Exclusive Q&A: Fortytwo CEO Ivan Nikitin on Why AI’s Future is Decentralized

Artificial intelligence is at a turning point. While today’s AI relies on massive, centralized models with soaring infrastructure costs, Fortytwo CEO Ivan Nikitin envisions a scalable, decentralized alternative.

Drawing from his early work in the AI industry, Nikitin saw the limits of conventional AI firsthand, leading to Fortytwo’s creation—a swarm-based AI model that leverages small, specialized models for greater efficiency, accuracy, and cost-effectiveness.

In this exclusive interview with BeInCrypto, Nikitin explains why centralized AI is unsustainable, how Fortytwo’s decentralized approach outperforms monolithic LLMs, and why its devnet launch on Monad’s testnet marks a breakthrough for AI scalability.

BeInCrypto: What led you to the AI field, and how did your journey shape the vision behind Fortytwo?

Ivan Nikitin: My journey into AI actually started in game development. I saw games as the perfect playground for self-learning AI agents, and that pursuit of evolving intelligence became my core mission. By 2006, while studying at DePaul University in Chicago, I assembled an indie game studio with a strong focus on AI research.

During that time, I met Vlad, now a PhD, who was then working on his first degree. Together, we developed our first AI-driven game tech—an evolving intelligence algorithm inspired by NEAT (NeuroEvolution of Augmenting Topologies). The moment we saw it working was pivotal. Almost a year later, Google DeepMind published their Agent57 research, demonstrating how their Reinforcement Learning approach outperformed human benchmarks on 57 Atari games—something we had already achieved with our own method. That was the moment we knew we were ahead of the curve, and the game was on.

Around the same time, the first Transformer-based language models emerged, like BERT in 2018. We expanded beyond game AI, diving into LLM-based conversational AI. But despite strong demand and promising tech demos, scaling AI remained a bottleneck. Even today, OpenAI’s top-tier partners face strict usage caps, and self-hosting high-load AI models is prohibitively expensive.

So, the real question became: Do we wait for Big Tech to build enough billion-dollar data centers, or do we rethink AI scalability? That’s how Fortytwo was born—a decentralized AI protocol designed to make scalable, cost-efficient inference a reality for developers like us.

BeInCrypto: The AI industry is at a crossroads, with centralized models demanding massive compute and capital. Fortytwo proposes a radically different approach—what convinced you that decentralized AI is the future, and how does it change the game?

Ivan Nikitin: A few years ago, compute scarcity wasn’t widely recognized outside the industry. New inference providers were emerging, pushing innovation across the tech stack. But then came clear signals—Sam Altman calling for $5–7 trillion in AI infrastructure, Europe estimating $1 trillion in power grid upgrades to support AI demand, and the U.S. recommissioning nuclear plants. All of this was happening while real AI applications were still in their early stages.

With increasing model capabilities, AI started moving beyond chat applications into new areas like coding copilots. Even in this niche, platforms like Cursor are already consuming more compute than Anthropic can provide. Reasoning models require even more inferences per request, further driving up costs and running into API limits.

At the same time, consumer devices are becoming more powerful, yet most of their compute remains underutilized. Local models offer an alternative but struggle to match the reasoning capabilities of larger centralized models. Instead of choosing between expensive proprietary AI and underpowered local models, Fortytwo takes a different path – connecting locally running models into a peer-to-peer global network where they reason together.

The shift toward decentralized AI is not just about efficiency – it’s about long-term viability. Centralized AI infrastructure is becoming an unsustainable arms race, requiring exponential investment with diminishing returns. AI calls for a scalable foundation that aligns with technological progress, not against it. By leveraging widely available compute rather than concentrating it in a few hands, we can ensure AI remains accessible, adaptable, and resilient.

BeInCrypto: Fortytwo’s swarm intelligence model claims to outperform monolithic LLMs on several key benchmarks while reducing costs. Can you break down how a network of small AI models achieves this efficiency without compromising accuracy?

Ivan Nikitin: The key to Fortytwo’s efficiency lies in moving away from the idea that bigger models are always better. Instead of relying on a single, massive LLM, we use a network of smaller, specialized models that collaborate. This not only makes inference cheaper but also improves accuracy in real-world applications.

Small Language Models (SLMs) consistently outperform generalist LLMs in specialized tasks like coding, math, law, and medicine. Rather than forcing one model to handle everything, Fortytwo intelligently routes queries to models optimized for specific domains. That alone boosts accuracy, but swarm inference takes it a step further.

Instead of relying on a single AI response, Fortytwo distributes tasks across multiple nodes, ranks the outputs, and compiles the most valuable and reliable ones into the final answer. Accuracy isn’t just a byproduct—it’s built into the process through dynamic peer validation.

And because this happens on a decentralized network rather than inside costly data centers, inference is dramatically more efficient—up to 35x cheaper than OpenAI o1. As more nodes join, the network scales organically, enabling entirely new AI use cases that demand high-volume API requests—whether it’s next-gen coding and other specialized copilots, conversational agents in games, or embedding LLM capabilities into IoT devices.

BeInCrypto: Your business model aims to disrupt the $25B+ AI inference market. What unique advantages does Fortytwo offer developers, enterprises, and node operators compared to traditional AI providers?

Ivan Nikitin: For developers, Fortytwo offers AI inference without the usual constraints – lower costs than centralized providers, no API rate limits, and continuously improving reasoning capabilities. As the network expands and models collaborate more effectively, its ability to handle complex tasks will advance, approaching frontier-level performance. Beyond efficiency, Fortytwo’s permissionless architecture ensures a more balanced approach to information, free from centralized control over what gets prioritized or filtered.

