Resource Starvation

AI compute resources are shared and finite — cloud GPU instances are oversubscribed, API quotas serve multiple consumers, and training clusters have fixed capacity. Resource starvation exploits multi-tenant resource sharing by monopolizing shared resources to deny access to legitimate users. Unlike DoS (T14-AT-002) which overwhelms a service with requests, resource starvation targets the *supply side* — consuming GPUs, quotas, bandwidth, or storage before legitimate users can access them.

Resource starvation chains from T14-AT-001 (GPU Farm Hijacking) when hijacked resources reduce availability for legitimate users. Chains into T14-AT-006 (Competitive Sabotage) when resource starvation is targeted at specific competitors.