RunComfy compute and growth docket#

Status: Decision log Scope: Current hosted sessions, serverless deployments and training/model APIs; older competitive assertions are historical.

Compute and procurement#

Current product: native ComfyUI sessions can save models, nodes and settings into reusable environments, then expose workflows as autoscaling endpoints. Training and ready-model generation are separate products sharing a balance. Published GPU rates include T4/A4000 $0.99/hour, A10G/A5000 $1.75, A6000 $2.50, L40S/L40 $2.99 (API-only), A100 $4.99, H100 $7.49 and H200 $9.59; Pro rates are discounted. Training H100/H200 prices differ and must not be substituted for inference. Session duration and serverless lifecycle billing are not identical just because both use seconds. Pricing, API product.

The inspected current sources disclose service orchestration, not a procurement contract. Earlier repository notes call Runpod white-labeling likely, then sometimes treat it as fact. That is an unverified hypothesis in this pass, not proof of an exclusive supplier, ownership or margin.

Acquisition and monetization#

PAYG needs no subscription; Pro is $29.99/month or $19.99 monthly equivalent billed annually, including $10 monthly credits, discounted eligible GPUs, setup CPU hours and larger persistent storage. Pro does not remove usage charges. A shared balance reduces switching friction between sessions, models and training. Current offer.

A searchable gallery of ready workflows, node/tutorial landing pages, classroom support and links reproducing a creator's environment provide discovery and sharing channels. Inference: successful template execution can lead users from exploration into repeat sessions and production APIs. No measured channel conversion is supplied. The API's existence is established; the old claim that the developer segment is unserved is obsolete. Product and workflow distribution.

Outcomes and implication#

No audited current user/revenue series or utilization was established. Claimed download-speed improvements and testimonials do not prove fleet economics. Compare setup portability, reliable API outputs, session idle cost and private model support; do not assert percentage savings without a matched workload. Supplier identity/title, wholesale rates, active paid users and conversion remain unresolved; no checkout or workload was tested.

Earlier profile is preserved. Sources retrieved October 4 local; batch manifest, coverage index.