Every cloud bill I've seen is at least double what the calculator said it would be. That's not hyperbole - I've watched companies budget $5k/month and pay $12k, then spend months trying to figure out where the money went.
What the Calculators Don't Tell You
The AWS pricing calculator is like getting a quote for a car that doesn't include the engine. It shows you compute costs but ignores data transfer between your services, which ends up being 30-40% of your bill.
When we moved from a monolith to microservices on AWS, our data transfer costs exploded from maybe $50/month to something like $2,800/month. The fucking calculator showed zero for that because AWS apparently assumes your services exist in complete isolation. Nobody told us about this during the migration - found out when the bill arrived. In reality, every API call between services in different AZs costs money, and with 12 microservices chattering constantly, those $0.01/GB charges add up fast.
Azure's calculator is slightly better but completely fucked if you need Windows licensing. The hybrid benefit thing is impossible to figure out - we ended up paying Microsoft Enterprise support $500/hour to explain our own licensing to us.
Google's calculator is the most honest of the three pieces of shit, but still misses operational overhead. You'll spend $1000/month minimum just monitoring and managing your cloud spend across these platforms. Datadog alone costs us $800/month.
AWS: The Hidden Fee Champion
AWS makes money by making their pricing impossible to understand. I've never seen an AWS bill that didn't have surprise charges nobody could explain.
What Actually Costs Money on AWS:
EC2 pricing looks reasonable until you realize t2.micro breaks under any real load. You end up on t3.medium instances minimum, and those cost $30/month each. Multiply by however many services you're running.
RDS Multi-AZ literally doubles your database bill but if you don't use it and your database goes down, you're fucked. So you pay it. Our PostgreSQL bill went from around $180/month to something like $360/month for this "feature" - learned that the hard way after a single-AZ outage took us down for 3 hours.
Application Load Balancer starts at $16/month but those LCU charges are brutal. We got hit with a $400 load balancer bill because we didn't understand how AWS counts "capacity units."
AWS Support is Expensive as Hell:
Business Support costs 10% of your monthly bill with a $100 minimum. So if you're spending $3k/month, you pay another $300 just to get someone on the phone when shit breaks. Enterprise Support starts at $15,000/month.
The free support is completely useless - they just point you to documentation. We paid $8,000 in support costs last year because their docs are deliberately incomplete for anything complex.
Data Transfer Costs Are Insane:
Cross-AZ transfer costs $0.01/GB each direction. Sounds cheap until your microservices generate 500GB/month talking to each other. That's $10/month per 500GB, and it adds up fast.
We ended up spending something like $1,200/month just on services in different availability zones talking to each other. Moving everything to a single AZ cut our bill by maybe 25% but increased our "all our eggs in one basket" risk - classic cloud tradeoff.
Azure: Licensing Hell Made Worse
AWS pricing is fucked, but Azure isn't much better - they just confuse you with different licensing bullshit instead. If you're already paying for Office 365 and Windows Server, Azure makes some sense. But their pricing is designed to confuse you into paying more.
What Azure Actually Costs:
VM pricing looks competitive but they force you to buy managed disks separately. Those add $50-200/month per VM depending on performance tier. AWS includes the disk in the instance price.
Azure SQL Database pricing is completely fucked. Business Critical costs $4.35 per vCore per hour. A 4-vCore database costs $3,132/month. We moved to PostgreSQL on VMs and cut our database costs 70%.
Premium SSD storage costs $0.12/GB/month. AWS gp3 is $0.08/GB/month. That adds up - 1TB costs $120 vs $80 on AWS.
Microsoft Licensing is Impossible:
Azure Hybrid Benefit supposedly saves 76% on Windows costs if you already have licenses. The requirements are so complex we hired a Microsoft partner for $15k just to audit our licensing and tell us what we could use.
Most companies end up paying both their existing Windows license costs AND full Azure rates because nobody can figure out the hybrid benefit eligibility.
Azure Support Costs Are Brutal:
Professional Direct support costs $1,000/month flat rate. That's fine if you're spending $20k/month but brutal if you're a small company spending $2k/month.
We ended up on Premier support at $25,000/month because Professional Direct couldn't help with our SQL Server licensing issues. Spent more on support than compute that year.
Google Cloud: Actually Reasonable (But Still Expensive)
GCP is the least fucked of the three platforms. Their pricing makes sense, they bill per-second instead of per-hour, and they don't have as many hidden fees. But it's still expensive as hell.
What Google Cloud Actually Costs:
Compute Engine pricing includes sustained use discounts up to 30% automatically. No bullshit reserved instance math - if you run a VM consistently, you pay less. Simple.
Cloud SQL charges per vCPU-hour and you only pay for high availability if you actually want it. Unlike AWS RDS where Multi-AZ is basically mandatory, GCP lets you run single-instance databases for dev/test.
Persistent Disk costs $0.04/GB/month for standard disks. Competitive with AWS gp3. The performance is actually better than AWS standard volumes.
GCP Has Less Surprise Billing:
Network pricing is straightforward - $0.01/GB between zones, $0.12/GB egress to internet. No complex tiers or surprise fees.
Their pricing calculator actually warns you about egress costs and includes realistic usage scenarios.
The Catch:
Google adjusts prices quarterly. Your $2,000/month bill might be $2,200 next quarter. It's transparent but makes budgeting a pain in the ass. Still better than AWS's "surprise you with random charges" approach.
The Shit That Adds Up Over Time
Storage Costs Never Stop Growing:
Your logs, backups, and user data accumulate every month. AWS S3 seems cheap at $0.023/GB until you realize you're storing 2TB of logs nobody ever looks at.
We started with maybe 100GB in S3. Two years later we had something like 3TB costing us around $70/month. Plus those fucking S3 request charges - $0.0004 per 1000 PUT requests sounds like nothing until your app uploads user content constantly and you're burning through millions of requests.
Google Cloud Storage is slightly cheaper at $0.020/GB but Class A operations cost $0.05 per 1000. Your backup scripts hitting the API constantly will surprise you.
Monitoring Costs Are Brutal:
AWS CloudWatch charges $0.30/metric/month. Sounds cheap until you have 200 custom metrics costing $60/month just for monitoring.
Azure Monitor charges $2.30/GB for log ingestion. We generated 300GB/month in logs and paid $690/month just to store them. More than our actual compute costs.
Compliance Costs Are Mandatory:
If you need SOC 2 or GDPR compliance, you can't avoid these costs. AWS Config costs $0.003 per configuration item. For 1000 resources, that's $3000/month in compliance monitoring.
Multi-Cloud is Expensive as Hell
Most companies use multiple cloud providers because they're terrified of vendor lock-in. It increases your costs by at least 50% but executives sleep better at night.
You need separate support contracts, separate monitoring tools, and staff who understand all the platforms. Data transfer between clouds costs $0.09-0.12/GB each way.
We run workloads on AWS and GCP. The operational overhead of managing both platforms costs us an extra $2000/month in tooling and staff time. Worth it for the flexibility, but expensive.
Bottom Line:
Google Cloud costs the least, AWS has the most shit available, Azure makes sense if you're already paying Microsoft taxes. All of them cost way more than their calculators suggest. Plan for 2x whatever the calculator says and you might not get totally fucked.