For node operators, running a Fortytwo node goes beyond simply contributing compute – it’s a chance to participate in building planetary-scale intelligence while optimizing for higher rewards. At the basic level, anyone can enhance their node’s knowledge by privately supplying it with PDFs or other documents. These resources remain local and are never shared with the network, but the model can factor in when reasoning and generating inferences. More advanced users – those with expertise in machine learning and data science – can integrate custom fine-tunes or private models, significantly increasing their earning potential. Crucially, these models and fine-tunes remain fully owned by the operator and are never openly exposed to the network.

BeInCrypto: Some argue AGI will require centralized supercomputers, but Fortytwo takes a different path. How does your decentralized swarm intelligence model contribute to AGI’s evolution, and what are the biggest challenges ahead?

Ivan Nikitin: The assumption that AGI needs a massive centralized supercomputer comes from the idea that intelligence is best achieved by a single, all-encompassing model. But intelligence isn’t just about raw processing power—it’s about how knowledge is structured, shared, and refined. Fortytwo takes a different approach: rather than concentrating everything into one system, it allows AI models with different strengths to work together, creating a more adaptable and evolving intelligence.

We like to think of Fortytwo as a “Wikipedia that can think.” Just as Wikipedia is maintained by contributors who refine and expand specific topics, Fortytwo grows through contributions from its community. Some provide compute, while others may develop and maintain highly specialized models. For example, a machine learning expert could fine-tune a model solely for Rust programming, keeping it up-to-date with the latest standards and best datasets. Every Rust-related query benefits from that expertise, and the maintainer earns rewards for improving the network’s collective knowledge in that area.

This collaborative AI approach allows capabilities to emerge dynamically rather than being pre-defined by a single architecture. Instead of relying on a one-size-fits-all model, Fortytwo’s structure ensures that intelligence expands organically as new contributions come in. And because it’s decentralized, it scales without limits – every new node brings additional knowledge, compute, and self-improvement to the network.

The biggest challenges ahead involve coordinating a global network of models, ensuring low-latency inference for real-time applications, and navigating the regulatory landscape of decentralized AI. But the core idea remains: intelligence isn’t just about having more power – it’s about structuring and distributing knowledge in a way that keeps improving over time.

BeInCrypto: Your devnet is launching on Monad’s testnet as part of the “Hitchhiking to AGI” program. How does Monad’s blockchain architecture enhance the scalability and trustless execution of your decentralized AI?

Ivan Nikitin: When we started working on Fortytwo, we didn’t immediately know which network to build on. After all, how could we talk about scaling AI if we were bottlenecked by network throughput?

Monad’s 10,000 TPS throughput and low fees provided the kind of performance needed for high-demand use cases like ours. While Fortytwo’s nodes operate as a peer-to-peer network, trustless coordination wouldn’t be possible without a high-speed blockchain to record node performance and maintain an incremental historical reputation system.

This is essential because Fortytwo is designed for open participation, with no centralized validation or whitelisting of AI models. We track peer evaluation outcomes on-chain to ensure network integrity without introducing noticeable latency. Monad enables this level of security and efficiency, making it a natural fit for our case of scaling AI through decentralization.

BeInCrypto: One of the biggest challenges with AI today is reliability—hallucinations, inaccuracies, and biases. How does Fortytwo’s decentralized swarm inference model address these issues better than centralized alternatives?

Ivan Nikitin: One of the fundamental problems with centralized AI is that it relies on a single model, meaning its biases, inaccuracies, and limitations are baked in. If the model gets something wrong, there’s no built-in way to cross-check its output. Fortytwo takes a different approach by distributing inference across multiple independent models and letting them validate each other.

Instead of relying on a single response, Fortytwo ranks multiple answers from different nodes and selects the best-validated one. This peer evaluation process naturally reduces hallucinations and incorrect outputs by ensuring that only the most reliable responses are used.

Another advantage is model diversity. A centralized model is locked into its training data, making it prone to specific biases. Fortytwo’s network, on the other hand, leverages a variety of fine-tuned models that specialize in different domains. The result is a more adaptive and balanced system that isn’t restricted by the blind spots of a single AI.

Finally, the network self-improves over time. Since nodes are rewarded based on accuracy, models that consistently produce low-quality responses lose influence, while the best-performing ones become more dominant. This creates an ongoing feedback loop where the system continuously optimizes itself, something centralized AI struggles to achieve.

BeInCrypto: Looking five years ahead, what does success look like for Fortytwo? Will decentralized AI become the norm, or do you see coexistence with centralized models?

Ivan Nikitin: Success means proving that decentralized AI isn’t just an alternative—it’s a better way to scale intelligence. By 2030, we expect a major shift toward permissionless, swarm-driven AI as developers, enterprises, and users recognize that relying on a few centralized providers isn’t sustainable. The growing limitations of centralized models will drive the industry in this direction, and Fortytwo will be part of that transformation.

Beyond adoption, real success means AI infrastructure owned and governed by its users. Instead of being controlled by a handful of corporations, intelligence should be shaped by those who contribute—whether by running compute, improving models, or fine-tuning domain-specific expertise. Fortytwo’s model enables organic, community-driven growth, where AI evolves based on real-world contributions.

Centralized models won’t disappear entirely, but they will become just one piece of a broader ecosystem. The real shift will come from AI that continuously refines itself, with millions of interconnected nodes reasoning collectively at a planetary scale. That’s where I see Fortytwo in five years—powering an open, self-sustaining intelligence network.

From this conversation with Ivan Nikitin, it is evident that AI is at a turning point. Scalability, cost, and reliability are forcing the industry to rethink its foundations. Whether decentralized AI becomes the dominant model or evolves alongside centralized systems, the shift is already happening. Fortytwo is proof of that, challenging long-held assumptions and showing that intelligence doesn’t need to be confined to massive data centers to be effective.