Presented by:

Want to appear here? Talk with us

Wasted cloud spend rose to 29% this year, which was the first increase in five years (Flexera).

FinOps Weekly's Victor Garcia gets into where the remaining waste hides inside resources, why billing data never flags it, and how to decide what's safe to delete.

Live only at the end: he breaks down audience-submitted waste on air. 22 minutes, 11am ET / 5pm CET, hosted by Eon.

AI costs are becoming a real budget line item, and this article breaks down how to manage them. The author calls this practice Tokenomics.

Think of it like a five-layer stack, from computer chips at the bottom to smart routing rules at the top.

Each layer builds on the one below it. Savings at the chip and hardware level flow upward and multiply.

The inference layer, which handles caching and batching, offers the biggest chance to cut costs right now.

Model selection matters too. Sending simple requests to cheap models and only using expensive ones when needed can save a lot of money.

Aim to send 80 percent of requests to cheaper models.

Keep failed downgrade attempts under 5 percent.

Track spend per user so you know who is driving costs.

As companies spend more on AI, treating token costs like cloud costs makes sense.

The same discipline that controls cloud bills can control AI bills too.

Read More

ChatGPT Voice costs drop 75% for Enterprise and Edu workspaces, falling from 5 credits per minute to 1.25 credits per minute for clearer, more predictable spend.

GPT-Live 1 launches with per-second billing at $0.05 per minute, letting teams measure voice API costs more precisely.

New analytics link AI usage to business value, helping FinOps teams build clearer chargeback and ROI reports for ChatGPT Work and Codex.

Gemini Enterprise pay-as-you-go now covers all invoiced Cloud Billing projects, giving more teams flexible, usage-based purchasing instead of fixed subscriptions.

New Gemini Enterprise subscriptions drop bundled Code Assist access, so teams should recheck entitlements to avoid surprise tooling costs.

Claude Code adds chargeback rate multipliers up to 10x, letting teams apply internal markup while keeping usage data accurate for billing.

Cost and token data now shows exact tools and agents used, making it easier to attribute spend and optimize by workload.

Read All AI Provider Updates

We sit down with AWS experts Adam and Loïc to discuss AI cost optimization on AWS and how FinOps practices can help organizations gain visibility, allocate spend, and improve the efficiency of their AI workloads.

AI bills can trick you before you even see a real invoice.

A FinOps writer walked through a simple math problem that shows how easy it is to overestimate AI costs.

He looked at one million model requests using Claude Haiku 4.5. A flawed cost estimator put the bill at $15,000.

Fixing one small error, how cached tokens get priced, dropped that number to $7,800.

That is a $7,200 gap caused by nothing more than mixing up two pricing categories.

AI providers split charges into different buckets: normal input, cached input, and output. Cached tokens usually cost much less.

If your tool lumps cached tokens in with normal ones, your estimate balloons, and you might cut a feature or switch models for no real reason.

The fix is not just recounting tokens either. Providers can bill for things you never see, like hidden "thinking" tokens, so even a careful recount can miss real charges.

To actually confirm a bill, you need five things: the call record, the usage breakdown, your recount method, the price sheet, and the real billing statement.

Only when all five match can you trust the number. Check your estimator's math before you touch the budget.

A phantom $7,200 problem is not a savings opportunity, it is a bug.

Read More

Measuring AI's return on investment is still messy for most companies, and this article shows why.

Many enterprises struggle to connect AI productivity gains to real business value like revenue or cost savings.

A Gartner analyst points out that companies often see efficiency in areas like customer service or finance operations, but cannot translate that into clear dollar figures.

Around half of AI projects get dropped after testing due to bad data, rising costs, or unclear value, according to a 2026 Gartner report.

Smaller companies seem to be finding more success.

An advertising agency called OBI Creative used AI tools to cut back-and-forth work with clients, boosting profit margins and growth by 20% year over year, even with higher overhead costs.

Cornell University took a different path, focusing less on cost or token usage and more on how AI supports its core mission of teaching and research.

Companies should decide what value they need before starting an AI project, not after.

Most projects should show a return within one year. Vendors also need to sell outcomes, not just tools.

The bottom line: AI can pay off, but only when leaders define what success looks like from day one.

Read More

AI agents run on tokens, and tokens cost real money.

This article breaks down where those costs come from and how teams can cut them without losing quality.

Here are the key ideas for FinOps teams watching AI spend:

Write shorter, structured prompts. Plain English instructions can burn 1,000 to 2,000 tokens before the real question even gets asked. Switching to compact formats like JSON cuts that waste.

Match the model to the task. Simple jobs like sorting or tagging can run on cheap, light models. Save the expensive, powerful models for hard reasoning steps only.

Stop repeating yourself. Instead of sending full conversation history to every agent, summarize it. Less repeated text means fewer tokens and faster replies.

Use outside storage for agent notes. Keep working details in a database instead of the AI's memory window. This keeps costs down while still giving a full record of what happened.

Cache prompts that do not change. Reusing saved instructions avoids paying for the same setup over and over, as long as the cache stays current.

Load tools only when needed. Loading every possible tool for every request adds cost. Pulling in only the right tools for the job saves tokens fast.

The bottom line is simple. Token spending adds up quietly across every layer of an AI system, and small fixes at each layer can lead to real savings.

Read More

Measuring the real value of AI is tricky, and this framework from the Tokenomics Foundation gives FinOps teams a clear way to do it.

The core idea: separate value claims from ROI. A value claim shows the benefit of an outcome against a baseline.

ROI needs cost added in, using Total Cost of AI.

The framework breaks value into three groups.

Direct dollar wins, like new revenue or avoided spend.

Labor gains, like freed-up hours or automated tasks.

Softer outcomes, like better quality, faster delivery, lower risk, or happier employees. Every claim follows four steps.

- Observe the raw numbers.

- Net them into one trusted figure.

- Convert that figure into dollars or another useful unit.

- Book it into the right ledger line, so no dollar gets counted twice.

- The code review example makes this real.

An AI tool cut review time from 45 minutes to 27 after accounting for extra triage work.

That saved time became freed capacity for one team, and avoided contractor spend for leadership.

Only one of those got booked as a dollar, since they described the same win. Track value the same careful way you track cost.

Clear rules stop teams from double counting or overstating what AI actually delivers.

Read More

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

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