Follow Value, Not Costs
The one thing you need to check in your cloud applications
Victor Garcia
August 10, 2025
Together with PointFive
We brought together 70+ practitioners from across the FinOps community to build the largest knowledge base of cloud inefficiencies
Explore 200+ peer-validated inefficiencies across AWS, Azure, GCP, OCI, Snowflake, and Databricks.
Each inefficiency includes a detection checklist, relevant billing model, and clear remediation steps. Filter instantly by account, service, inefficiency type, or cost driver and start saving fast.
Join your peers and contribute inefficiencies you've uncovered to keep growing the knowledge base.
FINOPS
Follow Value, Not Costs: A FinOps Transformation Story
Erik Norman tells a story that will make you question everything about how we think about cutting costs.
His team spent months trying to squeeze every penny from a retail software backend that was already running on bare minimum resources. Nothing worked because they were already at rock bottom.
Then they achieved a 98.4% cost reduction. Not by cutting more resources, but by asking a simple question: what value does this system actually provide?
The answer was shocking. The system was handling over 10 million API requests per month because the frontend team had built it to constantly ask "any updates yet?" every single second.
When stores closed at the same time each night, they all tried to upload data at once, creating massive traffic spikes.
The fix was simple. Instead of constantly asking for updates, they switched to a notification system that only sent messages when something actually changed.
The stampede of simultaneous uploads got spread out over time. Server usage dropped from maxed out to mostly idle.
FINOPS FOR AI
Understanding Hidden Costs in GenAI Token Pricing
When companies look at GenAI pricing, they often focus on the simple cost per million tokens. But this is like judging a car's total cost by only looking at gas prices while ignoring the engine type, how you drive, and maintenance needs.
The real costs hide behind those clean price sheets that vendors show you. Token pricing works more like a complex puzzle than a simple math problem.
Not All Tokens Cost the Same
First, input tokens and output tokens have very different prices. When you send data to an AI model, that's cheap. But when the model creates a response, that costs three to five times more.
The Hidden Cost That Kills Budgets
The biggest surprise cost comes from Context Window Creep. Most AI systems have no memory between conversations. To keep a chat going, you must resend the entire conversation history with every new message.
A simple customer service chat might start with 5 tokens for "What's your return policy?" The bot responds with 50 tokens. When the customer asks a follow-up question with 6 tokens, the system must send all 61 tokens from the conversation so far. As the chat continues, this number grows fast and costs explode.
Ways to Cut Costs
Smart companies use several tricks to lower their AI bills. Caching stores common responses so the system doesn't need to process the same requests over and over.
For tasks that don't need instant answers, batch processing offers huge discounts - often 50% or more. But you have to wait up to 24 hours for results.
Why Cheap Models Cost More
The cheapest AI model often ends up being the most expensive choice. The total cost for one successful result can be much lower than the accumulated cost of fighting with a cheaper model.
Where You Buy Matters
The platform you choose changes everything about costs. Cloud providers like Amazon, Microsoft, and Google wrap AI models in enterprise features but charge more. Different vendors can have 30% price differences for the same model.
FINOPS EVENTS
Mastering AI Economics
AI workloads are powerful but they’re also expensive. In this webinar, we’ll break down practical strategies to manage, develop, and reduce unit costs across your AI stack.
Learn how top engineering and FinOps teams are aligning performance with budget by optimizing architecture, tracking true cost per model, and using practical insights to stay ahead of runaway spend.
What we’ll cover:
Understanding unit costs: What they are, why they matter, and how to track them
Cost-efficient architecture: Design patterns and trade-offs that lower compute and storage bills
Data & model strategy: How to optimize what you train, when, and where
FinOps for AI: Building transparency and accountability into fast-moving AI teams
Speakers
Vaibhav Sharma
David Gross
Alon Savo
Host: Victor Garcia
September 9th - 6:00 PM CEST / 10AM EST
FINOPS LIFECYCLE
You Build It = You Budget It?
Most companies follow a "you build it, you run it" rule for their software. But this ownership rarely includes costs.
When did your phone last ring at 3 AM because your app was spending too much money? Probably never.
We get alerts when systems go down, but not when they burn through cash. Here's why costs often take a back seat to keeping systems running:
Visibility matters more. When your app crashes, customers notice right away. When costs spike, it might take weeks to show up on reports.
Costs are harder to understand. A system being "up" or "down" is simple. Costs change based on many factors and can be tricky to predict.
Startups focus on growth first. When you're trying to find customers, spending money to keep things running feels less important than making sure nothing breaks.
Some cost fixes are easy and don't need much teamwork. Others, like changing how the whole system works, need everyone involved. The biggest wins come from catching cost problems early, before they become expensive.
