#finops

[cloudwithaman]

The Redshift discount everyone knows, and the fine print nobody mentions

Everyone knows Redshift on-demand pricing is brutal, and everyone knows reserved nodes fix that. What rarely gets said out loud: Redshift reservations are some of the least forgiving commitments AWS sells. No size flexibility, no convertible option, no marketplace to offload one you don't need. Before you sign up for three years, that's worth sitting with.

Start with the node type, not the discount

Redshift has three node families in play right now. RG is Graviton-powered and the one AWS now recommends for new builds up to 2.4x faster than RA3 with 30% lower price per vCPU, plus a built-in engine for querying Iceberg tables in S3 without Redshift Spectrum's per-terabyte scan fees.

RA3 is still fully supported for reservations. Compute and storage are billed separately here, so your bill has two moving parts even after you reserve.

DC2 is legacy at this point local SSD storage, recommended only under 1TB compressed, and reserved node support is thinner than the other two.

If you're on RA3 and weighing a 3-year commitment, run the RG migration math first. Locking into RA3 pricing before checking whether RG is the better long-term fit is an easy way to overpay.

The part that makes Redshift different

Compare this to RDS, where a reserved instance flexes across sizes within a family. Redshift has none of that:

No size flexibility: a reservation for ra3.4xlarge covers only ra3.4xlarge. Nothing else.

No convertible class: once you commit to a node type, you're locked to it for the term.

No resale market: Redshift reservations are explicitly excluded from the AWS Reserved Instance Marketplace. A stranded reservation just sits there costing money.

Stack those three together and a wrong purchase isn't a rounding error, it's tens of thousands of dollars committed to a configuration you no longer use, with no way out before the term ends.

What actually protects you here

The payment math (No Upfront, Partial Upfront, All Upfront) matters less than getting the node type and size right in the first place. A few things worth doing before you buy anything:

Watch CPU, disk I/O, and query queue depth for at least 30-60 days, ideally including a peak period.

If this is your first Redshift reservation, start with 1-year No Upfront. The gap between 1-year and 3-year savings is real but modest, a few points while the cost of the wrong 3-year bet is not.

Separately model Concurrency Scaling. Reserved nodes don't cover it, and it bills at on-demand rates once you exceed the daily free credit.

If you're managing reservations across RDS as well as Redshift, the sizing and payment-option logic carries over closely; this RDS Reserved Instances breakdown walks through the engine-by-engine version of the same decision. And if spend-based commitments feel like a better fit than locking into a specific node type, AWS Database Savings Plans are worth understanding as the more flexible alternative for parts of your database footprint.

Where the buyback guarantee changes the calculus

Given there's no marketplace and no convertible option, the standard advice is "buy conservatively and eat the risk." Usage.ai's Insured Flex Reserved Nodes take a different approach: if a reservation gets stranded, cluster resized, migrated, or shut down the unused commitment gets bought back as real cashback. That removes the main reason teams under-commit on Redshift in the first place.

If you want to see what your own cluster's reservation math looks like before committing to anything, the savings calculator is a fast way to check.

The full breakdown node families, payment option math, and the complete purchase sequence is in the original guide.

Anyone else sitting on an RA3 cluster and putting off the RG migration decision because the reservation math feels too fiddly to untangle?

[kellton]

The Rise of FinOps: Balancing Performance & Cloud Spend

Discover the Rise of FinOps & how it transforms cloud management. What is FinOps? It’s a cultural practice uniting engineering & finance to optimize cloud spend.

[xexiar]

Groceries. Sprint 1 of shopping.

Groceries. Sprint 1 of shopping.

[ekremdum-blog]

Cloud cost optimization for business — where cloud spending leaks, the right-sizing, waste-elimination, and pricing practices that recover it, and how to build ongoing FinOps discipline.

Cloud Cost Optimization: Cutting Waste Without Cutting Performance

[marketingactivo]

FinOps: cómo recortar el desperdicio en la nube y alinear gasto digital con valor

La nube se presentó como la gran aliada de la eficiencia: infraestructura flexible, escalabilidad bajo demanda y pago por uso. Sin embargo, en muchas organizaciones esa promesa convive con una realidad menos visible: costes crecientes, recursos infrautilizados y gasto difícil de atribuir a un área, producto o campaña. En ese contexto, FinOps se ha convertido en una respuesta cada vez más…

FinOps: cómo recortar el desperdicio en la nube y alinear gasto digital con valor

[cloudwithaman]

The Azure bill nobody can explain

Microsoft just posted $75 billion in Azure revenue. Meanwhile your finance team wants to know why your bill keeps climbing 20% a quarter with zero user growth to show for it.

It's not you. Azure runs on thousands of SKUs, each with its own regional pricing, tiers, and commitment rules. Microsoft's docs explain what things cost. Almost nothing explains why you're overpaying, or how to stop.

