CE runs on Kubernetes with Horizontal Pod Autoscaler configured to trigger at 60% CPU utilization. Your flood causes CPU to spike to 65%. Within 3 seconds, the HPA has provisioned 2 new replica pods. Within 8 seconds, 4 new nodes. Within 30 seconds, load per node is back to 15%. Your attack's effect on each individual node decreases exponentially as you scale up, because CE scales faster than you can add attack nodes. You need to grow your botnet faster than AWS can provision virtual machines. You can't.

Sending expensive HTTP/2 requests requires TLS handshakes, TCP connections, and complex request construction. Each of your attack requests costs ~0.5ms of your CPU to generate. CE's response (including rate-limiting and rejection): ~0.05ms. Cost ratio: 10:1 in CE's favor. Your botnet burns $1 of compute for every $0.10 CE spends defending. The auto-scaling cost for CE during your entire attack: approximately $0.12. Your attack cost: ~$12. You're paying 100x more to attack than CE pays to defend.

Expensive endpoints (/api/search, /api/export) are protected by circuit breakers with request complexity limits. A search query with a wildcard? Complexity score: HIGH. Maximum complexity score per IP per minute is budgeted. Exceeding the complexity budget triggers exponential backoff: first violation = 1s wait, second = 4s, third = 16s, fourth = permanent block. Complex queries from unknown IPs hit this limit in the first 5 requests. After that, they get 503 immediately with zero backend computation.

Application-layer floods typically repeat similar query patterns (limited creativity in botnet payloads). CE's request deduplication layer recognizes identical or near-identical requests within a 60-second window and returns cached results. Your "expensive" operation is computed exactly once per 60 seconds regardless of how many times you request it. Sending 100,000 requests for the same expensive operation? 1 computation, 99,999 cache hits. Total extra CPU cost: 0.001%.