Managed service providers (MSPs) looking to build a practice around optimizing artificial intelligence (AI) infrastructure should start paying closer attention to the Tokenomics Foundation, which has now been formally launched.
As an arm of the Linux Foundation, the Tokenomics Foundation is focused on bringing greater clarity to the often opaque billing models used to charge for AI infrastructure resources delivered through the cloud.
Founding members include JPMorgan Chase, BNY, GoDaddy, Hitachi, Lenovo, Oracle, SAP, ServiceNow, Broadcom, Accenture, IBM, Cast AI, and Flexera.
Creating a common language for AI consumption
In a draft document dubbed “Big-T Notation,” the Tokenomics Foundation has already outlined a structured framework for classifying tokens. That effort is significant because AI token consumption occurs both when submitting prompts and when models generate responses. Without a standardized way to classify tokens, comparing AI infrastructure costs and performance across providers becomes difficult.
The foundation is also working to establish standardized definitions for concepts such as token value, token density, and the distinctions between input, output, reasoning, and cached token types.
Building a better model for AI cost management
Beyond terminology, the Tokenomics Foundation is developing a comprehensive cost-of-AI reference model. The framework is based on a nine-layer cost stack that places token charges in context with compute, storage, data, networking, embedding, engineering labor, and training expenses.
The foundation is also exploring alternative pricing approaches that could shift billing from token-based consumption to a model based on application programming interface (API) calls.
A growing opportunity for MSPs
Given the rising cost of AI, it’s a matter of time before more organizations turn to MSPs for help managing AI budgets. While many AI pilot projects are underway, businesses find they can realistically fund only a limited number of initiatives. The more efficiently they use available AI resources, the more projects they can ultimately move into production.
Organizations are also increasingly focused on identifying instances of “tokenmaxxing,” where AI use cases are deployed with little regard for business value, resulting in the unnecessary consumption of limited AI resources. As noted previously, many businesses are looking for ways to reduce wasted AI spending and resource utilization.
Of course, some organizations are choosing to deploy AI models in on-premises environments that they control. This eliminates the need to monitor token consumption. Even in those scenarios, however, they still need visibility into which projects are using infrastructure resources most efficiently.
What is certain is that the era of funding virtually every AI initiative has ended. As AI adoption grows and processor resources remain limited, organizations are becoming more focused on AI infrastructure costs. Where there is operational pain, there is opportunity for MSPs. The challenge is helping customers understand AI token usage. This includes what resources are being consumed, by whom, how often, and at what cost.
Photo: Studio Romantic / Shutterstock