Why Azure is its own special kind of hard

Azure shares AWS's complexity, then adds its own layer on top. Reserved VM Instances lock you into a specific VM series and region move from East US to West US, or swap series, and that reservation stops paying off.

Savings Plans trade some of that discount for flexibility: commit to a fixed hourly spend and it floats across regions and VM families. Reservations save more (up to 72%) if your setup never changes. Savings Plans save a bit less (up to 65%) but survive migrations. Most teams, unsure which to pick, end up choosing neither and just eat pay-as-you-go pricing.

Then there's Hybrid Benefit. If you're sitting on Windows Server or SQL Server licenses with Software Assurance, stacking that benefit with Reserved Instances can get you to 80% off. Figuring out which licenses qualify and staying compliant, though, is basically a part-time job.

Where the money actually leaks

Five services eat most Azure budgets:

Virtual Machines (40–60% of spend): Reserved Instances cut costs up to 72% on 3-year terms, but most workloads don't stay stable that long. Spot VMs save up to 90% for fault-tolerant jobs, with 30-second eviction notice.

Azure SQL Database (15–25%): Reserved capacity saves up to 33%, or up to 80% stacked with Hybrid Benefit. Reserve based on a growth guess and you're either paying for unused capacity or scrambling for more.

Storage Accounts (10–15%): Hot, cool, cold, and archive tiers each carry different access costs and minimum retention windows. Move data between tiers carelessly and transaction fees can wipe out the storage savings.

AKS (10–20%): AKS itself is free; the VMs underneath aren't. Clusters routinely run full-size VMs 24/7 for workloads that only spike during business hours.

App Services and Functions (5–15%): A Premium v3 P1 plan runs $400/month whether it serves 10 requests or 10 million. Running one plan per app multiplies that waste fast.

The commitment paradox

Here's the trap: the deepest Azure discounts require 1- or 3-year commitments, but nobody can predict their infrastructure three years out. So teams either overcommit and get stuck, or stay on pay-as-you-go and quietly overpay every month.

This is the exact problem Usage.ai's Insured Commitments were built around, they purchase optimized reservations and Savings Plans on your behalf, and if usage patterns shift, buy back what's unused in cash. You get the commitment-level discount without carrying the commitment-level risk.

Mistakes worth avoiding

A few patterns show up constantly:

Moving 10TB from Hot to Cool tier and getting surprised by transaction fees that eat the savings

Over-reserving SQL Database capacity, which can't be exchanged or refunded, 50–60% coverage is the safer starting point

Running ten apps on ten separate App Service plans instead of consolidating onto two or three

Missing egress charges ($0.087/GB after the first 5GB) on a misconfigured backup job

Buying reservations without checking Hybrid Benefit eligibility first, and leaving free savings on the table

The connection itself takes about five minutes: create an account, confirm a few Azure permissions, and paste in a couple of IDs from the Azure Portal. From there, the scope and configuration of your commitments matters as much as whether you buy them at all, a poorly scoped Savings Plan can sit underutilized in one subscription while the rest of your org pays full price.

Companies going through this analysis typically find 30–50% in monthly savings they didn't know were sitting there. On a $150K/month Azure bill, that's $45K–$60K a month, found in minutes rather than months of spreadsheet work.

You can read the full breakdown, including the exact setup steps, here.

What's the last surprise line item that showed up on your Azure bill?

[cloudwithaman]

 Multi-Cloud Cost Optimization Guide | AWS, Azure, GCP Savings

Ask anyone running infrastructure across AWS, Azure, and GCP what their least favorite part of the job is, and "reconciling three different cloud bills" comes up fast. Each provider prices things its own way, and that mismatch quietly turns cost management into a much bigger job than it looks like on paper.

Why Three Clouds Never Add Up Cleanly

AWS leans on granular metering plus commitment discounts like Savings Plans and Reserved Instances. Azure sticks to Reserved VM Instances, which are more rigid. GCP uses Committed Use Discounts with a spend-based twist. Two workloads that look identical on paper can behave completely differently once you factor in how each cloud actually bills for them.

That mismatch shows up everywhere: billing files that don't match, tags and labels that don't align, and teams working off different versions of "what we're spending." Add in ownership split across engineering, finance, and ops, and you get a lot of people looking at the same spend and reaching different conclusions.

The Real Reason Multi-Cloud Teams Overspend

It's not a lack of understanding. Most teams know commitments can save 20–70% on compute and databases. The real blocker is risk perception, nobody wants to be the person who over-committed and got stuck paying for capacity nobody used. So teams default to on-demand pricing far more than they need to, and that caution compounds across three clouds instead of one.

