Glassity at Tokenomicon + FinOps X Amsterdam

What Tokenomicon and FinOps X are about

The event brings together two groups that usually work separately.

FinOps teams check whether the AI bill matches the budget. Tokenomics looks at whether the same AI work can be done for less money, without the results getting worse.

In most companies, one team does one of these jobs and nobody does the other. We see this in almost every company we talk to.

Why companies can't see who spent their AI budget

A company usually has a FinOps team, and they are good at their job. They split cloud costs between teams and they know who owns which service.

Then the AI bill arrived, and their tools can't read it. Cloud cost tools don't understand tokens.

The engineers understand tokens very well. They know which model costs the most and why the bill went up last week. But this information stays with them. It never reaches the reports that finance and management use.

So nobody can answer two simple questions. Which team spent this AI budget? And how much did this service cost last month, including the AI it used?

Why AI cost attribution is hard

AI usage data is stored in engineering tools. A gateway is the tool engineers use to send requests to AI models, and it records which key or user made each call. But it was set up to keep things running, not to report costs, and people in finance never open it.

The invoice from the AI provider has no names on it. You get one total. If three teams share an account, they share one line on the bill. Only one gateway, LiteLLM, has a team field. Everywhere else, there are mostly API keys.

Cloud costs and AI costs also come in different formats. Cloud billing has a standard called FOCUS, but there is no agreed way yet to put AI usage into it. FOCUS 1.5, planned for December, will add one. Until now, teams who wanted to combine the two had no common way to do it.

What Glassity is launching on 22 September

On 22 September, the day the event opens, we're launching AI cost attribution in Glassity.

It reads token counts and costs from the tools a team already uses: LiteLLM, Helicone, OpenRouter, Cloudflare AI Gateway, Bifrost, Amazon Bedrock or Microsoft Foundry. It shows them every hour, by source, key, model and provider. Input, output, cache and reasoning tokens are counted separately wherever the source reports them.

It writes this data in FOCUS format, using columns the standard already has, in the same place as your cloud billing data. So you can look at both together, and ask the Glassity Agent questions about them in plain language, like "which team spent the most on AI last month?" or "why did our AI costs go up last week?"

This doesn't solve everything. Knowing who spent the money is one part of the work. Knowing what the AI actually gave you back is harder. For the work our own Glassity Agent does, we already show what it cost in AI next to what it saved on the bill. For teams' own AI use, nobody has solved that part well yet, including us. But you can't start on it until your cloud costs and AI costs are in one place.

Meeting Ernesto in Amsterdam

Ernesto will be at the reception on the evening of the 22nd and at the full programme on the 23rd. If you're working on AI cost attribution, he's happy to share how we approach it and what we've learned so far.

You can also write to him at ernesto@glassity.cloud.

If you can't come to Amsterdam, the FinOps Foundation is streaming the event as part of its September Virtual Summit.