Presented by

Want to appear here? Talk with us

Model Rightsizer is an open-source Claude Code sub-agent from CloudZero. It scores every task on capability need versus cost pressure, then routes it to the smallest model that clears the bar. Nothing defaults to the priciest tier because nobody asked.

Week one on CloudZero's own agents:

Quora just shared a real-world case study on moving core services from x86 to AWS Graviton chips.

About 40 percent lower costs on their feed backend and a 12 percent faster homepage. They did this with zero downtime, which is the hard part most companies worry about.

The team found that list price savings alone only pointed to 22 percent cost cuts. Real savings came out higher because fewer servers were needed to handle the same traffic.

Rebuilding old code libraries for the new chip type was the most time consuming part, more than the migration itself. The team stresses that keeping systems updated regularly is cheaper than a big rushed fix later.

They also built safety nets like shadow testing and slow rollout stages before shifting real customer traffic. This let them catch bugs and measurement errors early, including one hidden bug in their own performance tracking tools

Read More

A live conversation with me on the cloud waste your cost tooling can't see. I break down audience-submitted waste live in the closing segment.

📅 September 23rd, 2026
🕚 5:00 PM Spain / 11:00 PM ET

Amazon Bedrock lowers GPT-5.6 Sol pricing by up to 33%, cutting costs for high-volume AI workloads through November 2026.

Cost Anomaly Detection now watches third-party Bedrock models like Claude, flagging surprise AI spend by account and region.

Read All AWS Updates

Gemini Enterprise adds pay-as-you-go billing, so teams can run AI agents without a subscription or upfront cost.

New Flexible Savings Plans cut Gemini token costs by 10% to 20% based on a one- or three-year spend commitment.

Read All GCP Updates

Azure VM vCore Customization is now generally available, letting teams turn off unused CPU threads to avoid paying for capacity they don't need.

Microsoft shared a new AI FinOps guide for Foundry, covering cost attribution, caching, and budget controls for production AI agents.

Read All Azure Updates

Join FinOps professionals at the FinOps Weekly Summit 2026 and discover how to:

Transform FinOps from reactive cost control into strategic business value

Build scalable unit economics for the AI era

Learn proven strategies from organizations managing billions in cloud spend

📅 October 20 & 21, 2026 - 9AM ET / 3PM CEST

We’ll deploy Kestra locally, configure a FinOps workflow using a pre-built blueprint, monitor Azure Batch resources and automatically scale idle resources down to reduce unnecessary cloud costs.

GitHub Copilot just changed how it charges for AI use. It now runs on usage-based billing, which means costs can go up or down each month.

Level one: turn on GitHub's own budget tools first. These let you set spending caps per user, per team, or across the whole company. They work everywhere, including the GitHub website.

Level two: for teams that need tighter control, route Copilot through an Azure gateway called API Management, connected to Azure AI Foundry. This gives real-time limits per developer, based on actual token use, not just monthly dollar caps. It also means the spending shows up on your own Azure bill, with detailed tracking in Azure Monitor.

This upgrade only works inside code editors and command line tools, not on the GitHub website itself.

Free credit boosts from GitHub end on September 1. After that, overage costs kick in for real.

Read More

Cloud bills rarely spike because of what your team is actually using. They spike because of what keeps running when nobody is watching. A recent audit of a staging environment costing about $4,000 a month found the same problems again and again.

Idle compute: Servers and containers running all night and on weekends, even when almost nobody uses staging outside business hours. A simple on and off schedule cut compute costs by 60 to 70 percent.

NAT Gateway fees: These charge for both being turned on and for every byte of traffic that passes through. Free S3 and DynamoDB endpoints can remove a lot of that traffic instantly.

Logs that never expire: Without a retention policy, log storage grows forever and quietly adds up to hundreds of dollars. Setting short retention windows, then archiving older logs to cheap S3 or Glacier storage, keeps history available without the ongoing cost.

Most waste hides in idle time, unnecessary network paths, and forgotten storage settings.

Read More

Teams often ask for more CPU and memory than they need, unused storage, load balancers, and old namespaces keep racking up charges, poor scheduling wastes node capacity, mixed pricing models make budgeting hard.

The fix has three parts: First, get clear visibility into what each team and app actually spends. Second, automate optimization using tools like Kubecost, VPA, and HPA. Third, build a culture where finance and engineering work together on cost decisions.

It shows how to install Kubecost for cost tracking. It shows how to set up autoscalers so pods and nodes match real demand. It even points to AI tools that can suggest smarter resource limits over time.

The payoff is real. Companies following this approach can cut cloud bills by 30 to 40 percent. That frees up money for new projects instead of wasted infrastructure.

Read More

Visit FinOps Weekly Hub

Find all Jobs in our FinOps Jobs directory or join FinOps Weekly Jobs Newsletter below to have a weekly email with the most relevant updates every Monday

Use code: FINOPSWEEKLY_20 to get an instant 20% discount on the most prestigious certification bundles:

FinOps Certified Practitioner

FinOps Certified Engineer

FinOps Certified FOCUS Analyst

FinOps for AI

Get Certified for Less

Join the Slack community where the real FinOps conversations happen