NVIDIA
H200 SXM
- Rent
- Buy
- Location
- France
- Memory
- 141 GB HBM3e
- Bandwidth
- 4.8 TB/s
The go-to GPU for training and serving large language models. Its HBM3e memory fits bigger models on a single GPU.
- AI training
- Inference
- Generative AI
- HPC
Rent or buy GPUs
Rent or buy the right GPU for your AI, inference or high-performance computing projects, hosted in partner data centers across Europe.
Inventory updated on
Available GPUs
NVIDIA
L40S
- Rent
- Location
- Netherlands
- Memory
- 48 GB GDDR6
- Form factor
- PCIe
A versatile GPU with a strong price/performance ratio: inference for mid-sized models, image and video generation, 3D rendering.
- Inference
- Generative AI
- 3D rendering
AMD
Instinct MI300X
- Buy
- Location
- Finland
- Memory
- 192 GB HBM3
- Bandwidth
- 5.3 TB/s
Very large memory capacity to host models with hundreds of billions of parameters. An alternative to NVIDIA for training and HPC, backed by the open ROCm stack.
- AI training
- Inference
- HPC
SambaNova
SambaRack SN40L-16
- Buy
- Location
- Worldwide
- Chip
- SN40L RDU
- Power draw
- 16.4 kW max per rack
- Cooling
- Air
SambaNova racks installed in your data center and fully managed by SambaNova. Air-cooled SN40L RDU chips serve the largest open-source models (Llama, DeepSeek, Qwen) at high speed.
- Inference
- Generative AI
- Agentic AI
SambaNova
SambaManaged Container
- Buy
- Location
- Worldwide
- Deployment
- 30 days
- Scalability
- Up to 1 MW
- Cooling
- Air
A turnkey AI inference data center, delivered in a container and live in 30 days. SambaNova manages the models, your customers' billing and maintenance.
- Inference
- Generative AI
- Agentic AI
Intel
Gaudi 3
- buy
- Location
- Worldwide
- Memory
- 128 GB HBM2e
- Bandwidth
- 3.7 TB/s
Intel's AI accelerator, an alternative to GPUs for inference and training, with built-in Ethernet networking to scale out.
- AI training
- Inference
- Generative AI
NVIDIA
GB300 NVL72
- buy
- Location
- Worldwide
- Configuration
- 72 Blackwell Ultra GPUs + 36 Grace CPUs
- GPU memory
- Up to 21 TB HBM3e per rack
- Cooling
- Liquid
The Blackwell Ultra version of the NVL72 rack, with more memory per GPU. Built for reasoning models and the heaviest inference workloads.
- AI training
- Inference
- Generative AI
No GPU matches these filters. Adjust them or get in touch: we can source other models.
Frequently asked questions
Can I rent or buy the GPUs?
Both, depending on the model. Each card shows whether the GPU is available to rent, to buy, or both. Renting suits one-off or fixed-term projects; buying suits long-term needs hosted in colocation.
Where are Datalok's GPUs hosted?
GPUs are hosted in partner data centers across Europe. The country is shown on each card, so you can pick a location that fits your latency and data sovereignty requirements.
What does "on order" mean?
The GPU is not available right away: it is sourced once your request is confirmed. The lead time is stated in your quote.
Can't find the GPU you need?
Our partner data center network grows every week. Tell us what you need, to rent or to buy, and we'll find the capacity.
Describe my needs