Presented by

Want to appear here? Talk with us

Here's the thing. Your FinOps team finds the savings. They put them in a dashboard. And then? Nothing happens.

It's not a tools problem. We asked practitioners directly, 56% ship less than half of what they find. And 77% say the same thing: it's a workflow and ownership mess. Nobody owns it. No ticket gets created. No PR gets opened. The money just sits there.

The real gap? It's between the people who find the savings and the people who can actually ship the fix. That's a human problem. A process problem. And most teams don't even know where exactly it breaks.

So we made a simple checklist. Six stages. Thirty checkpoints. You go through it, you see where your pipeline is broken and what to fix first.

A company spent $200,000 on AWS CloudTrail logs that nobody needed or looked at. CloudTrail costs went from $1,000 per month to $10,000 in one account and $30,000 across the organization. The root cause was a single configuration setting from years ago that logged every single S3 action across all buckets in all accounts.

The system was recording 8.98 billion S3 events per month in just one region.Every time any system read a file from S3, it created a log entry at $0.10 per 100,000 events. S3 read operations make up 85-95% of all S3 activity in most production environments.

Changing one configuration field from "All" to "WriteOnly" cut costs by roughly $300,000 per year. The second step was limiting logging to only buckets that contain sensitive data like personal information or credentials. This brought data event costs close to zero while keeping full audit coverage where it matters.

A security tool was added to the system and recommended full S3 logging for better visibility. The engineering team followed the recommendation without understanding the cost implications at scale. What cost $1,000 per month in 2024 became a $360,000 annual expense in 2025 as the infrastructure grew.

When security teams introduce new tools, finance and operations teams need to be part of the conversation from day one. The most expensive line item on your cloud bill might be a well-intentioned setting that someone enabled months ago and never looked at again.

Three questions should always get answered: What configuration changes does this tool recommend and what will they cost at our scale? What AWS permissions does it need? Who owns the monthly bill for this tool's impact?

Read More

Database Savings Plans now support Amazon OpenSearch Service and Neptune Analytics, offering up to ~35% savings. These plans cover serverless and provisioned modes across regions.

Amazon Bedrock introduced TimeToFirstToken and EstimatedTPMQuotaUsage CloudWatch metrics.

Graviton4 availability expanded to C8gd and M8gd instances. Delivering ~30% better performance over Graviton3, these instances offer superior price-performance for compatible workloads across Spot and Savings Plans.

Read All AWS Updates

PostgreSQL Elastic Clusters now support Terraform, Bicep, and Ansible (GA).

Read All Azure Updates

Conversational Analytics in BigQuery now labels agent-generated queries and improves partition support.

Cloud Storage Rapid Buckets (zonal) are now GA.

Read All GCP Updates

​Join us at the heart of Hawthorne as we kick off our first-ever LA FinOps Weekly Meetup at the lively Common Space Brewery!

Hosted by FinOps Weekly Regional Leader Diana Molski, this event is your chance to dive into the world of FinOps while enjoying good company and great brews.

​We promise an evening of engaging conversations and a relaxed vibe.

First beer on us, limited slots!

See you there!

📅​ Date: April 9, 2026
🕗​ Time: 5:00pm PT
📍 Common Space Brewery | Hawthorne, California​

Discover the future of IT cost management and the evolution of FinOps with the adoption of a Universal Spend Ledger with Sasha Kipervag (Ternary CEO)

Listen on Apple Podcast

Listen on Spotify

Azure just made big changes to how you can swap and refund your cloud reservations. If you buy Azure services in advance to save money, you need to know about these new rules.

You can swap reservations within the same product family. For compute services, you can exchange between Virtual Machines, Dedicated Host, VMware Solution, and Nutanix BareMetal. For SQL services, you can swap between Managed Instance, SQL Database, and Elastic Pool.You can change regions, switch from one year to three years, or adjust quantities.

Azure planned to stop allowing exchanges for compute reservations in January 2024.They extended that deadline indefinitely, but it will end someday.

