Key Players and Roles
On one end of the tokenomics supply chain are the organizations that make and sell AI tokens and the infrastructure beneath them. On the other are the businesses that buy those tokens and have to turn them into something worth more than they cost. Inside any organization running AI at scale are a handful of people already doing the work of managing it.
The following mapping is organized by category and by function, and not by company. The categories are not rigid: most players span more than one bucket of tokenomics, production, consumption, or value, and the same organization is often both a supplier and a consumer, buying tokens from a model lab while selling an AI product built on top of them.
How this evolves
This map is early, and it will change. Categories that look distinct today will merge, new ones will appear, and the roles will keep shifting as the work gets named and specialized. Refining all of it, the categories, the roles, and the standards that connect them, is work the Tokenomics Foundation takes on in the open, through vendor-neutral working groups. This is a starting point and a shared language, not a final word.
The supply side
The supply side is everything involved in producing a token and getting it to the point of use. It is best understood from the ground up, because each layer constrains the ones above it.
Energy and data center capacity
The physical foundation: power, cooling, and the buildings that house compute. This layer has become the binding constraint on how fast the rest of the ecosystem can grow, with power availability and long build timelines setting the real ceiling on token production.
Where it sits: production.
Silicon and hardware
The chips that perform the computation, along with the memory and interconnect that increasingly determine real-world cost. Decisions made here, particularly around memory capacity and bandwidth, shape what a token costs to produce long before it reaches a price list.
Where it sits: production.
Cloud platforms and capacity marketplaces
The providers that aggregate hardware into rentable capacity and, increasingly, into marketplaces where tokens and model access are bought and sold. They set much of the commercial structure, the commitment terms, the regions, the pricing models, that consumers have to plan around.
Where it sits: production and consumption.
Specialized inference providers
A newer category focused specifically on serving models efficiently, often at lower cost or higher performance than general-purpose clouds for particular workloads. They expand the sourcing options available to consumers and put competitive pressure on token pricing.
Where it sits: production.
Model labs
The organizations that train the models, both the frontier labs and the open-weight community. They are the direct producers of tokens and the source of most of the pricing signals the rest of the ecosystem reacts to.
Where it sits: production, with a direct line into value generation through how they price.
The platform and tooling layer
The software that sits between raw model access and a working application: orchestration, routing, gateways, observability, evaluation, and the memory and retrieval systems applications depend on. This layer is where a large share of real AI spend accumulates, and where much of the day-to-day work of managing it happens.
Where it sits: consumption.
The consumer side
The consumer side is everyone who buys tokens and has to turn them into something worth more than they cost. Two postures matter most here, and they follow different logic.
Enterprises building on AI
The largest group, and the one with the most varied needs. The key distinction is between product AI, the customer-facing features a business charges for or differentiates on, and internal AI, the tools that make the business more efficient. Product AI is judged on growth and margin and can justify more spend; internal AI is judged on cost saved. The same company usually runs both, and confuses them at its peril, because applying internal-AI cost discipline to a product-AI growth bet is a good way to starve the thing that was supposed to pay for everything else.
Where it sits: consumption and value.
Software vendors repricing on top of tokens
The companies whose own products now run on AI and who have to decide how to price that to their customers. They sit at the far end of the supply chain, where the cost of a token, the efficiency of their consumption, and their pricing model all meet in a single margin. Their pricing decisions are what carry token economics out into the broader market.
Where it sits: value.
Roles: mapping the work to people
Tokenomics is not yet a job title. It is a set of functions, most of which already exist in some form inside organizations that run AI at any scale. The roles below describe the work, the question it answers, and the titles it tends to live under today. Use them to find the people on your teams who are already doing this.
Token sourcing and procurement owner
Owns the decision of where tokens and capacity come from and on what terms: direct, marketplace, or bundled, and how much to commit to up front. The question they answer is how to secure supply without locking the organization into the wrong cost floor.
Common roles today: cloud procurement, vendor management, or a cost practitioner extending into AI.
Where it sits: production.
Inference and serving engineer
Owns how models actually run: the serving stack, quantization choices, batching, and the memory behavior that drives real cost. The question they answer is how to serve a given quality of output for the least resource.
Common roles today: ML platform engineer, ML infrastructure engineer, or a backend or SRE role that has absorbed model serving.
Where it sits: consumption.
Model routing owner
Owns which model handles which request, and the policies that route work to the cheapest model that can do the job well. The question they answer is how to capture routing savings without breaking the caching that makes those savings real.
Common roles today: often the same platform or ML engineer above, or an emerging dedicated function in larger AI organizations.
Where it sits: consumption.
AI cost analyst and forecaster
Owns visibility: who is spending, on what, whether it maps to value, and what the spend will be next quarter. The question they answer is whether the organization can see and predict its AI costs well enough to make decisions. This is the role most likely to already exist under a familiar name.
Common roles today: FinOps practitioner, cloud cost analyst, or a finance partner aligned to engineering.
Where it sits: consumption.
Governance and policy owner
Owns the guardrails: budgets, access policies, and the rules that decide what is allowed and who can override them. The question they answer is how to keep spend accountable without governance so tight it kills useful experiments.
Common roles today: engineering leadership, platform governance, or a cost center of excellence.
Where it sits: consumption.
Value, monetization and pricing owner
Owns how AI features are packaged and priced, and how the cost of tokens flows through to what customers pay. The question they answer is whether the product makes money once the token bill is accounted for.
Common roles today: product management, pricing strategy, or a business or revenue lead.
Where it sits: value.
Executive sponsors
Owns the top-level question of whether the investment is justified and how it ties to business strategy. The question they answer is the one the boardroom asks: what is all this worth, and is the return there.
Common roles today: CTO, CIO, CFO, or a dedicated AI leadership role.
Where it sits: spans all three.
Few organizations have seven distinct people here, and that is fine as this is an evolving space. The value is in checking that each function has an owner, even an implicit one, because the gaps tend to be where cost quietly escapes. A team with a strong serving engineer and no monetization owner can run AI very efficiently and still lose money on it.
How players and roles meet across the buckets
The map and the roles describe the same supply chain from two angles. Follow a single token through it. It starts as energy and capital, turned into compute by the hardware and capacity layers and into an actual token by a model lab, with the sourcing owner deciding which of those paths the organization buys through. It moves into use, where the serving and routing engineers decide how efficiently it is spent, the cost analyst tracks whether the spend is buying value, and the governance owner keeps it inside budget.
Token usage ends at the point of sale, where the pricing owner decides what the intelligence it produced is worth to a customer, and the executive sponsor judges whether the whole chain returned more than it cost.
The reason to see it as one chain is that the pieces are wired together. A commitment made badly at the supply end shows up as a margin problem at the selling end. Efficiency won in the middle is what creates the room to price competitively. No single player or role manages tokenomics on its own; the discipline is in how they connect.