Beam as cloud orchestrator - Per-second serverless, per-machine on-demand, BYOC

Fetched 2026-10-04. Beam is a cloud orchestrator with three billing models. Key structural facts: (1) Serverless GPU Tasks: billed by the millisecond, only while code runs. RTX 4090 $0.000192/sec, H100 $0.000972/sec, RTX 5090 $0.000303/sec. CPU $0.0000125/sec/core, RAM $0.0000021/GiB. (2) On-demand machines: one flat price per machine including vCPU+RAM+NVMe. H100 $1.83/hr, A100 $1.36/hr, L40S $0.76/hr, RTX 4090 $0.44/hr. (3) BYOC: $0.019/hr per vCPU + $0.009/hr per GB RAM management fee, compute billed to your own cloud account. (4) No egress or bandwidth fees on any plan - key differentiator. (5) Storage $0.021/GB/month over 1TB. (6) Free tier: $0/month + usage, 5 GPU containers. (7) Team: $89/month. Beam's per-second billing is the finest granularity available in the market - most clouds bill per-hour or per-second with 60-second minimums. The on-demand machine rates are competitive with raw cloud providers because Beam includes vCPU, RAM, and NVMe in the price.