Dedicated Capacity

Dedicated Capacity for production workloads, with monthly contracts first and capacity economics made explicit.

BatchIn positions this as sales-assisted Dedicated Capacity, not a self-serve hardware mall. The core offer is private serving capacity with delivery support, SSH/root-level control, and predictable monthly economics.

Talk to infrastructure team Review pricing and models

Monthly capacity

From ¥8,000 / month

Capacity-hour

From ¥1.4 / capacity-hr

Access

SSH / root-capable delivery

Commercial motion

Sales-assisted Dedicated Capacity

Built for procurement, not toy benchmarking

Teams buying Dedicated Capacity care about contractability, queue stability, deployment control, data-residency fit, and long-run unit economics. That is the frame for this page.

Monthly capacity is the primary commercial unit; hourly numbers are a derived planning aid.
Every SKU is framed as an 8-GPU cluster baseline to match real project sizing.
Delivery can be paired with BatchIn Managed routes, Hybrid fallback, audit traces, and Private Cluster controls when the workload needs it.

Operational control stays with the customer team

Serverless is not always the right abstraction. Dedicated Capacity is for teams that need runtime ownership, custom stacks, no-cloud mode, and explicit infrastructure boundaries.

SSH access and operator-grade delivery for custom runtimes and deployment pipelines.
Suitable for model training, long-context inference, private batch lanes, regional deployment, and regulated workloads.
Commercial path includes sizing, deployment planning, and capacity reservation rather than instant checkout.

Renewable-powered delivery available

Why teams ask for Dedicated Capacity

H200 SXM

42,000 / month

High-memory Hopper lane for flagship MoE and long-context inference.

H800 SXM

36,000 / month

Regional Hopper lane for teams balancing supply and flagship throughput.

A800 80GB

20,000 / month

A pragmatic production choice for balanced performance and regional deployment.

A100 80GB

22,000 / month

Mature Ampere option when software compatibility matters more than peak headline specs.

H100 SXM

40,000 / month

Operator-standard Hopper deployment lane for high-end production inference.

L40S 48GB

8,000 / month

Better economics for image, video, embedding, and lighter multimodal loads.

B200 SXM

Contact sales

Reserved for next-generation Blackwell demand and handled via custom quotation.

H20 96/141G

12,000 / month

Regional Hopper option for long-context and controlled delivery requirements.

Current 8-GPU capacity menu

Monthly CNY pricing is primary. Capacity-hour and cluster-hour values are provided as planning references only.

Capacity SKU	Monthly (8 cards)	Capacity-hour	Cluster hour	VRAM	Architecture	Best for	Availability
H200 SXM	42,000 CNY	7.2 / GPU-hr	57.6 / cluster-hr	141GB HBM3e	Hopper	Flagship MoE inference, long-context, training prep	Sales-assisted
H800 SXM	36,000 CNY	6.2 / GPU-hr	49.6 / cluster-hr	80GB HBM3	Hopper	Regional flagship inference	Sales-assisted
A800 80GB	20,000 CNY	3.4 / GPU-hr	27.2 / cluster-hr	80GB HBM2e	Ampere	Balanced production inference	Available
A100 80GB	22,000 CNY	3.8 / GPU-hr	30.4 / cluster-hr	80GB HBM2e	Ampere	Stable software stack, mature production deployment	Available
H100 SXM	40,000 CNY	6.8 / GPU-hr	54.4 / cluster-hr	80GB HBM3	Hopper	High-end production inference	Sales-assisted
L40S 48GB	8,000 CNY	1.4 / GPU-hr	11.2 / cluster-hr	48GB GDDR6	Ada Lovelace	Image, video, embedding, lighter multimodal	Available
B200 SXM	Contact sales	Contact sales	Contact sales	192GB HBM3e	Blackwell	Next-gen flagship demand Custom quotation only	Custom quotation
H20 96/141G	12,000 CNY	2.1 / GPU-hr	16.8 / cluster-hr	96GB / 141GB	Hopper	Regional Deployment and long-context	Available on request