PyTorch, or Docker myself
I want an OpenAI-compatible endpoint without operating the stack
I want help building a private AI system on my data
I want CPU-only compute for general-purpose work
Build it once. Use it forever.
Set up the environment one time, save it as a template, and relaunch the exact same workload whenever you need it. No rebuilding from scratch, no drift between runs. Choose a container or virtual machine, pick a region, select GPU or vCPU, choose an OS or template, add your SSH key, configure HTTPS, TCP, or UDP if needed, and connect.
SSH into a VM
Run Docker
Open a Jupyter notebook
Serve a model with vLLM
Expose an API over HTTPS
Run PyTorch experiments
Use ComfyUI
Run background jobs
Reuse templates for repeat workloads
# Connect to your instance
ssh ubuntu@your-instance
# Run your stack in a container
docker run --gpus all -p 8000:8000 your-image
# Or serve an open source model with vLLM
python -m vllm.entrypoints.openai.api_server \
--model Qwen/Qwen2.5-7B-Instruct \
--port 8000
# Relaunch the same environment, every time
curl -X POST https://api.hivenet.example/v1/instances \
-H "Authorization: Bearer $HIVENET_API_KEY" \
-d '{"template": "my-vllm-template", "gpu": "rtx5090", "region": "fr"}'
Built for real teams and real automation.
Compute with Hivenet supports the way technical teams actually work: shared access, shared billing, role separation, and programmatic control for when console-only workflows stop being enough.
Team access without shared logins
Create an organization, invite teammates, assign roles, and switch between personal and organization workspaces.
Shared billing
Keep organization credits separate from personal credits and let the right person manage top-ups and payment methods.
Role-based access
Separate who owns billing, who manages members and resources, and who creates and operates instances.
Public Compute API
Create, start, stop, terminate, list, tag, and update instances programmatically. Manage SSH keys, billing, organization workflows, and quota requests through versioned API paths.
Built for automation
Use request IDs, machine-readable errors, pagination, rate-limit signaling, and the OpenAPI specification to connect Compute to scripts, CI/CD workflows, and internal tools.
Control the instance, not just the billing plan.
Use SSH, templates, OS images, GPU or vCPU instances, region choices, and per-second billing to run workloads your way. For teams with larger or production workloads, Hivenet can help review fit, architecture, and migration path before you commit.
Here is why you should trust usPredictable GPU rates
Per-second billing
France, UAE, and USA regions
SSH access
Templates and OS images
GPU and vCPU options
Clear distinction between raw compute and managed inference
Workload guidance for larger deployments
FAQ
Common questions
Bring the workload.
Get the instance that nails it.
Start with vCPU for general-purpose compute, RTX 4090 for testing and research, RTX 5090 for specialized AI, or RTX 6000-series for enterprise-scale workloads. Talk to us if you want help choosing the right path the first time.
