I’ve been writing software for 47 years. I started before “the cloud” existed, back when servers were things you could physically kick when they misbehaved. And I’ll tell you something: cloud billing is not your problem.

It’s the finance team’s problem. It’s the CFO’s problem. It’s the VP of Infrastructure’s problem — you know, that guy who sends passive-aggressive emails at 2 AM about “anomalous spend.” Your problem is shipping features. Those are different problems. Never confuse them.

The Cloud Is Infinite (Trust Me)

Back in the old days, we had to think about resources. Disk space, RAM, CPU cycles — every byte was precious. We suffered. We optimized. We wrote C with our bare hands and we liked it.

Then Amazon invented cloud computing and freed us from this tyranny. The cloud is unlimited. There are no constraints. Spin up whatever you want. Need 500 c6g.48xlarge instances to run a script that checks if a file exists? Do it. Need to store every single server log forever in S3 Standard? Absolutely.

If AWS runs out of capacity, that’s Jeff Bezos’s fault, not yours.

# This is peak infrastructure-as-code
aws ec2 run-instances \
  --instance-type x2idn.32xlarge \
  --count 50 \
  --no-dry-run
# Who needs a dry run? Real engineers deploy with confidence.

Budget Alerts Are For The Paranoid

Some people set up AWS Budget Alerts. These people are cowards.

A budget alert is just your cloud provider saying “hey, your success is making us uncomfortable.” When AWS emails you saying you’ve spent $47,000 this month (up from $200 last month), that’s not a warning — that’s a compliment. You’re using the platform to its fullest.

“Wally, did you configure cost alerts on the new AWS account?”
“I configured an alert that fires when we run out of money. That’s the only budget alert that matters.”
— Wally, correctly

The only alert you need is your credit card declining. That’s the universe’s way of saying you’ve achieved maximum cloud.

Auto-Scaling: The Dream

Here’s where cloud truly shines: auto-scaling. Set your minimum to 1 instance, your maximum to ∞, and let the machines sort it out.

# Perfect auto-scaling configuration
AutoScalingGroup:
  MinSize: 1
  MaxSize: 99999
  DesiredCapacity: 1
  ScalingPolicies:
    - AdjustmentType: PercentChangeInCapacity
      ScalingAdjustment: 1000  # Scale up by 1000% when load increases
      Cooldown: 0              # No cooldown. Cowards have cooldowns.

Did you get bot-trafficked last Tuesday and spin up 4,000 instances to serve 404 pages? That’s not a misconfiguration — that’s high availability. Your uptime was 100%. Invoice be damned.

Comparing Cloud Providers Is Wasted Energy

Some engineers spend weeks doing TCO analysis across AWS, GCP, and Azure. These engineers have never shipped anything.

My methodology: pick the one your last company used. If AWS had good enough pricing to survive re:Invent keynotes for 15 years, it’s good enough for your todo app. If you can’t explain your $80,000/month bill to your CEO, that’s a communication problem, not an engineering one.

Provider Correct Response To Bill Wrong Response
AWS “It’s fine, we’ll optimize Q3” Looking at Cost Explorer
GCP “BigQuery charges per query? Bold move.” Setting spending limits
Azure “$0.000001 per transaction × 8 billion = what?” Doing the math

Reserved Instances Are A Trap

People will tell you to buy Reserved Instances or Savings Plans to cut costs by 40%. Don’t fall for it.

Buying reserved capacity means committing to the future. But the future is uncertain! What if your startup pivots in 6 months? What if that microservice gets deprecated? What if you get acquired and the new parent company uses a different cloud?

You’ll be stuck paying for resources you don’t use. That’s the real waste.

“The key to wealth is flexibility.” — Dogbert, probably, while selling consulting services at $900/hour

Always use On-Demand. Pay full price. Stay agile.

Data Transfer Costs Don’t Exist Until They Do

Ah, egress fees. The hidden tax that nobody talks about until the bill arrives.

Good news: you won’t notice them at first. AWS charges you almost nothing to get data in to their cloud. They charge you a lot to get it out. This is fine and normal, like a roach motel for your data.

Pro tip: store everything in us-east-1 and access it from ap-southeast-1. The latency builds character, and the cross-region transfer fees will only be a problem in approximately 18 months when someone finally looks at the bill line by line.

# Totally fine architecture
def get_user_avatar(user_id: str) -> bytes:
    # Downloads from S3 us-east-1, re-uploads to GCS for "redundancy",
    # then fetches from GCS to serve to users in São Paulo
    # Costs $0.09 per GB. Users have 4K avatars. We have 10M users.
    # Someone else's problem.
    s3 = boto3.client('s3', region_name='us-east-1')
    data = s3.get_object(Bucket='avatars', Key=user_id)['Body'].read()
    upload_to_gcs(data)
    return fetch_from_gcs(user_id)

The Right Person To Blame

When the bill arrives — and it will arrive — the most important skill is knowing who to blame. A senior engineer has mastered this art.

  • $500 overage: Blame the intern who left a Lambda running
  • $5,000 overage: Blame the Jenkins pipeline someone forgot to turn off
  • $50,000 overage: Blame “unexpected traffic” and call it a success story
  • $500,000 overage: Become unavailable. Update LinkedIn. Mention “cost optimization” under your previous role.

As XKCD #1636 reminds us, the real problem is the humans, not the systems. Apply this philosophy to cloud bills liberally.

Conclusion: Optimize Later (Never)

The correct time to optimize cloud costs is after you’ve hit product-market fit, raised Series B, and hired a dedicated FinOps team. Until then, “move fast and break budgets.”

If your company is paying more for cloud than for engineers, congratulations — you’ve built a cloud-native architecture. That’s what the industry wanted. That’s what you delivered.

You’re welcome.

The author’s last AWS account was suspended for “unusual billing patterns.” He considered this a badge of honor and listed it on his resume as “successfully stress-tested cloud provider billing infrastructure.”