Notes from the team building C3X. Cost estimation for Terraform, the economics of cloud infrastructure, and how to ship FinOps tooling without a SaaS gate.
Fargate costs 15-50% more per vCPU and GB than EC2, but you pay for zero idle capacity. Here's where the crossover actually is, why dense fleets favor EC2, and when Fargate's premium is the cheapest line on the bill.
ALB and NLB both cost ~$16.43/month to exist plus LCU/NLCU usage at ~$0.006-0.008 per unit-hour. The price difference is small, the real lever is the LCU dimension your traffic stresses and how many balancers you run.
The AKS control plane is free (or ~$73/month for the SLA tier), but worker node VMs are the real bill. Here's how to cut AKS cost with the autoscaler, Spot node pools, right-sizing, reservations, and scale-to-zero.
Cloud Run bills per request-second and scales to zero; GKE bills for nodes 24/7. The cheaper option comes down to utilization. Here's the crossover, the min-instances trap, and where Autopilot fits.
MemoryDB costs more than ElastiCache for the same node type and adds a per-GB data-written charge ElastiCache doesn't have. Here's the price difference, the durability you're buying, and when each is the right call.
RDS Proxy bills per vCPU of the database it fronts, continuously, whether or not connections flow. Here's how the charge works, when it pays for itself, and when it's overhead you can skip.
Amazon MWAA bills a continuous base environment fee with no scale-to-zero, so even an idle environment costs ~$350/month. Here's the cost breakdown, why the floor exists, and how to keep it down.
SQL Managed Instance has a 4-vCore minimum, so even an idle instance starts around $1,460/month on General Purpose. Here's the full pricing breakdown, the Business Critical premium, and the four levers that actually cut the bill.
Aurora I/O Optimized eliminates per-I/O charges but adds ~30% to instance/storage rates. The break-even is around 25% I/O share of total cost. Here's the math, when to switch, and how to apply it in Terraform.
The standard 5 tags every resource should have, the activation gotcha that hides historical data, enforcement via Terraform default_tags and Tag Policies, and the anti-patterns to avoid.
AHUB lets organizations with Windows Server or SQL Server licenses (with Software Assurance) save up to 40% on Windows VMs and 55% on SQL workloads. Stacks with Reserved Instances for 70-80% total savings.
SUDs are automatic (30% off steady workloads). CUDs require commitment (25-55% off). They don't stack. Here's the math, the break-even, and the optimal strategy: 3-year flexible CUDs on baseline, SUDs above, Spot for elastic.
Lambda wins for sporadic and event-driven workloads. Containers win for steady high-volume traffic. The crossover is around 2-3M requests/day for typical APIs. Here's the math, the workload archetypes, and a decision framework.
For 95% of production workloads, gp3 is the right choice (cheaper, performant enough). Use io2 only for over 16K IOPS sustained, 99.999% durability, or Multi-Attach. Here's the pricing math and the migration playbook.
How we serve a public AWS/Azure/GCP pricing API with 2.7M SKUs from a single small Hetzner VPS. Schema design, scraper architecture, and six months of operational lessons.
Hourly rates, performance per dollar, and migration paths for AWS general-purpose EC2 families. When to migrate to m7i, when to skip straight to Graviton m7g, and when older families still make sense.
Lambda is the most mis-estimated AWS service. Here's why cost tools show $0 by default, what data to feed them for accurate numbers, and the four optimization levers that actually move the bill.
Autopilot bills per-pod (vCPU, memory, ephemeral storage). Standard bills per-node. The break-even is around 50-55% node utilization. Here's the actual pricing math with n2-standard-4 nodes, when each mode wins, and how to migrate without surprises.
PUT is $0.005/1K. GET is $0.0004/1K. For high-frequency small-object workloads, request fees can exceed storage fees by 100x. Here's the full request pricing matrix across all storage classes, when fees dominate, and the architectural patterns that keep them low.
Right-sizing cuts steady-state EC2 spend 20-40% in one pass. The 20/50 rule (average under 20%, peak under 50%) finds candidates. Here's the full pipeline: gather metrics, decide size, validate against periodic peaks, deploy in waves with monitoring.
B-series is burstable. D-series is general. E-series is memory-optimized. F-series is compute-optimized. Family selection drives the bill more than size. Here's the per-vCPU and per-GB pricing math, the decision flow, and the workloads each family wins on.
Ingestion at $0.50/GB dominates CloudWatch Logs bills. Source-side filtering, structured logging, IA tier, and S3 migration for high-volume groups typically cut the bill 50-80%. Here's the full breakdown and migration playbook.
Estimate Azure costs from Terraform configurations including Reserved Instances, Azure Hybrid Benefit, Spot VMs, and Premium SSD tier mapping. Completes the cloud trilogy alongside AWS and GCP.
Three categories of AWS Free Tier (12-months-free, always-free, trial offers) with the actual limits for each service. What stays free forever, what expires, what surprises new accounts.