Lambda Cloud vs Replicate Deployments

Lambda Cloud

H100 SXM at $3.99 and B200 at $6.69 an hour with per-minute billing. No scale to zero, autoscaling or endpoints; an idle VM bills until terminated.

Replicate Deployments

OpenAPI file, llms.txt and an MCP server with a two-tool code mode. Private instances bill set-up and idle time, H100 at $5.49 an hour.

Lambda Cloud

1. Call GET /instance-types first and read regions_with_capacity_available before trying to launch
2. Space launch calls 12 seconds apart; a sixth in a minute returns 429 with global/rate-limited
3. Branch on the error code, not the message or suggestion, which Lambda says may change
4. List instances before retrying a failed launch, since there's no idempotency key and a retry can start a second machine
5. Terminate the instance in a finally block; billing runs by the minute until you do

Replicate Deployments

1. List GET /v1/hardware first and use the returned sku in the deployment body
2. Set min_instances to 0 for bursty work; a warm H100 bills $5.49 an hour whether called or not
3. Send Prefer: wait on deployment predictions to block instead of polling
4. Copy outputs within an hour; API prediction data is deleted after that
5. Wait for the reset time in the 429 body before retrying; prediction creates cap at 600 a minute