Cost model
A BYOC deployment produces two separate bills:- AWS: The data plane runs in your own AWS account, so AWS bills you directly for the infrastructure it creates.
- LangSmith: LangChain invoices you based on your contract and usage.
Billable AWS services
Every data plane creates the following resources in your account. AWS charges for them at its own rates, independently of your LangSmith contract.Data transfer
There are a few data transfer cost drivers in BYOC:- Cross-availability-zone traffic, since the data plane runs highly available across zones.
- Egress through the NAT gateway, for example if calling model providers, or an external API through a webhook.
- Public IPv4 addresses, which AWS charges per address-hour. The regional NAT gateway uses one, and an internet-facing load balancer uses another when public ingress is enabled.
Service limits
AWS enforces default quotas per region and per account. Defaults are often too low for production on new or lightly used accounts, so check the following before you deploy, and request increases where needed:- EC2 vCPU quotas: At minimum, 64 vCPU of On-Demand capacity, covering the 3 system nodes and Karpenter workloads.
- Subnet IP capacity: Room for pods, nodes, EKS ENIs, load balancers, and at least 12 interface endpoints per application subnet.
- VPC endpoints: Capacity for 12 interface endpoints and 1 S3 gateway endpoint.
- Load balancers: Capacity for 2 Network Load Balancers and their target groups.
- EKS: Capacity for 1 EKS cluster.
- Storage: EBS volume and storage capacity for ClickHouse, ZooKeeper, node disks, and other persistent workloads.
- Databases: Capacity for 2 RDS instances and 3 ElastiCache replicas.
- PrivateLink and networking: Capacity for 1 endpoint service and 1 regional NAT gateway.
See also
Connect these docs to Claude, VSCode, and more via MCP for real-time answers.