papers/hanzo-ai-chain/sections/03-ai-coin.tex

\section{The AI Native Coin}
\label{sec:aicoin}

\subsection{Token vs.\ Coin}

LUX is the native coin of the Lux primary network and pays for security and
gas across all Lux chains. AI is the native coin of the Hanzo Network and
is independent of LUX. The two coins are not the same asset, and neither
is a wrapper of the other. AI is minted on the AI Chain, paid for compute
on the AI Chain, burned proportionally to AI Chain fees, and staked by AI
Chain operators. LUX is unaffected.

This separation is intentional. Security inflation (LUX) and AI compute
inflation (AI) have different elasticities and different stakeholder
sets; bundling them into one token would distort both. The architecture
parallels how A-Chain (LP-134) separates the gas token from
application-layer orchestration tokens such as AI and ZOO.

\subsection{Supply Schedule}

\begin{table}[h]
\centering
\small
\begin{tabular}{lr}
\toprule
\textbf{Parameter} & \textbf{Value} \\
\midrule
Genesis supply              & 1{,}000{,}000{,}000 AI \\
Block reward (initial)      & 4 AI / 500\,ms block \\
Halving interval            & every 4 years \\
Long-run terminal supply    & 2{,}000{,}000{,}000 AI \\
Activation block reward     & 2025-12-25 (block 0) \\
\bottomrule
\end{tabular}
\caption{AI supply parameters at genesis. The block reward halves every
four years, asymptoting at the terminal supply.}
\label{tab:supply}
\end{table}

The genesis supply is allocated as follows: 30\% public sale and ecosystem
liquidity, 25\% operator and agent rewards reserve, 20\% Hanzo Industries
treasury (4-year linear vest), 15\% research and core contributors
(4-year linear vest, 1-year cliff), 10\% Hanzo Foundation grants.

\subsection{Block Reward}

Each block emits the current epoch reward $R(t)$:

\begin{equation}
  R(t) = R_0 \cdot 2^{-\lfloor t / T_h \rfloor}
\end{equation}

where $R_0 = 4$ AI, $T_h = 4$ years, and $t$ is wall time since activation
(2025-12-25). The reward is split:

\begin{itemize}[leftmargin=1.2em]
  \item 60\% to compute operators in proportion to attested
        per-block work (\S\ref{sec:markets});
  \item 30\% to validators in proportion to QuasarCert participation
        weight (LP-105);
  \item 10\% to the agent rewards pool, paid out as
        capability-weighted micro-grants to active agents
        (\S\ref{sec:agents}).
\end{itemize}

\subsection{Fee Burn}

Every paid action on the AI Chain carries a base fee in AI: inference
calls (per token), training bounty escrow, GPU lease payments, registry
mutations, and standard EVM gas. The base fee follows the same
EIP-1559-style mechanism the EVM uses, but with a different burn split:

\begin{equation}
  \text{base\_fee\_burned} = 0.7 \cdot \text{base\_fee}, \quad
  \text{base\_fee\_to\_treasury} = 0.3 \cdot \text{base\_fee}.
\end{equation}

The 70/30 burn ratio is calibrated such that, at sustained network usage
above a threshold $U^\star$ (defined as 50\% of theoretical block-fee
capacity), net issuance becomes negative; AI becomes deflationary under
heavy load. Below $U^\star$, the 60/30/10 block reward dominates and net
issuance is positive, subsidizing operator bootstrap.

\subsection{Staking and Validator Yield}

AI Chain validators stake AI to participate in Quasar 3.0 consensus on
this chain. Stake mechanics inherit from LP-105 §3 (Quasar Validator
Set): a validator must contribute to all three lanes (P, Q, Z) to earn
rewards; lane-incompleteness slashes proportionally.

Operator staking is a separate, second-tier deposit governing right to
list compute capacity (\S\ref{sec:markets}). Operator stake is locked
against attestation fraud: a single proven false TEE quote slashes 100\%
of the operator's stake, distributed to the verifier and to the burn
address.

\begin{table}[h]
\centering
\small
\begin{tabular}{lcc}
\toprule
 & \textbf{Validator} & \textbf{Operator} \\
\midrule
Min stake                & 100{,}000 AI & 10{,}000 AI \\
Lane participation reqd. & P + Q + Z    & TEE quotes \\
Slashing for downtime    & 0.1\%/epoch  & — \\
Slashing for equivocation& 100\%        & — \\
Slashing for false quote & —            & 100\% \\
Target net yield (year 1)& 7--9\%       & 12--18\% \\
\bottomrule
\end{tabular}
\caption{Validator vs.\ operator stake. Validators secure consensus;
operators sell compute. Both stake AI; both can be slashed; yields differ
because operator workloads carry capital and operating costs that
validator-only operation does not.}
\label{tab:stake}
\end{table}

\subsection{Worked Example: Cost of an Inference Call}

A caller submits an inference job to a 70B-parameter HMM at a posted
price of $p = 0.000\,002$ AI per output token. The job produces 512
output tokens. Total payment $P = 0.001\,024$ AI, decomposed as:

\begin{itemize}[leftmargin=1.2em]
  \item Operator share: $0.85 \cdot P = 8.704 \times 10^{-4}$ AI;
  \item Model owner royalty: $0.10 \cdot P = 1.024 \times 10^{-4}$ AI;
  \item Base fee: $0.05 \cdot P = 5.12 \times 10^{-5}$ AI, of which
        $3.584 \times 10^{-5}$ is burned and $1.536 \times 10^{-5}$ is
        routed to the treasury.
\end{itemize}

The operator's GPU produced a TEE quote covering this job; the receipt
is recorded in the next block's QuasarCert. The caller can verify the
receipt on any device with a single Groth16 pairing check.