Tokenomics

What the word covers when it is used precisely

Tokenomics is the economic design of a digital asset. Fidelity Digital Assets puts the scope plainly: the framework determines how value flows through a network, which participants benefit, and what flows may strengthen or weaken that value over time.¹ That is a wider brief than most people expect from the word. It covers issuance. It covers who gets what and when. It covers what the token is for, and it covers who is allowed to change any of it later.

Academic work lands in the same territory with different vocabulary. A published token economy design method treats tokenomics as one of three coupled design dimensions, alongside incentives and governance, and defines it as the rule set for issuance, distribution, allocation and burning.² A 2025 framework paper describes it as the strategic design and management of token economies, listing utility, supply mechanisms, distribution strategies and incentive alignment as the components in scope.³

Our own framing is narrower than either. Tokenomics enhances value creation, it does not manufacture it. If the underlying business does not produce something people pay for, no supply curve fixes that, and the design work becomes an elaborate way of moving a shortfall into the future. The token is infrastructure. The business is the engine.

Six parts, and the fact that they are coupled

Supply is the total that can ever exist, the amount that exists now, and the schedule connecting the two. Distribution is who holds it, on what terms, behind which lockups. Utility is what the token does that nothing else in the system can do. Incentives are the rewards paid to buy behaviour the protocol needs. Governance is who can change the parameters and by what process. Valuation is how the market prices the whole arrangement, which is where fully diluted valuation and circulating market cap stop agreeing with each other.

The coupling is what gets missed. Shorten a vesting schedule and float rises, which lowers the ratio between fully diluted valuation and market cap, which changes what a market maker has to absorb at listing, which changes the price at which your earliest incentives were denominated, which decides whether those incentives still buy the behaviour they were sized for. One parameter, five consequences. A pie chart shows none of them.

This is why we treat a token economy as something to simulate rather than something to write up. Five coupled variables cannot be reasoned through by argument, and the interactions that matter are the ones nobody predicted in the meeting.

What separates a design from a pie chart

The allocation chart answers exactly one question: who received what percentage. Every question that decides whether the design survives contact with a market sits outside it. When do those allocations unlock, and does the schedule cluster or stagger. Who can change the emission rate after launch, through what process, with how much notice. Is the treasury behind a multisig, a timelock, a governance vote or one admin key. Does the valuation being quoted refer to circulating market cap or to fully diluted value, because the gap between those two is routinely an order of magnitude.

Industry teaching material now organises the discipline into six pillars covering supply and monetary policy, distribution and allocation, vesting and unlocks, utility and demand drivers, incentives and game theory, and governance and control, with explicit evaluation questions about unlock transparency and who holds parameter change rights.⁴ Four of those six are invisible on a pie chart.

The tell we look for is whether a design states its own failure conditions. A real one says: this holds while monthly sinks exceed monthly emissions, here is the measurement, and here is what we do in the quarter that stops being true. A deck says the token accrues value from protocol growth and moves to the next slide. Across the projects we have advised, the second version is far more common, and its presence predicts a rough first year better than any other single signal we track.

A worked pass on one allocation table

Take a hypothetical one billion token supply, with numbers chosen to be ordinary rather than extreme. Team 18 percent, investors 20 percent, ecosystem emissions 25 percent, treasury 25 percent locked, and 12 percent circulating at launch. Team and investor tranches use the common structure: a twelve month cliff releasing a quarter of the allocation, then linear monthly vesting across the following three years. Ecosystem emissions run evenly across four years.

Run it to month twelve. Circulating supply is the 120 million launch float plus twelve months of emissions at 5.21 million a month, so roughly 182 million tokens. On the cliff date the team cliff releases 45 million and the investor cliff releases 50 million. That is 95 million tokens arriving on one day against 182 million circulating: a 52 percent increase in float, on a date that was fixed the moment the term sheet was signed.

Month thirteen onward is the quieter problem. Team vesting adds 3.75 million a month, investor vesting 4.17 million, emissions another 5.21 million. Call it 13.1 million tokens a month entering a market with 277 million circulating. At a hypothetical fifty cent price that is 6.6 million dollars of potential monthly supply. If protocol fee revenue routed to buybacks is 200,000 dollars a month, the structural bid covers about three percent of it. Nothing in that paragraph is a forecast. It is division.

That ratio is the number we want on the table in the first meeting, and it is the number an allocation chart is structurally incapable of showing. The fix is rarely a different pie. It is a longer schedule, a smaller raise, a real sink, or a business that earns more.

The other tokenomics: per-token cost in AI systems

Search results for this word now split across two unrelated fields, so it is worth naming which one you are in. In AI, token economics refers to the cost structure of running language models, where a token is a metered slice of text and providers price input and output tokens separately. A 2026 arXiv paper uses the exact phrase for that subject, modelling the pricing of tokens and computation in foundation models.⁵ Adjacent work formalises inference economics as the cost of GPU time per token generated.⁶

The two senses share the word and almost nothing else. A blockchain token is a transferable unit of ownership or access, with a supply schedule, holders, a market price and often governance rights attached. An AI token is a unit of consumption. It has no holder, no supply cap, no secondary market and no governance. Vesting and float mean nothing in the AI sense. Cost per million and cache hit rate mean nothing in the crypto one.

Both are real disciplines and the collision is nobody's fault. If you arrived here looking for the cost side of AI products, we keep that subject on its own page at /ai-tokenomics/ rather than blending the two. Everything else on this page uses the crypto sense.

How these designs actually fail

The most common failure is not a broken mechanism. It is a mechanism working exactly as written against demand that never shows up. Emissions buy usage, usage reads as traction, emissions taper, usage leaves with them. The design was not wrong. It was funded by its own inflation and mistaken for product-market fit.

The second is velocity. A token people must hold to use, then sell the moment they have used it, turns over fast and holds no floor. Utility that can be rented per transaction creates no reason to hold anything, and bolting a staking reward onto it only pays people to delay the same sale.

The third is governance concentration that only becomes visible under stress. Quorum thresholds set for convenience at launch turn into a single holder veto once participation drops, and participation drops as soon as voting costs gas and nothing controversial is on the ballot. Then something controversial appears and the threshold is already captured.

None of these show up in the first three months. All of them are legible in the design before launch, which is the whole argument for doing the work before the term sheet rather than after the listing.

What we settle before an allocation table exists

In order: what the business sells and to whom, because that decides whether a token is load bearing or decorative. Then whether the token is needed at all, tested against a version of the product that does not have one. Then supply and emissions, sized against the sinks that will absorb them rather than against a target raise. Then distribution and vesting, checked for cliff clustering across every tranche at once rather than tranche by tranche. Then governance and admin powers, written out as a capability matrix naming who can change what. Valuation comes last, because it is an output.

Reverse that order and you get the pattern we see most often: a raise sized first, an allocation table reverse engineered to fit it, and a utility story written afterwards to justify both. The mechanism work then spends its entire budget compensating for decisions made before anyone opened a spreadsheet.

One boundary on all of the above. None of this is investment advice and none of it is a view on whether any specific token is worth holding. It is design reference for the people building the thing.