The Small Business Guide to AWS Costs (It’s Cheaper Than You Think)

“AWS will bankrupt us.” I hear this once a week from SMB owners considering automation. They’ve heard the horror stories. But for a well-architected serverless workload, AWS is actually cheap. Let me show you the real numbers.

The Free Tier (Your Secret Weapon)

AWS offers a free tier that resets monthly. For the first 12 months, and then permanently for many services:

Lambda: 1 million function invocations per month. Free. If your automation runs 10,000 times per month (a lot), you’re at 10,000 invocations. You have 990,000 left over. Processing invoices once a day? 30 invocations/month. Essentially free forever.

DynamoDB: 25 GB of storage and 25 write capacity units. Free on on-demand pricing. For most SMBs, a DynamoDB table that stores metadata — conversation history, job status, extracted data — uses less than 1 GB. Well under the free tier.

S3: 5 GB of storage per month. Free. Your invoice PDFs, images, documents — if you’re storing less than 5 GB, it’s free.

API Gateway: 1 million API calls per month. Free. If you have a webhook that receives data, processes it, and returns a response, this is free.

CloudWatch: 10 GB of logs per month. Free. Your Lambda logs, error tracking, everything — free.

When You Go Beyond Free Tier

Let’s say your automation is wildly successful and you blow past the free tier. Good problem to have.

Lambda: $0.0000002 per invocation after 1M/month. So 10 million invocations per month (unreasonably high for an SMB) = 9 million chargeable invocations × $0.0000002 = $1.80/month.

DynamoDB (on-demand): $0.25 per million write capacity units, $1.25 per million read capacity units. If you’re writing 100,000 items per month and reading them 200,000 times: ~$0.03 in writes + ~$0.25 in reads = $0.28/month.

S3 (storage): $0.023 per GB per month. 100 GB of invoices (a lot) = 100 × $0.023 = $2.30/month.

S3 (requests): $0.0004 per 1,000 PUT requests, $0.0000004 per 1,000 GET requests. 1 million reads per month = $0.0004. Negligible.

API Gateway: $3.50 per million API calls after 1M free. 2 million calls = 1 million chargeable × $3.50 = $3.50/month.

CloudWatch (logs): $0.50 per GB ingested after 10 GB free. 50 GB per month = 40 chargeable × $0.50 = $20/month.

Claude API: This is from Anthropic, not AWS, but it’s part of your bill. $3 per million input tokens, $15 per million output tokens. Processing 500 invoices per month, 100 tokens in + 50 tokens out = $0.53/month.

A Real Example: Invoice Processing at Scale

Let’s say your automation is humming. You process 1,000 invoices per month. Each triggers a Lambda, writes to DynamoDB, reads from your PO database, and logs everything to CloudWatch.

Add a buffer for unexpected costs, monitoring tools, or secrets management: $5/month is realistic.

Compare that to the alternatives:

• Zapier: $100–300/month
• Custom engineering team: $8,000–15,000/month salary (even junior)
• Outsourced data entry: $1,500–3,000/month

AWS is the cheapest option by far.

The Scary Stories (And Why They Don’t Happen)

Story 1: Runaway Lambda

A badly written Lambda loops infinitely, invokes itself 1 million times in an hour, and racks up a $1,000 bill.

Why this doesn’t happen: Set up billing alerts. AWS emails you when your bill hits $10. You kill the Lambda immediately. Cost: $2, not $1,000. Also, well-written Lambda with timeout, limits, and error handling prevents this entirely.

Story 2: Cryptocurrency miner

A hacker compromises your AWS credentials and spins up 100 GPU instances to mine crypto.

Why this doesn’t happen: Use IAM roles. Your Lambda doesn’t have credentials to create EC2 instances. Even if a hacker gets in, they can only do what that specific role allows. Also, billing alerts catch the $10K spike in 5 minutes.

Story 3: DDoS-triggered costs

Your API Gateway gets hammered with traffic. AWS scales automatically. Your bill explodes.

Why this doesn’t happen: AWS DDoS protection is automatic. Also, you can set API Gateway to throttle at a specific request rate, so costs are capped.

Cost Optimization (Actually Free)

• Right-size compute. Lambda defaults to 128 MB memory. If you only need 256 MB, you’re not wasting anything. If you need 3 GB, upgrade — it costs $0.0000017 more per invocation. Proportional.
• Use on-demand DynamoDB. Don’t guess capacity; pay for what you use.
• Delete old data. Old S3 files, old CloudWatch logs — set a retention policy (30 days by default). Don’t keep forever.
• Use Lambda layers for shared code. Reduces package size, slightly reduces cold start time.
• Monitor with CloudWatch dashboards. See what’s actually happening. Turn off verbose logging if you don’t need it.

Budgeting

Set a monthly budget in AWS Billing. Tell it: “Alert me if I’m on track to spend >$50/month.”

AWS will email you if you’re trending toward that limit. For an SMB automation, hitting that limit means something is broken — runaway process, bug, or you’re wildly successful and need to rethink.

Bottom Line

AWS is expensive if you’re running servers 24/7 for a team of 1,000. AWS is cheap if you’re running serverless automation for an SMB.

For a typical invoice processing, document intake, or reporting automation, you’re looking at $2–15/month in infrastructure costs. Add a competent engineer (one-time, $5–10K), and your all-in cost is under $15K to build and under $200/year to run.

That ROI — saving 50+ hours of manual labor every month — is hard to beat.