Reshaping AI Development: How Bittensor Promotes Sustainable Development of Open Source and Decentralized AI
The Bittensor AI project attempts to utilize web3 token mining to make open-source AI development more sustainable, verifiable, and efficient.Author: IOSG Ventures
Introduction
AI development has made tremendous breakthroughs in recent years due to advancements in data, computing power, and algorithm research, especially with the emergence of OpenAI GPT-4, which represents the arrival of foundational LLM large models, driving productivity improvements and transforming social efficiency.
However, the drawbacks of closed-source large models represented by GPT-4 have also become apparent, namely, centralized models often have limitations on third-party integrations, which undermine the scalability and interoperability of AI agents based on centralized models.
As a result, open-source large models like the Llama series have gained increasing popularity among researchers, but open source does not equate to transparency, and it also faces many challenges.
The main dilemma is that open-source AI development offers no economic incentives for most contributors. Even though some competition rewards exist, they are usually one-time, and subsequent improvement and development work still require passion, unless a large community of followers is built after reaching a certain scale, which could lead to more revenue opportunities and more contributors continuing to improve.
Therefore, the AI project Bittensor attempts to utilize web3 token mining to make open-source AI development more sustainable, verifiable, and efficient. Through Yuma Consensus, it aims to align resources with research parties (Miners), validators (Validators), and AI project parties (Subnet Creators), making the entire AI research process more transparent and decentralized, allowing anyone to contribute to AI and earn deserved rewards.
The performance of tokens in the secondary market also confirms people's expectations, with prices rising from over $50 in September 2023 to over $500 in December 2024, achieving a tenfold increase!
Recently, Bittensor's investor and founder of Digital Currency Group established an accelerator named Yuma, specifically to incubate subnet projects within the Bittensor ecosystem, and serves as CEO, demonstrating his confidence and potential in the Bittensor project.
Source: Coindesk
Of course, the success of any project cannot be achieved without facing skepticism. Since the inception of Bittensor, there has been a lot of FUD. In this article, we summarize many unanswered questions and attempt to understand Bittensor's future positioning and potential in the decentralized AI space through research and analysis.
What is Bittensor?
Bittensor was founded in 2021 by a team from Toronto, Canada, including Jacob Robert Steeves, Ala Shaabana, and Garrett Oetken.
Bittensor is a decentralized AI infrastructure used by AI developers to build and deploy machine learning models or other AI-related developments. Many Web3 AI projects, regardless of whether they have their own blockchain, can connect to Bittensor's blockchain "subtensor" and become part of a subnet.
What is a Subnet?
Subnets form the core of the Bittensor ecosystem, with each subnet being an independent incentive-based competitive market. Anyone can create a subnet, customize the tasks it will perform, and design incentive mechanisms (in machine learning terms, the incentive mechanism can be understood as the target loss function, guiding model training towards ideal outcomes). By paying a registration fee (priced in TAO), one can create a subnet and receive a netuid for that subnet. Note that a subnet creator does not need to undertake the operational tasks within the subnet but can delegate the rights to operate those tasks to others.
Operating tasks within the subnet provides another way for others to participate, namely by joining an existing subnet. If joining an existing subnet, there are two ways to participate: as a subnet miner or a subnet validator. Besides paying a registration fee (priced in TAO, and validators also need to stake TAO), one only needs to provide a computer with sufficient computing resources and register that computer and their wallet to a subnet, while running the subnet creator's provided miner module or validator module (both modules are Python code within the Bittensor API).
How Does the Competitive Market of Subnets Work?
The operation of subnet competition works as follows: suppose you decide to become a subnet miner. Subnet validators will assign tasks for you to complete. Other miners in the subnet will also receive the same type of tasks. Once all subnet miners complete their tasks, they submit the results to the subnet validators.
Subsequently, subnet validators will assess and rank the quality of the tasks submitted by subnet miners. As a subnet miner, you will receive rewards (priced in TAO) based on the quality of your work. Similarly, other subnet miners will also receive corresponding rewards based on their performance. At the same time, subnet validators will also receive rewards for ensuring that high-quality subnet miners receive better rewards, thus driving the continuous improvement of the overall quality of the subnet. All these competitive processes are automated based on the incentive mechanisms coded by the subnet creator.
Source: Steps on how Subnet Creator defines Incentive Mechanism
The incentive mechanism ultimately judges the performance of subnet miners. When the incentive mechanism is well-calibrated, it can create a virtuous cycle where subnet miners continuously improve the tasks required in competition.
