Axie Infinity is the case study every crypto game should be required to read before designing a token. Not because it failed — for a period it was the most used application in the entire industry — but because it failed in a way that is fully explained by its own mechanics, and the same mechanics are being reimplemented today.
What Axie actually did
Two tokens. AXS for governance, and SLP, earned by playing, spent on breeding new Axies. A scholar system in which NFT owners lent teams to players in exchange for a share of earnings, which turned the game into a job for players in lower-income countries.
The emissions side grew faster than any sink in the game. Breeding consumed SLP, but player growth meant more breeding, which meant more Axies, which meant more SLP being earned. As long as new entrants kept buying in, the price held and scholar earnings stayed above the cost of playing. When that inflow slowed, SLP collapsed, scholar earnings fell below the cost of playing, players left, and the economy unwound in months. The structure was not a bug: any economy whose payouts depend on recruiting new capital is a Ponzi with a game attached.
The design errors, generalised
Four failures recur across every dead game economy, and a project failing two of them is very likely to fail entirely.
Emissions outrunning sinks. If the system produces tokens faster than it destroys or absorbs them, the price of the earn token is a countdown, not a variable.
A single earn token with no non-speculative demand. SLP was only wanted because it was worth something. Nothing in the game required holding it for reasons other than price.
Growth that requires recruitment. If the economy needs new players to fund existing players, the growth rate is the solvency condition, which is an unsustainable place to be.
Yields denominated in the falling asset. Payouts set in a token whose price falls as the player count rises are a mechanism that accelerates its own collapse.
What a sustainable economy needs
Four components, and the absence of any one of them is what a post-mortem will eventually call the cause.
Sinks that scale with progression and burn rather than recycle. Upgrades, breeding, entry fees, consumables, and cosmetic exclusivity remove tokens from circulation permanently. A sink that transfers tokens to another player is not a sink.
Multiple tokens with distinct jobs. A fixed-supply governance token, a volatile earn token, and a stable in-game unit behave differently under stress, and separating them lets you adjust one without repricing everything.
NFT utility outside speculation. The asset has to be useful in the game. An NFT whose only function is a claim on future players is a share in the recruiting scheme, labelled as an item.
Emission curves that decay on a schedule. A published decay is a commitment; a governance vote to cut emissions is an admission that the original numbers were wrong.
The technical stack changed, and it matters
The infrastructure has genuinely improved in ways that remove the biggest onboarding blockers.
ERC-4337 paymasters can sponsor gas, so players do not need to hold a native token to act — historically the single largest drop-off point in any crypto game.
Embedded wallets remove seed phrases for a mainstream audience, and account abstraction lets the game manage keys in a way that is still user-owned.
Off-chain game state with on-chain settlement keeps the cost of play reasonable, by only putting assets and outcomes that matter on the chain.
Chain choice is now a real trade. A general-purpose L1 gives composability; an app chain or a dedicated gaming rollup gives cheap, predictable throughput. The right answer depends on whether your game needs to interact with the rest of DeFi, which most games do not.
The current models
Five patterns have survived, and each works for a different reason. Esports and wagering works because the stakes create demand that is not speculation. Social collection works because status is a real consumer of supply. Mobile-first casual works because the audience is large and the per-player cost is low. Infrastructure works because it earns fees from games rather than from players. And virtual worlds with land works when the land has a use that is not rent.
The test that settles it
Ask whether removing the token would leave a game people still play. If the answer is yes, the token is a monetisation layer and can be designed accordingly. If the answer is no, the token is the product, and the economy will behave like one — which means it will be measured against every other speculative asset rather than against other games.
References
Next, read Yield Farming Risks 101. More in Web3 Applications.