Cost Overview

AIQLick's AWS infrastructure runs in eu-north-1 (Stockholm) with a small footprint optimized for an early-stage platform. This page breaks down the monthly cost structure and documents the optimization strategies in place.

Monthly Cost Breakdown

Compute (EC2)

Instance	Type	vCPU	RAM	Approx. Cost/mo
aiqlick-backend	t3.small	2	2GB	~$15
aiqlick-ai	t3.small	2	2GB	~$15
aiqlick-backend-dev	t3.micro	2	1GB	~$7.50
aiqlick-ai-dev	t3.micro	2	1GB	~$7.50
aiqlick-jitsi	t3.xlarge	4	16GB	~$60
EC2 Subtotal				~$105/mo

note

Prices are approximate on-demand rates for eu-north-1. Actual costs may vary slightly based on data transfer and EBS usage.

Database (RDS)

Instance	Class	Engine	Storage	Approx. Cost/mo
aiqlick-postgres	db.t4g.small	PostgreSQL 16.6	50GB gp3	~$25
aiqlick-postgres-dev	db.t4g.micro	PostgreSQL 16.10	20GB gp3	~$13
Storage (gp3, 70GB total)				~$8
Automated backups				~$4
RDS Subtotal				~$50/mo

Storage & CDN

Service	Details	Approx. Cost/mo
S3	6 buckets (uploads, dev-uploads, backups, deployments, docs, docs-dev)	~$5
CloudFront	4 distributions (storage + docs, prod + dev), PriceClass_100	~$5-10
Storage Subtotal		~$10-15/mo

Container Registry (ECR)

Repository	Image Size	Approx. Cost/mo
aiqlick-backend	~240MB	~$1
background-tasks	~699MB	~$2
aiqlick-meeting	~142MB	~$0.50
aiqlick-jigasi	~505MB	~$1.50
ECR Subtotal		~$5/mo

AI Services (Bedrock)

Model	Pricing Model	Cost Driver
Claude (Anthropic)	Per input/output token	Agent chat, text rewrite, meeting insights
Titan Embed v2	Per input token	RAG document/query embedding

Bedrock costs are pay-per-token and scale directly with platform usage. At low usage, costs are minimal ($5-20/mo). At scale, this becomes the dominant cost category.

info

All Bedrock token consumption is tracked in the AIOperationLog table with input/output token counts and mapped credit costs. The credit system charges users at configurable rates per model via CreditCostConfig.

Other Services

Service	Details	Approx. Cost/mo
Amplify	Build minutes + Next.js SSR hosting (dev + prod branches)	~$10-20
Secrets Manager	3 secrets at $0.40/secret/month	~$1.20
Rekognition	DetectFaces API calls (per-image pricing)	~$1-5
Transcribe	Real-time streaming (per-second pricing) — meetings + voice STT	~$2-10
Polly Neural	Text-to-speech for voice chat ($16/1M chars)	~$1-5
CloudWatch	Logs and basic monitoring	~$2-5
Data Transfer	Inter-region + internet egress	~$5-10
Other Subtotal		~$25-55/mo

Total Infrastructure Cost

Category	Monthly Range
EC2	~$105
RDS	~$50
S3 + CloudFront	~$10-15
ECR	~$5
Amplify	~$10-20
Other (Secrets, CloudWatch, Transfer)	~$10-20
Total (excluding Bedrock)	~$180-200/mo
Bedrock (usage-dependent)	$5-100+/mo

Current Optimization Strategies

The infrastructure is designed for cost efficiency at the current stage of the platform.

Burstable Instances (t3 Family)

All EC2 instances use t3 burstable instances, which accumulate CPU credits during idle periods and spend them during load spikes. This is significantly cheaper than fixed-performance instances for workloads with variable CPU demand.

Instance Type	On-Demand	Equivalent Fixed (m5)	Savings
t3.micro	~$7.50/mo	~$35/mo (m5.large min)	~78%
t3.small	~$15/mo	~$35/mo (m5.large min)	~57%
t3.xlarge	~$60/mo	~$70/mo (m5.xlarge)	~14%

Tiered Dev Environments

Development instances use smaller (and cheaper) instance classes than production:

EC2: t3.micro (dev) vs t3.small (prod) -- 50% cost reduction
RDS: db.t4g.micro (dev) vs db.t4g.small (prod) -- ~48% cost reduction
EBS: 8GB (dev) vs 20-30GB (prod) -- smaller disk allocations

CloudFront PriceClass_100

Both CloudFront distributions use PriceClass_100, which limits edge locations to the cheapest regions (North America and Europe). This is appropriate since the primary user base is in the Nordics and Europe.

Single-AZ RDS

Both RDS instances run in a single Availability Zone without Multi-AZ failover. This halves the RDS compute cost. For the current stage of the platform, the trade-off of slightly higher downtime risk during AZ failures is acceptable.

External Redis

Redis Cloud is used instead of Amazon ElastiCache. A small external Redis instance is cheaper than the minimum ElastiCache node (cache.t4g.micro at ~$12/mo) and requires no VPC configuration overhead.

No NAT Gateway

EC2 instances have public IPs and route directly to the internet, avoiding the ~$32/mo fixed cost of a NAT Gateway per AZ.

