Presented by

Want to appear here? Talk with us

Still reporting spend after the damage is done?

Flexera shows cloud spend after it happens.

FinOpsly is built for proactive AI cost governance so teams can control spend before it turns into waste & measure the ROI of their AI initiatives, including tracking customer profitability and margin.

See how FinOpsly helps you:

Estimate before launch: Model AI workloads, data pipelines, and infrastructure costs before they go live.

Govern more than cloud: Control spend across AWS, Azure, GCP, Kubernetes, Snowflake, Databricks, GPUs, models, and AI workloads.

Act with confidence: Turn savings opportunities into approved, policy-governed actions.

Eligible Flexera customers may qualify for up to 6 months free.

An engineering team saw their LLM bill climb fast after launching an AI assistant. They found the same waste that hits most companies.

Studies show many teams waste 40 to 60 percent of their AI spend on poor setup choices. The team found and fixed five main problems.

- They routed easy tasks to cheaper models and saved big money on routine work.

- They cached repeated prompt text, cutting costs by 90 percent on those parts.

- They added semantic caching so repeat questions worded differently still hit the cache.

- They trimmed extra text sent to the model, which also made answers better.

- They compressed wordy prompts without hurting accuracy.

Together, these five steps cut their token bill by about 80 percent.

The biggest lesson is simple. Track where your money goes, fix the cheapest problem first, and check quality before rolling out changes.

This approach turns AI cost control into a normal, manageable part of running a product instead of a surprise expense each month.

Read More

Claude Code now caps subagents and enforces spend limits, stopping runaway parallel tasks from blowing past your budget. Once you hit the limit, new subagents get blocked and running ones stop automatically.

A Bedrock billing bug is fixed, so mapped model IDs now bill at the correct rate. This means cleaner invoices and more accurate cost tracking for teams doing chargeback.

Claude Opus 5 is now the default model, with clear pricing at $10 per million input tokens and $50 per million output tokens, plus a 1M token context window. Upfront pricing makes it easier to plan and forecast costs.

Gemini 3.6 Flash is now live globally, while Gemini 3.5 Flash is being phased out of that region. Teams using 3.5 Flash should start planning their move now to avoid disruption.

Transparent thinking is now generally available in Gemini Enterprise, showing real-time reasoning and tool use. It also speeds up time to first token, helping teams debug costly reasoning chains and improve response speed.

GPT-5.6 Luna is now 80% less expensive: Luna now costs $0.20 per million input tokens and $1.20 per million output tokens.

GPT-5.6 Terra is now 20% less expensive: now costs $2 per million input tokens and $12 per million output tokens.

Fast mode for GPT 5.6 Sol: Up to 2.5x faster responses. introducing Fast mode in the API, which replaces Priority Processing offering. For For new requests, pass service_tier="fast".

Read All AI Provider Updates

Big AI context windows can feel like an easy fix for search, but this article shows why they can turn into a cloud cost problem fast.

The author warns that sending huge blocks of raw company text into a top AI model may work in a pilot, but can slow down apps and raise spend at scale.

For FinOps teams, the key lesson: filter first, spend later.

A smarter setup uses a low-cost first step to check user access, search the right files, and cut the results from 50 chunks to about 5.

Only then should the system send the best text to the costly AI model. The math makes the point clear.

At 500,000 queries a month, sending 40,000 input tokens per query could cost about $63,750 a month before other costs.

Cutting that input by 90% can make spend far easier to plan.

This is a strong read for teams moving GenAI from demo to production because it links system choices to budget risk.

Do not pay premium AI rates to read text your model may not need.

Read More

Join this FinOps for AI webinar to learn how to control AI costs, improve visibility, implement governance, and justify AI investments with confidence.

📅 August 20, 2026
🕚 6:00 PM Spain / 12:00 PM ET

Discover the FinOps Value Map, a powerful tool that connects cloud spend to real business outcomes, revealing hidden waste and value across your tech stack.

AI agents can look perfectly healthy while quietly draining your budget. That is the problem one developer ran into while building automated AI agents.

Requests were succeeding, models were responding, and all the normal monitoring tools showed a green light.

But behind the scenes, the agent was stuck in a loop, burning money with every retry.

Most monitoring tools track tokens used, not dollars spent. Knowing an agent used 128,000 tokens does not tell you much. Knowing it cost you 263 dollars in a single day tells you it is time to act.

To fix this, the developer built an open source tool called BurnRate.

It turns raw token data into real dollar costs and ties that spending back to the specific agent, model, or service responsible.

It also catches common ways agents waste money, like:

- Repeating the same task over and over

- Using an expensive model when a cheaper one would work

- Sending bloated, oversized prompts

- Missing cache savings it should be getting

- The most useful part is a feature called Cost Guard.

It watches spending in real time, investigates the cause when costs spike, and automatically slows down the agent responsible, no human needed to step in first.

As more companies let AI agents work on their own, tracking the dollars behind them will matter just as much as tracking their performance.

Read More

Picture leaving a helper working all weekend, and coming back Monday to a bill you never saw coming. That is exactly what is happening with a new AI coding trick called the Ralph loop.

Engineers set an AI agent to loop on its own, fixing and writing code all night. It sounds cheap, about ten dollars an hour. But that small number hides a big problem.

The loop runs nonstop, and it resends huge chunks of code and text on every single pass. That adds up fast, sometimes into six figure bills on large projects.

The loop often sends every task, even simple ones like running a test, to the most expensive AI model available.

There is no cost control built in by default. Worse, the agent talks straight to the AI provider. Nobody sees the spend until the bill lands at the end of the month.

Fifty developers running this on fifty laptops means fifty separate bills nobody is watching in real time.

The fix is simple in idea, harder in practice.

- Put a checkpoint between the agent and the AI provider.

- Route simple tasks to a cheap or free local model.

- Send only the hard tasks to the expensive model.

- Track token usage in real time, broken down by model and team, so spend shows up on a screen instead of a bill.

The bigger lesson here goes beyond one clever coding trick.

Any team letting AI agents run unsupervised needs cost visibility and controls in place before the automation starts, not after the invoice arrives.

Read More

Many teams think AI spend is simple: pay for what you use, token by token. But the real bill starts much earlier. It includes compute, data platforms, and the infrastructure needed to keep everything running.

New data from Cloudability looked at more than 1,000 companies and found a clear pattern. The more a company spends on AI, the more providers it tends to use.

Companies spending under $10,000 use multiple providers 19% of the time. That number jumps to 48% once spend crosses $10,000. It climbs to 76% at the $100,000 mark.

And once a company spends more than $1 million on AI, 88% are juggling multiple providers.

The grocery bill gets attention, but rent, utilities, and gas add up fast too. AI works the same way.

The fix is not choosing one provider and hoping to stay simple. It is building strong basics early: tagging, tracking, and forecasting that work no matter how many providers show up.

Companies that set this up now will not need to rebuild their systems every time a new AI tool joins the mix.

Plan for multiple providers before you are forced to.

Read More

The job market is hungry for certified professionals who can prove results. Don't let your company's budget leak due to a lack of specialization.

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