← Back
GCE vs Cloudflare
Infrastructure cost comparison based on Dolphin AI's original deployment
Source Found in codebase
Assumed Estimated/inferred
What We Found in the Repos
GCP Project Configuration Source
| Project ID |
dolphin-ai |
gce-clusters/dev/config.json:3 |
| Cluster Name |
dolphinai-dev |
gce-clusters/dev/config.json:4 |
| Region/Zone |
us-central1-f |
gce-clusters/dev/config.json:5 |
| Container Registry |
gcr.io/dolphin-ai/ |
tugboat/README.md:10 |
Domains & SSL Source
| Production App |
demo.dolphinai.com |
gce-clusters/dev/specs/demo-relay-ingress.yaml:12 |
| API Endpoint |
api.demo.dolphinai.com |
gce-clusters/dev/specs/demo-go-ingress.yaml:12 |
| SSL Provider |
Let's Encrypt via kube-lego |
gce-clusters/dev/specs/kube-lego-deployment.yaml |
AWS GPU Infrastructure Source
| Damage Detection GPU |
52.200.249.199:5000 |
api/Dockerfile:80 |
| Instance ID |
i-7e0b77e7 |
api/Dockerfile:84 |
| Shingle Detection GPU |
34.226.23.26:5000 |
api/Dockerfile:87 |
| Instance ID |
i-01cdb295550706beb |
api/Dockerfile:90 |
Monthly Cost Estimates
Costs below combine sourced infrastructure details with assumed usage patterns and current (Jan 2026) cloud pricing.
Google Cloud (Original Stack)
| Item | Details | Monthly | Source |
|---|
| GKE Cluster |
~3 n1-standard-2 nodes (assumed) |
$219 |
Assumed |
| GKE Management |
$0.10/hr per cluster |
$73 |
GCP Pricing |
| Load Balancer |
Ingress from demo-relay-ingress.yaml |
$45 |
Source |
| Cloud SQL |
db-n1-standard-1, ~100GB (assumed) |
$97 |
Assumed |
| Elasticsearch |
Self-hosted on GKE (from docker-compose) |
$73 |
Source |
| Cloud Storage |
~500GB aerial imagery (assumed) |
$10 |
Assumed |
| Egress |
~100GB/mo (assumed) |
$12 |
Assumed |
| AWS GPU (Damage) |
EC2 i-7e0b77e7 (from Dockerfile) |
$700 |
p2.xlarge assumed |
| AWS GPU (Shingle) |
EC2 i-01cdb... (from Dockerfile) |
$700 |
p2.xlarge assumed |
| Total (Original) |
$1,929 |
|
Cloudflare + Minimal GCP (Modern Alternative)
| Item | Details | Monthly | Source |
|---|
| Cloudflare Pages |
Static frontend hosting |
$0 |
CF Pricing |
| Cloudflare CDN |
Global edge + DDoS protection |
$0 |
CF Pricing |
| R2 Storage |
500GB, zero egress |
$7.50 |
CF Pricing |
| Cloud Run (API) |
Go API, auto-scaling |
$25 |
Assumed usage |
| Cloud SQL (small) |
db-f1-micro |
$15 |
GCP Pricing |
| Search (Typesense) |
Replaces Elasticsearch |
$29 |
Typesense Pricing |
| ML Inference |
Pay-per-call (Replicate/Modal) |
~$10 |
Low volume assumed |
| Total (Modern) |
$87 |
|
~$1,842/month potential savings
Heavily assumption-dependent. GPU costs dominate the original stack.
Modern pay-per-inference eliminates idle GPU costs.
Key Assumptions & Caveats
- GKE node count: Assumed 3 nodes based on typical small production clusters. Not found in config files.
- Instance types: Assumed n1-standard-2 for GKE, p2.xlarge for AWS GPUs. Instance IDs found but not types.
- Storage volume: 500GB assumed for aerial imagery. No actual usage data available.
- Traffic/egress: 100GB/month assumed. No analytics or billing data available.
- GPU uptime: Assumed 24/7 on-demand. May have used spot/preemptible or scheduled shutdown.
- Modern ML costs: Assumed low inference volume (~$10/mo). Actual usage could be much higher.
- Prices: Based on January 2026 public cloud pricing. 2017-2021 prices were different.
Source Files Referenced
| gce-clusters/dev/config.json |
GKE cluster config, project ID, service definitions |
| gce-clusters/dev/specs/*.yaml |
Kubernetes deployments, ingress rules, services |
| api/Dockerfile |
AWS GPU IPs, instance IDs, environment variables |
| lagoon/docker-compose.yml |
Local dev services (MySQL, ES, neural-nets) |
| tugboat/README.md |
Container registry, deployment process |