Minimal EBS Storage

EBS volumes use gp3 (the cheapest general-purpose SSD tier) with conservative sizing:

Instance	EBS Size	Rationale
Backend (prod)	20GB	Docker images + logs
AI (prod)	30GB	Larger ML-related images
Dev instances	8GB	Minimal footprint
Jitsi	20GB	Docker Compose services (6 containers)

Future Optimization Options

As the platform scales, several cost optimization strategies become worthwhile:

Reserved Instances / Savings Plans

For instances running 24/7, Reserved Instances or Compute Savings Plans offer 30-40% savings over on-demand pricing.

Option	Commitment	Typical Savings
1-year Reserved Instance (no upfront)	1 year	~30%
1-year Reserved Instance (all upfront)	1 year	~38%
3-year Reserved Instance (all upfront)	3 years	~55%
Compute Savings Plan	1-3 years	~30-50%

tip

Compute Savings Plans are more flexible than Reserved Instances because they apply across instance families, sizes, and regions. They are the recommended option once instance usage patterns stabilize.

ECR Lifecycle Policies

ECR currently retains all image tags indefinitely. Adding lifecycle policies to automatically expire old images would reduce storage costs and keep repositories clean.

{
  "rules": [
    {
      "rulePriority": 1,
      "description": "Keep only last 10 images",
      "selection": {
        "tagStatus": "any",
        "countType": "imageCountMoreThan",
        "countNumber": 10
      },
      "action": {
        "type": "expire"
      }
    }
  ]
}

Graviton (ARM) Instances

AWS Graviton processors (ARM-based) offer approximately 20% better price-performance than equivalent x86 instances. Both the backend (Node.js) and background-tasks (Python) workloads are compatible with ARM.

Current	Graviton Equivalent	Savings
t3.small (x86)	t4g.small (ARM)	~20%
t3.micro (x86)	t4g.micro (ARM)	~20%
t3.xlarge (x86)	t4g.xlarge (ARM)	~20%

note

The RDS instances already use Graviton (db.t4g family). EC2 instances are the remaining migration target.

ALB + Auto Scaling

Currently, each service runs as a single Docker container on a single EC2 instance. When traffic justifies it, moving to an Application Load Balancer with Auto Scaling Groups would provide:

Automatic horizontal scaling based on CPU/memory thresholds
Health-check-based instance replacement
Zero-downtime deployments via rolling updates
Multi-AZ redundancy

The trade-off is added complexity and the base cost of an ALB (~$16/mo + per-request charges).

RDS Multi-AZ

Enabling Multi-AZ on the production RDS instance would provide automatic failover to a standby replica in another Availability Zone. This doubles the RDS compute cost but provides higher availability for production workloads.

Spot Instances for Dev

Development EC2 instances could run on Spot pricing for up to 90% savings. The trade-off is that AWS can reclaim the instance with 2 minutes notice, which is acceptable for non-production environments.

Bedrock Cost Management

Bedrock (LLM inference) is the only usage-based cost that scales with platform activity. The platform tracks and manages these costs through several mechanisms:

Cost Tracking

Every AI operation is logged in the AIOperationLog table with:

Model ID and operation type (CV_EXTRACTION, JOB_PARSING, MATCHING, AGENT_CHAT, etc.)
Input and output token counts
Calculated credit cost based on CreditCostConfig rates

Credit System Mapping

Users pay for AI operations through the credit system:

1 credit = $1 of token consumption
Credit costs are configurable per model via the CreditCostConfig admin table
The BillingGate in background-tasks performs a pre-check before each AI operation and deducts credits after completion
Failed operations are not charged

Cost Control Mechanisms

Mechanism	Description
Credit balance limits	Users cannot exceed their purchased or allocated credits
Plan-based allocation	Each subscription plan includes a monthly credit allocation
`BillingGate` pre-check	Rejects operations when balance is insufficient
`BILLING_ENABLED` env var	Gradual rollout of billing enforcement
Per-company config	`CompanyBillingConfig` allows per-company credit allocation overrides

Monitoring

# Check Bedrock invocation metrics
aws cloudwatch get-metric-statistics \
  --profile Administrator-842697652860 \
  --region eu-north-1 \
  --namespace AWS/Bedrock \
  --metric-name Invocations \
  --start-time $(date -u -v-7d +%Y-%m-%dT%H:%M:%S) \
  --end-time $(date -u +%Y-%m-%dT%H:%M:%S) \
  --period 86400 \
  --statistics Sum \
  --output table

Monthly Cost Breakdown​

Compute (EC2)​

Database (RDS)​

Storage & CDN​

Container Registry (ECR)​

AI Services (Bedrock)​

Other Services​

Total Infrastructure Cost​

Current Optimization Strategies​

Burstable Instances (t3 Family)​

Tiered Dev Environments​

CloudFront PriceClass_100​

Single-AZ RDS​

External Redis​

No NAT Gateway​

Minimal EBS Storage​

Future Optimization Options​

Reserved Instances / Savings Plans​

ECR Lifecycle Policies​

Graviton (ARM) Instances​

ALB + Auto Scaling​

RDS Multi-AZ​

Spot Instances for Dev​

Bedrock Cost Management​

Cost Tracking​

Credit System Mapping​

Cost Control Mechanisms​

Monitoring​