ARK Tokenomics

Fair launch. Zero tokens exist at genesis. No premine, no investor allocation, no team tokens, no airdrop. Every ARK in circulation is minted against verified inference work.

Supply

Parameter	Value
Hard cap	1,000,000,000 ARK
Decimals	9 (1 ARK = 10⁹ ark_atom)
At genesis	0 ARK
Practical maximum	~795M ARK (geometric halving converges)

Emission schedule

ARK is minted per-epoch (3,600 blocks ~ 1 hour) against verified inference output. Annual emission halves every 4 years after Year 4:

Year	Annual emission
0 (launch)	120,000,000 ARK
1	105,000,000
2	90,000,000
3	80,000,000
4-7	50,000,000
8-11	25,000,000
12-15	12,500,000
16+	Halves every 4 years

Unspent epoch budgets roll forward — no emission is ever burned or lost.

Reward distribution

Every settled inference job distributes rewards across five recipients:

Recipient	Share	Purpose
Compute node	80%	The operator who ran the model
Verifier	7%	Re-executed the job for correctness
Treasury	5%	Governance-controlled fund
Delegators	5%	Pro-rata to stakers on the compute node
Burned	3%	Deflationary pressure

Simple split. The network has two node types (validator and compute) with no routing middlemen. Compute nodes earn the largest share because they do the work — running GPU inference for every request.

Reward multipliers

Block rewards scale based on model characteristics and node performance (all integer arithmetic, no floating point):

Model size — 1x (7B), 1.5x (13B), 2.5x (30B), 4x (70B), 10x (400B+)
Model category — 1x text, 0.3x embedding, 2x image, 1.5x audio, 3x multimodal
Latency bonus — up to 1.2x for beating expected TTFT
Uptime bonus — 1.1x for >95% uptime over 30 days

Bootstrap period

arknet launches with zero tokens in circulation (fair launch). To solve the chicken-and-egg problem, the chain starts in a bootstrap period:

Parameter	Value
Duration	6 months (~15.5 million blocks)
Validator target	100 unique validators
Exit condition	Whichever comes last — both must be met
Min stake during bootstrap	0 — anyone can participate
Inference cost during bootstrap	Free — rate-limited per wallet
ARK minted during bootstrap	~60,000,000 ARK from emission

During bootstrap, compute nodes earn ARK from block emission for serving free inference requests. This bootstraps the token supply from zero — same model as early Bitcoin mining. By the end of bootstrap, there's enough ARK circulating for the paid economy to activate naturally.

No early-mover monopoly. The validator target is set to 100 (not 5 or 10) specifically to prevent one operator from spinning up many nodes and exiting bootstrap early. The 6-month floor ensures broad distribution regardless of how fast validators join.

Staking

After bootstrap, every role requires stake:

Validators — secure consensus, run seed/relay, earn block-production rewards
Compute nodes — stake proportional to model size served
Verifiers — stake to earn verification rewards

Delegation supported: stake on any node operator and earn the 5% delegator cut pro-rata.

Unbonding period: 14 days. Redelegate cooldown: 1 day.

Slashing

Dishonest behavior loses stake:

Failed deterministic verification — compute node produced wrong output. 10% of stake slashed.
Double-signing — validator signed conflicting blocks. 5% slashed.
Downtime — validator missed too many blocks. 0.1% slashed per offense.

Slashed amounts split: 50% burned, 30% to reporter, 20% to treasury.

Fee market

EIP-1559 style: base fee adjusts per block based on gas utilization. Base fees are burned. Tips go to the block proposer.

Dynamic pricing

Per-token inference prices adjust at each epoch boundary using a utilization-based oracle with adaptive EMA smoothing. High demand raises prices; low demand lowers them. Floor: free tier. Ceiling: governance-adjustable.

Governance

On-chain proposals with deposit (10,000 ARK), discussion period (7 days), voting period (7 days). Quorum: 50%. Approval: 66%. Veto threshold: 33%. Passed proposals can adjust protocol parameters, emission rates, and fee targets.