Rayon Labs: A subnet leader on Bittensor (TAO)?

This analysis is written by Victor.crypto, here is his X account.

Presentation of Rayon Labs

What is Rayon Labs?

Rayon Labs is a company developing Artificial Intelligence (AI) services on top of the Bittensor (TAO) infrastructure. The team currently operates three subnets: Chutes (64), Gradients (56), and Nineteen (19), which rank among the most capitalized in the protocol. Each of these subnets addresses a key use case: inference, fine-tuning, or high-speed inferencing.

Their ambition is to offer an open-source, decentralized, and higher-performing alternative to Web2 solutions, while maintaining a seamless user experience. Unlike many Web3 projects, Rayon Labs prioritized delivering real products over an aggressive marketing strategy. This approach allows them to show real traction on some services (which we will present later).

Rayon Labs also stands out through the quality of its integration with the Bittensor ecosystem. All their products are interconnected and powered by the protocol's resources, enabling an efficient value loop between end users, miners, and validators.

A few words about Bittensor

As a reminder, Bittensor is a decentralized protocol designed to transform the exchange and use of AI resources. It relies on a model of Subnets, specialized networks focused on one or more missions (inference, model fine-tuning, GPU computation, etc.), which function like startups.

→ Want to learn more about Bittensor (TAO)? Check out our full presentation:

The Bittensor ecosystem is structured around TAO, a token that powers staking, rewards, and governance. Since the Dynamic TAO update, each subnet has its own native token, creating a market where users can choose which subnet to delegate their TAO to.

Miners ensure the functioning of the subnets and the execution of tasks, while validators evaluate their performance. Now, the best subnets theoretically attract more TAO staking from users, thus receive more emissions and more resources to grow. This model creates a mechanism of continuous competition where only the best survive.

→ Everything you need to know about Dynamic TAO (dTAO), a major update for Bittensor:

An experienced technical team

Founded by two recognized figures in the Bittensor ecosystem, Namoray and BonOliver (a French co-founder), Rayon Labs is supported by about fifteen full-time collaborators. The team has demonstrated strong execution capabilities, having launched several functional products with refined UX and deep integrations (API, dashboards, GPU infrastructure) within just a few months.

Some figures are public, like Jon Durbin on subnet 64, while others prefer to remain anonymous. This strategy sometimes raises questions, but the quality of the deliverables and the speed of deployment largely mitigate any concerns.

Today, Rayon Labs positions itself as the undisputed leader of Bittensor in terms of product quality, user adoption, and technical expertise. In the next sections, we will present the main products developed by Rayon Labs.

---

Chutes: Subnet 64

General presentation

Chutes is Bittensor’s subnet 64 and Rayon Labs’ flagship solution. It is an open-source inference service that enables large-scale deployment and execution of AI models.

The product is based on two fundamental elements: an API to allow developers to access various AI models, and computing power to run these models, delivered by its own network of miners (composed of GPUs).

Concretely, Chutes aims to offer a decentralized alternative to major Web2 services like OpenAI’s API, while guaranteeing better accessibility, greater model diversity, and superior performance.

Chutes also stands out for its very responsive integration of the latest innovations in the AI sector. Indeed, they were the first to offer models like DeepSeek V3 and have recently been publicly praised by actors like OpenRouter for their execution speed.

Finally, the platform is designed to soon accommodate advanced features like Trusted Execution Environment (TEE), a privacy technology allowing queries to be executed without ever exposing user data to miners. This could open the door to numerous B2B use cases.

Adoption and key metrics

Since its launch at the end of January 2025, Chutes' adoption has been spectacular. The platform processes more than 30 billion tokens per day, amounting to several million daily requests, with a continuous growth curve.

Chutes is now one of the main inference providers on the OpenRouter platform, alongside giants like Anthropic and far ahead of players like Together AI or Nebius. Their free offering enabled them to quickly acquire a user base, and many projects continue using their models even after the start of monetization.

Currently, over 100,000 users use the Chutes API, meaning OpenRouter now accounts for only half of their traffic.