Data latency makes it worse. Native cloud recommendations often refresh every few days, sometimes longer. By the time a signal shows up, the window to act on it may already be gone. If you're weighing term length on the AWS side, this breakdown of 1-year vs. 3-year commitments is a useful starting point before you commit to anything long-term.

What Actually Works

A few practices show up consistently in teams that get multi-cloud spend under control:

Normalize billing data into one taxonomy, so EC2, Azure VMs, and GCE all roll up as "compute" instead of three separate stories

Rightsize the instances and VM sizes that make up the bulk of spend usually 20–30 resource types drive 70–80% of the bill

Set commitment coverage targets per cloud based on actual workload stability, not a blanket percentage. Azure's cost structure in particular rewards a more cautious approach than AWS does

Put guardrails in place; required tags, budget alerts, anomaly detection so drift gets caught early instead of at the end of the month

None of this is exotic. It's just consistency applied across three systems that were never built to talk to each other.

Where Automation Actually Helps

This is the part that's genuinely hard to do by hand: forecasting workload behavior across three clouds accurately enough to commit with confidence, every time. Usage.ai's approach is to refresh usage analysis daily instead of relying on native recommendations that can be a week old, then handle the actual purchasing once a team approves it.

The two features worth knowing about if you're evaluating this space: Flex Commitments, which give SP/RI-style discounts without the long lock-in, and cashback protection, which pays out if a commitment ends up underutilized instead of just eating the loss. If you want a quick gut-check on where your own AWS spend stands, the free savings calculator is a fast way to see it.

Multi-cloud isn't going away, and neither is the pricing chaos that comes with it. Normalizing your data and building a cloud-specific commitment strategy won't make the three clouds speak the same language, but it'll stop the mismatch from costing you money every month. You can read the full breakdown in the original guide.

Anyone else find that the under-commitment problem gets worse, not better, the more clouds you add?

[cloudwithaman]

GCP Cost Optimization Best Practices & Why They Don't Scale

Every GCP bill review starts the same way: rightsize the VMs, kill the idle instances, clean up old storage, set some budget alerts. It's the checklist everyone recommends. It works for a while. Then usage scales, and the bill keeps climbing anyway. You're not doing anything wrong. You've just hit the ceiling of what that checklist was ever built to fix.

Why the Checklist Stops Working

Rightsizing and autoscaling are the first things teams reach for low risk, quick wins. But they only remove excess capacity. Once your environment matures and workloads are already reasonably sized, there's nothing left to trim. These techniques change how much you consume, not how much each unit costs. That gap is easy to miss until spend keeps rising in a clean environment. (This is also the core problem across clouds, not just GCP see this multi-cloud breakdown of how AWS, Azure, and GCP all hit the same wall differently.)

Idle resource cleanup has the same shape. Shutting down forgotten dev environments and orphaned disks feels productive, and it is once. But in most production-heavy GCP setups, the bulk of the bill comes from things that have to run continuously. You can't clean your way out of always-on spend. (The AWS side of this how to actually find idle and underutilized resources follows the same detection logic if you're multi-cloud.)

Storage tiering helps too, but it's usually a smaller, more predictable slice of the bill than compute, and it plateaus fast.

The Part Nobody Wants to Touch: Discounts

This is where most teams stall. Google offers Sustained Use Discounts (automatic, no risk, capped) and Committed Use Discounts (deeper savings, but you're locked into a fixed usage level for 1–3 years).

CUDs are where the real money is and the real risk. Fall short of your committed usage and you're still paying for it. So teams do the rational thing: they under-commit, leaving savings on the table specifically to avoid being wrong.

That's not a technical problem. It's a forecasting and risk-tolerance problem, and the standard best-practices list has basically nothing to say about it.

So What Actually Scales?

Once the easy wins are gone, optimization stops being about configuration and starts being about how pricing decisions get made and revisited. A few things separate teams that keep saving from teams that plateau:

They re-evaluate commitment coverage constantly, not once a quarter

They separate who approves the commitment from who eats the risk if it's wrong otherwise everyone defaults to caution

They optimize for effective blended rate, not just total spend, since that number compounds in a way one-time cleanup never does

This tracks with a related look at why cloud cost management keeps failing even after teams have decent visibility and tooling, the gap isn't insight, it's turning insight into decisions that hold up as usage shifts.

It's part of why tools like Usage.ai exist. Its Flex-Commit Program automates commitment coverage across AWS, Azure, and GCP while offering cashback if a commitment goes underutilized, so teams aren't stuck choosing between "save more" and "sleep at night."

Best-practices checklists get you to baseline efficiency. Baseline efficiency isn't a cost strategy. Teams that keep saving as they scale are the ones treating pricing as an ongoing decision, not a box to check once. Full breakdown, including the risk-management mechanics, here.

Where's your team stuck — still finding waste to clean up, or already staring down the commitment-risk problem?