GPU compute & storage,built for AI teams.
Rent dedicated H100, H200, A100 and MI300X instances by the hour, and store your datasets right next to them. Transparent pricing, no annual contract required.
What you get
A simple way to rent serious accelerators.
Scout makes it straightforward to spin up modern GPU instances for training and inference, without negotiating an enterprise contract.
Dedicated GPU instances
Reserve whole accelerators by the hour. Your workload doesn't share a GPU with anyone else's.
High-bandwidth interconnect
Multi-GPU pods use the fastest interconnect each accelerator supports — NVLink on NVIDIA, Infinity Fabric on AMD.
Sensible security defaults
Per-tenant network isolation, customer-managed SSH keys, and no shared filesystem between tenants.
Fast time-to-launch
Pre-baked CUDA, PyTorch, vLLM and SGLang images. SSH or Jupyter as soon as your instance is ready.
Multi-region availability
Capacity across North America, Europe, and Asia-Pacific. Place workloads close to your data.
CPU + accelerator parity
Pair GPUs with modern CPU nodes for data pipelines, without leaving the private network.
Available accelerators
The accelerators teams actually want.
NVIDIA H100 80GB SXM5
- VRAM
- 80 GB HBM3
- Interconnect
- NVLink
- Throughput
- 1,979 TFLOPS FP16
NVIDIA H200 141GB
- VRAM
- 141 GB HBM3e
- Interconnect
- NVLink
- Throughput
- Optimized for inference
NVIDIA A100 80GB
- VRAM
- 80 GB HBM2e
- Interconnect
- NVLink
- Throughput
- 624 TFLOPS FP16
NVIDIA L40S
- VRAM
- 48 GB GDDR6
- Interconnect
- PCIe Gen4
- Throughput
- 362 TFLOPS FP16
Dataset storage
Store petabytes for the price of a coffee.
S3-compatible object storage co-located with every Scout region. Zero egress to your own GPU instances — pay only for what you keep.
No egress fees inside Scout · Free reads to attached GPU instances · Launch pricing, locked in for early teams.
# spin up an 8× H100 pod
scout launch \
--gpus h100:8 \
--image pytorch:2.4-cuda12.4 \
--duration 6h
→ instance scout-9af3 provisioning…
→ ssh root@scout-9af3.compute.scout.aiDeveloper experience
One CLI. One API. Less YAML.
Launch an instance, attach storage, and stream logs from your terminal. Bring your own container or use one of our pre-built ML images.
- Single-command launches with reproducible images
- Bring your own SSH keys; per-tenant network isolation
- Pre-installed vLLM, SGLang, TensorRT-LLM, Triton
- Per-second billing once your instance is running
Capacity is finite. Get in line.
Join the waitlist to reserve H100, H200, or MI300X capacity for your next training run. We onboard new teams in weekly cohorts.