Business model and monetization

Previously, Chutes’ usage was entirely free, allowing for this rapid growth. Since April 2025, Chutes has begun monetizing some of its models (with payment in TAO or fiat), while aiming to always remain cheaper than the competition.

All revenues generated on Chutes are used in an auto-staking mechanism, consisting of buying back the subnet’s native token to redistribute it to the network. This system aligns the interests of users, infrastructure (miners/validators), and the token.

---

Squad: AI Agent Platform

Squad is the new platform developed by Rayon Labs that enables anyone to create, deploy, and interact with AI agents. Designed as a no-code interface inspired by Figma’s user experience, Squad positions itself as an accessible solution for all users, intuitively and without programming skills.

Entirely built on the Chutes infrastructure (subnet 64), Squad benefits from direct access to AI models deployed by Rayon Labs, as well as the computing power provided by their GPU miner network.

Squad allows the design of a custom AI agent by assembling three main components: simply choose a control model (the "brain" of the agent), define its system message (which guides its behavior), and connect it to different tools (web search, image generation, audio processing, etc.). Within minutes, it is possible to build agents capable of interacting autonomously with their environment.

On the business model side, Squad offers a free version to test creating a first agent with a limited number of tools, and a Pro plan at $40 per month to unlock all features. This approach aims to build a broad user base while generating recurring revenues for Rayon Labs.

---

Gradients: Subnet 56

General presentation

Gradients is Bittensor’s subnet 56, developed by Rayon Labs to address a need that was still under-served in the ecosystem: fine-tuning AI models (i.e., improving them to meet a specific use case). While Chutes enables easy use of open-source models, Gradients offers to customize them based on specific datasets or instructions.

Concretely, Gradients allows users (developers, researchers, or companies) to easily launch fine-tuning sessions without managing heavy infrastructure. All technical aspects, from orchestration to GPU resource management, are handled by the subnet 56 miners.

The platform supports several types of fine-tuning: classic supervision, reinforcement learning with human feedback (RLHF), and model alignment for more complex use cases like specialized agents. All of this is accessible serverless, either via a simple API or an intuitive graphical interface, in continuity with the experience offered by Chutes.

Gradients thus positions itself as a decentralized and cost-effective alternative to proprietary Web2 solutions, offering more flexibility regarding the data used and training methods.

Adoption and key metrics

Although launched more recently than Chutes, Gradients quickly attracted interest from many Bittensor ecosystem users. The ability to adapt open-source models at a low cost, combined with a simplified UX, allowed the subnet to experience rapid adoption.

Activity volumes are still modest compared to Chutes, given the greater complexity of fine-tuning workflows, but they are growing steadily. The subnet has already established itself as the main fine-tuning solution on Bittensor.

The quality of the infrastructure, capable of massively mobilizing GPUs to train models in a distributed way, is also recognized as one of Gradients' strong points by early users.

Business model and monetization

Gradients follows a business model very close to that of Chutes: free basic access to encourage adoption, then progressive billing based on consumed resources (GPU time, model size, type of fine-tuning), payable in TAO or fiat.

Usage costs are low: $5 per hour of training, which is cheaper than Together AI and far cheaper than major solutions from Amazon and Google. Moreover, the pricing does not come at the expense of performance since, across around a hundred comparative tests, Gradients won all of them (important note: this research was conducted by the Gradients team).

As with Rayon Labs' other products, collected revenues are used in an auto-staking mechanism to support the subnet’s native token. This approach aligns the incentives between users, miners, validators, and token holders.

---

Nineteen: Subnet 19

General presentation

Nineteen is Bittensor’s subnet 19, developed by Rayon Labs to address a specific need: high-frequency inference. While Chutes is optimized for general use, Nineteen targets applications requiring ultra-fast response times and low-latency performance.

The subnet has been specifically designed to handle intensive query flows while maintaining high response quality. To achieve this, Nineteen relies on a network of specialized GPU miners, capable of ensuring very high availability and rapid execution, even on large models.

