What Is Tokenomics, And Why Your AI Infrastructure Is Now a FinOps Problem - Cast AI The 2026 State of Kubernetes Optimization Report is here 🎉 Download it to learn how teams are fighting waste.
Skip to content Blog
Cut your cloud bill
Book a demo<br>Sign in<br>Get started
News and insights<br>What Is Tokenomics, And Why Your AI Infrastructure Is Now a FinOps Problem<br>I was in the room when tokenomics became official. Here is what it means for every FinOps and infrastructure team.
Laurent Gil<br>Jun 10, 2026
Table of contents<br>Will display ToC list here.
When you are in a room with a thousand FinOps practitioners and a Goldman Sachs chart goes up on the main screen projecting usage of 120 quadrillion tokens in three years, you feel it differently than reading about it. Just think about it – 120 quadrillion token usage in the next 3 years, and we are currently at 6 quadrillion, it’s at least 20x growth.
That was yesterday morning at FinOps X in San Diego, where J.R. Storment announced the Tokenomics Foundation from the main stage and, in doing so, marked what I believe is the biggest evolution the FinOps discipline has ever seen.<br>J.R. Storment announcing the Tokenomics Foundation from the main stage at FinOps X, San Diego – June 2026Tokenomics, in the enterprise AI context, is the discipline of governing how energy and capital are converted into AI tokens, how those tokens are consumed efficiently, and how that spend connects to business value. Three layers: production, where your GPU infrastructure manufactures tokens; consumption, where model routing, caching, and prompt architecture determine what those tokens actually cost; and value, where spend maps to outcomes. This is why AI infrastructure is now a FinOps problem: the token bill starts long before the model provider invoices you. It starts in your Kubernetes clusters, your GPU fleet, and your autoscaling decisions.<br>Here are four things from yesterday that every FinOps and infrastructure team needs to hear.<br>1. The invoice is not the cost. And I see this every day.<br>Pooja Kumar, VP CTO Transformation and FinOps at Prudential Financial, said it from the main stage, and I agree completely: "AI is just another workload" is the most dangerous lie a FinOps team can tell itself. We have customers spending six to seven times more on GPU inference than on cloud. That gap does not show up on the model invoice. It lives in retry storms, in agentic chains where one prompt triggers 20 model calls beneath it, in GPU nodes reserved just in case, and sitting idle when the job finishes early. These are infrastructure failures before they are token failures. You cannot govern what you cannot see, and right now, most teams are only seeing the tip.<br>2. A token is not the same as another token. This is the unlock.<br>Frederik Pohl, VP, Head of FinOps and Data Solutions at SAP, and his colleague Maida Nazifi, Data Scientist at SAP, showed 12 months of exponential token growth with cost per token falling and total spend still doubling. As J.R. mentioned on stage, Pinterest is already tracing costs from silicon through to model routing with cost and performance data attached at every architectural decision. As J.R. noted from Adobe’s experience on stage, model routing is powerful until it breaks your cache: routing to a cheaper model that invalidates a warm cache can make it more expensive than the one you replaced.. These are second-order effects that do not show up in standard tooling. None of this surprises me. LLMs are not just different in quality; they are different in specialization. Roughly 15% of a developer’s work genuinely requires a frontier model, based on what we see across Cast AI customers. The other 85% does not. The teams winning right now are the ones letting the infrastructure decide which model runs which task. You describe the outcome. The system routes, grades, and iterates. You only care that you got an A. At Cast AI, this is how we build, and it is how we help our customers build. The era of manually picking models is ending the same way manual node provisioning ended. It is better because it is finished.<br>3. FinOps just got its biggest mandate. We need to all show up for it.<br>J.R. Storment announced the Tokenomics Foundation yesterday morning: vendor-neutral, inside the Linux Foundation, running alongside the FinOps Foundation. Token costs have done something cloud optimization never quite managed: they have put FinOps practitioners in the middle of boardroom discussions.Mike Eisenstein, Managing Director at Accenture, said it clearly on stage: "this is our moment." The question is whether your team shows up with the full stack answer or just the invoice analysis.<br>The full Day 1 keynote is below. Watch the room react in real time.
The infrastructure layer, the model selection layer, and the business outcome layer. All of it. The organizations building that governance foundation today will set the unit economics that define the next decade. Everyone else will...