You can refund reservations, but there is a $50,000 limit per year. This limit applies to your entire billing account, not per reservation. Right now there is no early cancellation fee, but Microsoft might add a 12 percent fee in the future.

When you exchange a reservation, the new one must be worth at least as much as what remains on your old one.

Some services are locked in with no exchanges or refunds allowed. These include Databricks, Synapse Analytics, Red Hat plans, SUSE Linux plans, and security tools like Defender and Sentinel. You also cannot swap between different product families. A Cosmos DB reservation cannot become a SQL Database reservation.

Microsoft calculates refunds based on remaining days in your reservation period. The money goes back differently depending on how you pay. Enterprise Agreement customers get credit added to their prepayment balance, valid for 90 days. Credit card customers get money back to their original card. Wire transfer customers see the refund applied to their next invoice.

Azure is pushing people toward savings plans instead of reservations for flexible workloads. Although they are hella complex. We got an edition with Shannon explaining them

Read More

A developer just shared how a single application bug ballooned their AWS bill by 15% in one month and what they did to fix it. A bug in the image processing code was creating duplicate files over and over again.

Storage went from 100TB to 600TB in just weeks. The S3 bucket held quadrillions of duplicate images because of versioning. Monthly AWS costs hit $60,000 and kept climbing. The problem was:

The fix required careful planning because the obvious solution would have made things worse. Deleting quadrillions of objects through normal API calls would trigger massive LIST operation charges. Instead, the team used their database to identify which images were actually needed.

They exported duplicate file names to a CSV file. Then they built a Python script that deleted 50,000 images per second without listing operations. S3 lifecycle policies handled the actual removal. The results came in three months.

Costs dropped from $60,000 to under $45,000 per month. That's a 40% reduction with no performance impact. The real lesson here is about visibility. Regular cost analysis caught the spike early. Database records provided the roadmap for cleanup. Custom tooling avoided expensive API operations.

Small bugs in production code can create massive cloud bills when they interact with storage and data transfer at scale.

Read More

Microsoft just fixed one of the most annoying problems in Azure Functions. And Alfonso is here to explain to us.

For years, if you wanted your Azure function to run inside a private network, you had to pay for a server that runs all day every day. Even if your function only ran twice a day.

That meant choosing between two bad options. You could use the Consumption plan and get true pay-per-use pricing, but your function had to sit on the public internet. Or you could use Elastic Premium and get private network access, but you'd pay for a pre-warmed instance running 24/7 whether you used it or not.

NOW: Flex Consumption gives you both things at once. Your function can run inside a private network AND scale to zero when nothing is happening.

Alfonso built a file automation system for a large company. Files land in Azure storage and get routed to either file shares or SharePoint based on which folder they arrive in.

Everything runs inside a private network. No public endpoints. No shared access keys. All authentication uses managed identities. The files arrive sporadically, maybe a few dozen times per day. Each function runs for less than 10 seconds.

With the old Elastic Premium plan, this would cost money every single hour of every day. With Flex Consumption, the cost is close to zero because Azure gives you 1 million free executions per month per subscription. For most low-frequency automation work, that free tier covers the entire workload.

If you run event-driven workloads that need private network access, Flex Consumption should be your default choice now. This applies to blob processing, queue-based jobs, scheduled tasks, or any internal automation that runs occasionally rather than continuously.

Read More

Manager of Product Support @CoreStack

As Manager - Product Specialist at CoreStack for over 2.5 years, I lead a team of support engineers to deliver exceptional customer and partner support.

Connect with the Professional of the Week

Naming conventions is essential to avoid a Cloud mess.

Tagging is one of the core skills to govern Cloud Resources.

Cloud Governance is a core skill for FinOps.

Learn both in our course Tagging & Naming conventions with our discount code only for FinOps Weekly Readers.

Use code “FW2026” for 30% off in our course

P.S. Azure FinOps Course already started. Thanks everyone who joined there.

Together with CloudThrottle. Check their Cloud Budget Simulator