This type of infrastructure is essential for use cases like real-time assistants, autonomous agents interacting live with their environment, or certain industrial applications where latency is critical.

In short, Nineteen offers pure performance on a few targeted models, while Chutes offers stability across a large number of models.

Adoption and key metrics

Launched in April 2025, Nineteen currently targets a more specific clientele than Chutes or Gradients but is already finding its market. Several teams developing complex AI agents on Bittensor (notably through Squad) have adopted Nineteen as their preferred backend to improve the responsiveness of their applications.

Although usage volumes are still lower than those of Chutes, the platform is experiencing stable growth driven by real technical needs, and it benefits from positive word of mouth among the most advanced AI developer circles.

Nineteen’s ability to maintain low response times despite progressive load increases is now considered one of its greatest achievements.

Moreover, recently, Mistral published a graph highlighting their performance and latency compared to OpenAI and Anthropic on a certain benchmark: Nineteen replicated the test and far exceeded Mistral’s performance.

Business model and monetization

Nineteen’s economic model, like the other Rayon Labs subnets, is based on partially free access to foster initial adoption. Soon, Nineteen will implement a pricing model based on usage (GPU time, number of requests, access priority).

The prices will likely be slightly higher than those of Chutes, consistent with the premium quality of service offered, but still aiming to remain very competitive compared to traditional Web2 solutions.

All revenues collected on Nineteen are also used to feed an auto-staking mechanism in favor of its native token, thus reinforcing the circular economy specific to each subnet operated by Rayon Labs.

---

The three tokens of Rayon Labs

With Bittensor’s Dynamic TAO update, each subnet can now issue its own native token. Rayon Labs has therefore launched three distinct tokens for its subnets, allowing users to directly stake on the subnet of their choice, exercise governance power, and capture part of the value generated by the network's activity.

As a reminder, Rayon Labs has implemented an auto-staking mechanism on its products: part of the revenues generated (for example via Chutes, Gradients, or Squad) is used to buy back the corresponding tokens on the secondary market and stake them continuously.

Today, Chutes’ token is by far the leader in terms of capitalization ($60 million), mainly because adoption is the highest and Squad is built on top of it. Gradients is also beginning to attract a user base thanks to its growing success in the fine-tuning sector, making it second ($30 million). Finally, Nineteen ranks third among Rayon Labs’ subnets ($15 million).

→ Want to understand how to invest in Bittensor’s subnets? Here’s our analysis:

---

Conclusion

Rayon Labs is now one of the most important players in the Bittensor ecosystem. Their approach — focused on quickly delivering tangible products and deeply integrating with the TAO infrastructure — has allowed them to take a clear lead over their competitors. Through their three subnets, Chutes, Gradients, and Nineteen, as well as Squad, Rayon Labs now covers essential use cases for open-source AI, from inference and fine-tuning to the orchestration of autonomous agents.

Among the main strengths of the project, we can highlight the technical quality of their infrastructure, their speed of execution, their ability to integrate the latest advances in the AI sector in real time, and an interesting economic strategy through auto-staking mechanisms. The diversification of their products also allows them to mitigate risks: even if one service slows down, other growth drivers can take over.

However, Rayon Labs faces several challenges. The inference and fine-tuning market is becoming increasingly competitive, especially with the rise of low-cost Web2 solutions. The team will need to keep innovating to maintain its lead while ensuring the scalability of its GPU infrastructure. Additionally, the dependency of their economic model on the Bittensor ecosystem remains a constraint: any unfavorable evolution of the protocol or the TAO price could impact their growth.

In short, Rayon Labs today has solid fundamentals to establish itself sustainably in decentralized AI. Their ability to execute quickly, their clear product vision, and their alignment of interests with users give them a rare competitive advantage in the Web3 ecosystem. Provided they maintain this level of excellence and anticipate the rapid evolution of the AI sector, Rayon Labs could become one of the first large-scale success stories in the decentralized AI infrastructure space.