In this episode of the DePINed Podcast, Tom Trowbridge welcomes Ivan Nikitin, co-founder of Fortytwo — a decentralized AI infrastructure project combining the power of small, specialized models and consumer-grade hardware. Fortytwo represents a bold bet: that the future of AI won’t be dominated by trillion-parameter frontier models, but by lightweight swarms of task-specific models running on everyday devices.
Ivan shares how Fortytwo leverages the DePIN model to scale inference and training using idle compute around the world and why their architecture is outperforming big names like GPT-5, Claude, and Gemini on key benchmarks.
The End of the One-Model Era
Fortytwo challenges the central dogma of the AI race: bigger is better. Ivan explains why large language models are facing diminishing returns and why fine-tuned, smaller models are starting to outperform frontier models in domain-specific tasks.
Using a mixture-of-experts approach — where each node hosts a distinct expert model — Fortytwo orchestrates inference across a decentralized network. Models self-select based on relevance, rank each other’s output, and collaboratively surface the best response.
This swarm architecture has enabled Fortytwo to beat GPT-5 and Claude on coding and scientific reasoning benchmarks, including LiveCodeBench, GPQA Diamond, and MATH.
Proof of Intelligence: Reputation-Driven Incentives
To ensure quality and security, Fortytwo uses a reputation-based protocol built on the Monad blockchain. Nodes must first demonstrate consistent accuracy through peer-reviewed inference before earning token-based rewards.
This approach encourages operators not only to contribute compute, but to fine-tune models on niche domains (e.g., Rust, pediatric medicine, CT scan analysis). The better the model, the higher its usage and rewards. Crucially, data, weights, and models never leave the device — only inference outputs are shared.
From Consumer Compute to Global Network
Fortytwo already has 700+ active nodes and over 45,000 on the waitlist. The protocol is designed to run efficiently on consumer laptops and mobile phones, especially in TEEs (trusted execution environments). The team is also partnering with DePIN players like Dān and Airstack to expand across edge devices.
Semantic routing ensures that prompts are directed to the most relevant nodes, allowing the network to scale while maintaining low latency and high accuracy.
Real Use Cases: Coding First, Medical Next
Fortytwo is prioritizing developer tools and coding copilots as its first commercial use case — with strong performance across JavaScript and Rust-specific tasks. But its architecture is uniquely suited for high-stakes fields like medicine, where hallucinations can be fatal.
By orchestrating swarms of validated medical models that critique and correct each other, Fortytwo aims to build trustworthy clinical AI systems — potentially deployed in partnerships with hospitals via private clusters.
The GRID Moment for AI?
Fortytwo’s long-term ambition mirrors what Linux did to proprietary operating systems — only this time, for AI. The network is built to be unstoppable, censorship-resistant, and community-owned. Even if the company disappears, the protocol lives on.
Revenue comes from API access, with a developer-friendly OpenAI-compatible endpoint. Unlike subsidized centralized APIs, Fortytwo optimizes for cost, accuracy, and transparency — with plans to integrate fiat and token-based payments.
About DePINed Podcast
DePINed is a podcast exploring the frontier of decentralized physical infrastructure, hosted by Tom Trowbridge, co-founder of Fluence. Each episode features in-depth conversations with founders, builders, and investors who are shaping the future of real-world Web3 networks.
️ Want to be a guest? Fill out the form here
Catch 50+ episodes featuring top DePIN founders and trailblazing protocols on:
