!

Introduction

By partnering with EigenLayer, Cartesi is the first and only blockchain technology stack capable of fully trust-minimizing large language model inference on-chain, with DeepSeek as a pilot case.

This marks a pivotal leap forward in the convergence of cryptography and AI, ushering in a new era in which blockchain can not only support AI, but also actively drive AI in a fully deterministic way.

Background

Traditional AI relies on centralized servers and opaque “black box” systems that users trust their output without being able to verify the underlying processes. This reliance on centralized infrastructure comes with significant pitfalls: limited transparency, a single point of failure that is vulnerable to censorship or tampering, and ultimately the inevitable reliance on trust.

Many projects claim to offer “decentralized AI” but still rely on proprietary models, centralized infrastructure, and trust-based systems. As a result, most so-called decentralized AI practices are still constrained by traditional frameworks that only use blockchain to record AI output, rather than using it as an active decentralized computing environment.

True decentralized AI must guarantee that the same input always produces the same output, and that anyone can verify it in a trustless environment. This paradigm requires provable certainty that the AI computation itself runs on-chain, rather than just logging the results to the ledger.

Cartesi’s strengths in AI

Cartesi is currently the only project that enables true blockchain-native AI, with a unique feature that smart contracts can directly perform AI calculations and ensure full transparency and verifiability.

Unlike traditional solutions such as Trusted Execution Environments (TEEs), which enhance security but still have a single point of failure, Cartesi provides a verifiable decentralized execution environment that enables fully on-chain AI inference.

Cartesi’s RISC-V virtual machine is uniquely positioned to enable developers to seamlessly run powerful AI libraries such as TensorFlow, PyTorch, and Llama.cpp in a decentralized environment, pushing the traditional limitations of blockchain in AI computing.

Cartesi’s partnership with DeepSeek

During the recent Cartesi x Eigen Experiment Week, Cartesi’s core team of engineers successfully built a prototype system that enables smart contracts to perform verifiable large language model inference. As a result, Cartesi became the first project in the crypto space to achieve a fully on-chain operation of DeepSeek. The prototype system also supports the operation of other open source LLMs such as Qwen, Smol, etc.

Inference is performed in Cartesi’s verifiable environment for maximum security by eliminating centralized intermediaries, guaranteeing the integrity of AI output and ensuring complete trustlessness. To date, Cartesi remains the only technology stack that supports fully trust-minimized LLM inference on-chain, and DeepSeek has proven its potential as a pilot case.

Future outlook

The impact of decentralized AI goes far beyond the technological breakthrough itself. The advent of truly deterministic AI will open up unprecedented possibilities in areas such as autonomous treasury management, intelligent DeFi systems, and fully autonomous governance, reshaping the decentralized ecosystem.

About Cartesi

is a powerful modular blockchain protocol that provides developers with a complete Linux environment and high-performance rollup technology designed to support the next generation of decentralized applications. By integrating Linux, the Cartesi Virtual Machine enables developers to build dApps using decades of battle-tested programming languages, tools, and codebases. Cartesi provides a separate rollup layer and dedicated compute resources for each dApp, significantly improving computational scalability while ensuring decentralization, security, and censorship resistance.