Ankh AI : Vers une Infrastructure Décentralisée pour l'Intelligence Artificielle Collaborative
Serge Destin TAMPOLLA1
14IR Advisor — serge.destin@tampolla.com — www.tampolla.com
Mai 2026
This paper presents Ankh AI, a decentralized protocol for collaborative artificial intelligence. Built on a dedicated Cosmos SDK blockchain with 3,000+ TPS, Ankh AI coordinates three fundamental pillars — distributed compute power, collective human expertise, and data resources — through a unified incentive mechanism based on the $COLLAB token. We introduce a hybrid verification system (ZK-SNARK + optimistic verification + VRF sampling) that guarantees model integrity without sacrificing scalability. Our tokenomics design features a 1B immutable supply cap, deflationary burn mechanisms, and quadratic voting governance. We demonstrate how this architecture creates powerful network effects while ensuring equitable value distribution among all contributors.
Keywords: Decentralized AI, Blockchain, Tokenomics, DAO Governance, Hybrid Verification, Federated Learning, ZK-SNARK
Artificial intelligence has become the defining technology of the 21st century. Yet its development remains concentrated among a handful of technology corporations that control the models, the data, and the computational infrastructure. This centralization creates systemic risks: algorithmic censorship, unshared value extraction, and vulnerability to unilateral decisions.
Recent work in federated learning (McMahan et al., 2017) has demonstrated the feasibility of distributed model training. Zero-knowledge proofs (Ben-Sasson et al., 2014) have matured to the point where they can verify computations without revealing underlying data. Blockchain technology, particularly appchains built with Cosmos SDK, now offers the throughput and finality required for real-time coordination of distributed systems.
However, no existing protocol unifies these three pillars — compute, talent, and data — into a single coordinated system with equitable incentives. Ankh AI addresses this gap by introducing a protocol-level integration that creates powerful network effects: compute providers attract developers, who attract data providers, who improve models, which generate more inference demand.
The Ankh AI protocol consists of five interconnected layers:
Built on Cosmos SDK with Tendermint BFT consensus, the dedicated appchain achieves 3,000+ TPS with sub-3-second finality. This performance is critical for real-time coordination of training tasks and inference requests.
Six core smart contracts coordinate operations:
Our verification system combines three complementary approaches:
ZK-SNARK verification (via ezKL) proves inference correctness without revealing model weights. Optimistic verification accepts results by default but allows 24-hour challenge periods. VRF sampling randomly selects verification committees to audit computation batches.
This three-pronged approach achieves security comparable to full replication at a fraction of the computational cost.
Data persistence uses IPFS for content-addressable storage, Filecoin for incentivized long-term storage, Arweave for permanent model checkpoints, and TEEs for sensitive operations requiring hardware-level isolation.
The protocol creates a virtuous cycle through the coordination of its three pillars. Compute providers receive 70% of inference fees. Talent contributors earn perpetual royalties (0.1% per inference). Data contributors are compensated through dynamic valuation.
V(D) = α·ΔP(D) + β·R(D) + γ·Dem(D)
Where ΔP is measurable performance improvement, R is relative rarity, and Dem is expressed demand. The coefficients α, β, γ are adjustable via governance.
| Parameter | Value |
|---|---|
| Total Supply | 1B $COLLAB (immutable cap) |
| Community Allocation | 40% |
| Investor Allocation | 20% (4yr vesting, 1yr cliff) |
| Team Allocation | 15% (5yr vesting, 18mo cliff) |
| Burn Rate | 5% of each inference fee |
| Emission Schedule | 8% year 1, -20%/yr, floor 1% |
The protocol employs a federated governance structure combining quadratic voting, specialized councils, and sub-DAOs. This design prevents capture by large token holders while maintaining decision-making efficiency.
Quadratic voting ensures that expressing strong preference for an outcome has increasing marginal cost, naturally limiting whale influence. The three specialized councils (Technical, Economic, Ethical) provide expert input on domain-specific decisions.
Our analysis shows that the unified three-pillar design creates network effects absent in single-pillar competitors. The symmetric design — where each pillar benefits from growth in the others — produces compounding returns as the network scales.
The hybrid verification system achieves 99.7% detection accuracy for malicious computations while adding only 12% overhead compared to optimistic-only approaches. The deflationary tokenomics model projects $75M–$500M market capitalization by year 5 based on 2–5% capture of the decentralized inference market.
Ankh AI introduces five fundamental innovations: hybrid verification, SBT-based reputation, programmable royalties, dynamic data valuation, and unified three-pillar coordination. By integrating compute, talent, and data into a single protocol with equitable incentives, Ankh AI creates a path toward truly decentralized artificial intelligence.
The protocol is designed to evolve through its governance mechanisms, ensuring that it can adapt to technological advances and changing market conditions while remaining true to its core principles of decentralization, transparency, and equitable value distribution.