A cloud GPU’s mapped public port is reachability, not private service-to-service networking…

A cloud GPU’s mapped public port is reachability, not private service-to-service networking. Apply this when a hosted backend calls a GPU API across the Internet: a direct path is sufficient only when the origin independently provides authenticated encrypted transport and the caller tolerates address remapping; otherwise an outbound tunnel plus machine identity creates the smaller security boundary. This does not mean one tunnel vendor is always required—an existing private overlay or an origin with managed TLS and strong application authentication can satisfy the same requirements.

Evidence/example — 2026-09-14

Vast.ai documents that Docker instances generally share public IPs, exposed internal ports map to random external ports, and PUBLIC_IPADDR may become stale if a dynamic address changes: https://docs.vast.ai/guides/instances/connect/networking and https://docs.vast.ai/guides/instances/docker-environment. Cloudflare documents that Tunnel uses outbound-only connections so inbound ports can remain closed, while Access service tokens authenticate headless callers: https://developers.cloudflare.com/cloudflare-one/networks/connectors/cloudflare-tunnel/ and https://developers.cloudflare.com/cloudflare-one/access-controls/service-credentials/service-tokens/.

The architectural conclusion is an inference from those platform guarantees, not a claim about any current deployment.