Smart companies are starting to think about costs during the design phase. They use tools that show cost estimates right in the code editor. Some even use AI to help write cheaper code from the start.
CLOUD PROVIDERS
Google First in FinOps Race !?
AWS
AWS delivered several cost-saving updates this week. You can now force-terminate stuck EC2 instances that remain in "stopping" state - preventing ongoing charges for attached resources like EBS volumes.
S3 improvements include tag-based access control for Access Points, enabling better security and cost management through attribute-based access control (ABAC). The updated Mountpoint for S3 CSI driver accelerates containerized application data access, reducing compute time and costs for data-intensive workloads.
Operational efficiency gains include organization-wide VPC Flow Logs enablement from a central account, SQS support for 1 MiB message payloads (reducing API calls), and AWS Resource Explorer now covering 120+ resource types for better resource discovery.
Performance boosts came via Aurora Serverless v2 delivering 30% better performance for transaction processing. AWS Budgets now offers improved cross-account cost visibility for billing-permitted users.
Microsoft Azure
Microsoft focused on FinOps stability with non-breaking updates in FinOps Hubs 1.2, ensuring data consistency during platform evolution. The FinOps Toolkit v1.2 brings enhanced cost analysis capabilities.
AKS cost analysis now provides detailed Kubernetes spending breakdowns by cluster and namespace.
Google Cloud
IDC recognized Google Cloud as a Leader in their Cloud FinOps MarketScape, highlighting their commitment to comprehensive cost management tools.
AWS
Vercel Cuts AWS Lambda Costs by Reusing Idle Time
Vercel found a clever way to cut AWS Lambda costs by up to 95 percent by fixing a big problem with how Amazon charges for serverless functions.
Here's the issue: AWS Lambda bills you for the entire time your function runs, even when it's just sitting there waiting for something else to finish. Think of it like paying for a taxi that's stuck in traffic - you're still paying even though you're not moving.
They built a tunnel system to connect their platform with AWS Lambda functions. Then someone had a bright idea: what if they could send multiple requests through the same tunnel to reuse the same Lambda instance?
The fix wasn't simple to build. They had to track how much CPU and memory each Lambda instance was using, monitor the 15-minute time limit, and add smart controls to reject requests when needed. But it worked.
One Vercel customer saw their monthly bill jump from $300 to $3,550 because their functions were making slow AI calls. After turning on Vercel's new "Fluid Compute" feature, they reported seeing immediate cost savings.
📺️ VIDEO
Expert Masterclass: FOCUS Success Case in Azure
In this Masterclass, we teach you how to design an event-driven architecture based on Azure Functions that processes billing information and divides costs using a combination of general rules and specific allocations that can be managed and pre-set in advance. Using FOCUS and Power BI
🎖️ MENTION OF HONOUR
Azure FinOps Book by Anderson Oliveira
Congrats to Anderson Oliveira for releasing his book, looking forward to its arrival so I can take a look. Here’s the information about it:
It starts by explaining what FinOps is and how Azure pricing works. Then it shows readers how to collect data about their cloud usage, create good labeling systems for their resources, and split costs between different departments.
The book teaches people to use Azure's built-in tools like Cost Management and Azure Advisor. These tools help find waste and fix spending problems quickly. It also covers writing reports that help managers understand where money goes.
Later chapters focus on finding ways to save money, setting up automatic controls, and using artificial intelligence to make better decisions. The book includes real examples and code that readers can use right away.
This practical guide helps IT professionals, finance teams, and managers work together to control cloud costs while still getting the computing power they need.
Professional Spotlight
Marco de Bianchi
AWS FinOps Master
Had a brief chat with Marco in FinOps X Barcelona, pleasure to be able to interchange ideas with people like him. Really great job he’s doing on AWS, saving customers good money!
That’s all for this week. See you next Sunday!
Don’t Miss Our Best Online Event of 2025
Join FinOps professionals at the FinOps Weekly Summit 2025 and discover how to:
Transform from reactive fire-fighting to strategic leadership — Learn the proven frameworks that top practitioners use to turn cloud cost management into a competitive advantage
Master AI-powered cost optimization — Get exclusive access to the latest automation tools and techniques that can reduce your cloud spend by up to 40% while accelerating innovation
Build bulletproof unit economics — Walk away with actionable playbooks for calculating true cloud ROI and proving business value to executives who control your budget
"But I don't have time for another webinar..."
This isn't another generic webinar. This is the only major FinOps event left in 2025, featuring battle-tested strategies from companies managing billions in cloud spend.
October 23rd, 2025 | 4:00 PM - 8:00 PM CEST
Limited seats available