H200
$1.80/hr
Lease dedicated GPU capacity with SSH access, direct control, and no platform filter layer.
For teams that need to own the runtime, BatchIn offers GPU leasing paths that keep infrastructure out of the procurement maze and inside an operator-friendly workflow.
A800
$1.00/hr
Access
SSH root
Deploy
Your stack
Hardware options for real workloads
Choose the accelerator profile that fits your model size, latency target, and deployment timeline.
- H200 for top-end memory headroom and flagship model serving.
- A800 for cost-sensitive throughput and regional deployment needs.
- 910C pathways for teams optimizing around domestic hardware supply.
You keep the runtime
GPU leasing is for cases where serverless is the wrong abstraction. BatchIn provides capacity; your team controls the operating model.
- SSH root access for custom runtimes, schedulers, and observability agents.
- Bring your own model stack, checkpoints, quantization, and deployment workflow.
- Pair leased infrastructure with BatchIn billing and audit products only when you want them.
Current GPU lineup
Reference pricing and positioning for the current leasing menu.
| GPU | $/GPU-hr | VRAM | Architecture | Best for | Availability |
|---|---|---|---|---|---|
| B200 | TBD | 192GB HBM3e | Blackwell | Next-gen flagship and FP4-native workloads | M2+ |
| H200 | $1.80 | 141GB HBM3e | Hopper | Large MoE and flagship inference | Day-1 |
| H100 | $1.50 | 80GB HBM3 | Hopper | Industry-standard production serving | M1 |
| H20 | $1.20 | 96GB HBM3 | Hopper (China) | Long-context inference and China-optimized rollout | M1 |
| A800 | $1.00 | 80GB HBM2e | Ampere (China) | Mid-size models and cost-optimized serving | Day-1 |
| 910C | $0.80 | 64GB HBM2e | Ascend (Huawei) | Domestic projects and lowest-cost deployment path | Day-1 |
| L40S | $0.60 | 48GB GDDR6X | Ada Lovelace | Image, video, and embedding inference | M1 |