ECO-AIM-AI-005 — Always-on inference endpoints¶

Category: AI/ML (AIM)
Family: AI (AI)
Layer: AI
Tier: 3
Severity: warning
Tags: ai, scale-to-zero
Legacy ID: ECO-AI-005

Summary¶

Always-on endpoints waste baseline compute when idle.

Idle accelerators still burn power; scale needs boundaries.

{
  "confidence": 0.6,
  "notes": "High for GPU-backed endpoints.",
  "type": "carbon"
}

{
  "languages": [
    "infra"
  ],
  "method": "trace"
}

{
  "guidance": "Autoscale; scale-to-zero where feasible; consolidate endpoints.",
  "tradeoffs": "Cold starts and scheduling complexity."
}

{
  "system_layers": [
    "ai"
  ]
}