License-Based, Not Per-Token: Why Our Pricing Model Changes the Math
Your AI budget for Q3 just came in 40 percent over forecast. Usage was higher than expected — which is exactly the outcome you were hoping for. Per-token pricing just punished you for deploying AI successfully. That's the math you're working against every quarter when you pay by the token.
We charge for the infrastructure, not the outcomes. The more value you get, the better our case for renewal — not a bigger invoice.
The Tax on Productivity
Per-token pricing contains an incentive problem that rarely gets named directly: every AI query costs money, which creates invisible pressure on employees to limit how often they use it. They don't skip queries because their manager told them to. They skip them because somewhere in the back of their awareness, they know that more usage equals a larger bill. That self-censorship defeats the entire purpose of deploying AI.
Call it what it is: a tax on productivity. Every time your team actually uses the AI well, your costs go up. The more effectively you've deployed it — the more workflows it's entered, the more employees have integrated it into their daily work — the higher the line item on the CFO's desk. Per-token pricing and maximum adoption are structurally incompatible goals for the same organization.
Finance teams building budgets on per-token models face variance that makes forecasting unreliable. Costs spike when quarterly reports are due. They increase when projects ramp up. They jump when new employees onboard and discover the tools. Month one: €15K. Month two: €22K. Month three: €31K. Each increase looks reasonable in isolation — usage is growing, which is supposed to be the goal. By month eight, the AI line item is the fastest-growing in the technology budget, and the CFO wants an explanation for a cost model that scales with employee engagement.
The Enterprise Math
At enterprise scale, per-token pricing becomes a different problem entirely. OpenAI's enterprise API pricing for GPT-4-class models runs €15-€25 per million input tokens — a "token" is approximately one word; a typical business email contains around 500 tokens (2024 pricing). A 1,000-employee organization where each employee runs ten substantive AI queries per day — each consuming around 5,000 tokens — generates 50 billion tokens per month. That's €1.25 million per month, before compliance infrastructure, security logging, EU data residency, and operational support.
For comparison: Leeloo's fixed license is €30K-€80K per month for the entire sovereign AI stack. That includes the AI models themselves, the routing and workflow infrastructure, data layer, security components, compliance controls, integrations, and operations support. No usage overages. No per-seat fees layered on top. No pricing change if your team has a busy quarter.
A financial services firm running sovereign AI for 200 analysts during quarterly close could generate 50 million tokens in a single month. Under per-token pricing, that's roughly €1,000 in tokens — but that firm also pays separately for data security, compliance logging, model hosting, and EU data residency. Under Leeloo's fixed license, quarterly close costs the same as the previous month. The most productive month of the year doesn't produce the largest AI invoice.
Consider the comparison: usage-based OpenAI Enterprise at enterprise scale runs €50K-€200K per month, variable, with data processed in the US, no ownership of any product layer, and pricing subject to unilateral change. Leeloo's fixed license: €30K-€80K per month, data on your infrastructure, EU jurisdiction, and a product layer you own outright.
Why Enterprise Software Is Licensed, Not Metered
Salesforce has never charged per CRM record accessed. SAP has never charged per ERP transaction processed. Oracle has never charged per database query executed. Enterprise software moved to per-seat or platform licensing decades ago, for one straightforward reason: usage-based pricing is incompatible with how enterprises plan, govern, and budget technology.
Infrastructure gets budgeted annually, depreciated over time, and evaluated against the capabilities it provides — not against the transactions it processes. When an enterprise buys an ERP system, they don't worry that a busy quarter will produce a larger ERP invoice. They know the cost on the first of January and on the thirty-first of December. That predictability is what makes it possible to treat the investment as infrastructure rather than a variable expense.
Modern AI infrastructure is no different. Organizations that have successfully scaled AI deployment removed the cost barrier to usage before rollout. Per-token pricing is a usage barrier dressed as a cost model. Every time an employee calculates — consciously or not — whether a query is worth running, the deployment has already partially failed.
Precedent exists: cloud storage moved from usage-based to fixed tiers. Enterprise SaaS moved from per-seat to platform pricing. AI infrastructure will follow the same arc. The organizations that build on fixed-license sovereign AI now own a financial model that scales with their organization rather than against it.
What Full Adoption Actually Looks Like
Fixed pricing produces a behavioral shift that consumption-based pricing never enables: employees stop thinking about the cost of running a query. They use AI for the small decisions as well as the large ones. They run a query to confirm something they already suspect. They draft a document, discard it, draft it again. They explore. That exploratory usage — the queries that don't have a clear ROI attached to them beforehand — is where the highest-value organizational learning happens.
When every query has an implicit cost, organizations optimize for measurable use cases and ignore the exploratory ones. The result is efficient on paper and impoverished in practice. Fixed pricing doesn't just change the invoice. It changes how an organization relates to its AI infrastructure.
Sales teams, relationship managers, and anyone preparing for client interactions feel the secondary effect most clearly. At a Luxembourg financial services firm running Wespher on the Leeloo Framework, pre-meeting preparation time dropped from 47 minutes to under four minutes. That change happened because employees ran the query every time, not just when they had budget certainty. The cost model made adoption possible.
The Ownership Clause
One aspect of Leeloo's pricing model rarely gets discussed before deployment but always matters after: if you stop licensing the Framework, you keep everything you built.
Front-end product layer, all customizations, all trained configurations, all data structures — yours permanently. Only the Framework components themselves require the ongoing license. The €30K-€80K per month is not payment for access to a service that disappears when you stop paying. It's payment for ongoing access to the underlying infrastructure and continued updates.
This matters for how organizations should think about the build-versus-license decision. Building sovereign AI from scratch costs €5-10M upfront, takes 24 or more months, and fails to ship more than 70 percent of the time, because the talent required — sovereign AI architects who can build production-ready infrastructure from zero — doesn't exist in the employment market. They start companies. Leeloo's implementation cost of €300K-€2M is not a license fee. It's the cost of getting a production-ready sovereign AI product in 8-12 weeks, including everything you'd have built in 24 months if you'd started from zero.
The Board Presentation That Writes Itself
AI as managed infrastructure investment, stable cost, owned product layer — that's the picture a fixed-license sovereign AI deployment presents to a board. AI as variable expense growing with adoption, no ownership, subject to vendor pricing changes — that's the alternative.
Picture the test: if AI usage doubles next year because the deployment is going well, should your AI costs double? On per-token pricing, yes. On Leeloo's fixed license, the same infrastructure handles the increased volume at the same monthly cost.
You know the number on the first of the month. It doesn't change if your team has a productive quarter, hires 50 new people, or runs a product launch that drives intensive AI usage. It's the same number on day one and day three hundred and sixty-five. That number is what makes it possible to deploy AI without a financial ceiling on adoption — and without going back to the board to explain why success cost more than forecast.