Conversely, poorly designed incentive mechanisms may lead to exploitation and shortcuts, adversely affecting the overall quality of the subnet and hindering the motivation of fair miners.
The specific work of each subnet miner depends on the original purpose for which the subnet creator established the subnet, which can be quite variable or specific. For example, the task of miners in subnet 1 is to respond to text prompts sent by subnet validators and provide the best completion results, while the task of miners in subnet 47 may be to provide storage.
Each subnet also has its unique research and commercialization direction, such as attempting to tackle technical challenges in a specific AI field, like decentralized AI training, verifiable inference, or providing essential infrastructure and resources for AI, such as GPU trading markets or data labeling services, or helping users identify AIGC deepfake technologies, like Subnet 34 - BitMind.
Currently, Bittensor has over 55 subnets, and this number continues to grow!
Source: IOSG Ventures
The Role of the Subtensor Blockchain
Clearly, the blockchain and project token TAO play a significant role in this series of competitions.
First, the Subtensor blockchain records all key activities of subnets in its ledger. More importantly, the Subtensor blockchain is responsible for determining the reward distribution for subnet miners and subnet validators. An algorithm called Yuma Consensus (YC) continuously runs on the Subtensor blockchain. Each subnet validator ranks the quality of work performed by all subnet miners, and all rankings from each subnet validator will be collectively sent as input to the YC algorithm. Generally, rankings from different subnet validators will arrive at Subtensor at different times, but the YC algorithm on Subtensor will wait for all rankings to arrive, typically every 12 seconds, to calculate rewards based on the input from all validators' rankings. These rewards (priced in TAO) will be deposited into the wallets of subnet miners and subnet validators. The Subtensor blockchain will continuously run the YC algorithm independently for each subnet.
The YC consensus algorithm mainly considers two factors: the first is a weight vector maintained by each subnet validator, where each element of the vector represents the weight assigned to subnet miners based on their historical performance, allowing each subnet validator to rank all subnet miners through this weight vector. The second factor is the amount staked by each validator and miner. The Yuma consensus on the chain will use this weight vector and stake amount to calculate rewards and distribute them among subnet validators and subnet miners.
The Bittensor API serves to transmit and connect the opinions of validators within the subnet and the Yuma consensus on the Subtensor blockchain. Additionally, validators within the same subnet will only connect to miners within that subnet, and validators and miners from different subnets will not communicate or connect with each other.
Source: Bittenso
Game Theory of Validators
To participate as a subnet validator or subnet miner, one must first register and stake. Registration means registering a key in the desired subnet to obtain a UID slot within that subnet, representing the right to validate that subnet. Note that subnet validators can simultaneously hold multiple UID slots and validate multiple subnets, but do not need to increase their stake amount; staking once allows them to choose multiple UID slots for validating multiple subnets (similar to the concept of restaking).
Therefore, to obtain the most rewards, staking validators will tend to choose to provide validation services for all subnets. However, not all staked validators have the right to genuinely provide staking services. Only the top 64 validators ranked by stake amount in a subnet are considered to have true validation permission for that subnet. This reduces the risk of malicious behavior by validators, as the stake amount becomes a high barrier and increases the cost of wrongdoing (at least 1000 TAO is required to set weights in a subnet). Each validator will try to build a good reputation and performance record to attract more TAO for delegated staking to increase their stake amount and become one of the top 64 validators in that subnet.
Once subnet validators and subnet miners (who do not need to stake) register their keys to the subnet, they can begin mining.
Unique Token Incentive Economy
All TAO token rewards are newly minted, similar to Bitcoin. Bittensor's $TAO has the same token economics and issuance curve as Bitcoin. TAO Supply: The total cap is 21 million, halving every 4 years.
Bittensor started with a fair launch, with no pre-mined TAO tokens or ICO. Currently, the network generates 7,200 TAO daily, with 1 TAO generated per block, approximately every 12 seconds. The total supply cap is set at 21 million, following a programmatic issuance plan similar to Bitcoin.
However, Bittensor introduces a unique mechanism where once half of the total supply is distributed, the issuance rate will be halved. This halving occurs approximately every 4 years and continues at each half boundary of the remaining tokens until all 21 million TAO tokens are in circulation.
Although TAO adopts Bitcoin's issuance curve and philosophy, due to its recycling mechanism, this curve is dynamically positive and not completely fixed like Bitcoin.
Recycling Mechanism:
The current daily token issuance for the cycle is 7,200 TAO (the same as Bitcoin's issuance during its first cycle from January 2009 to November 2012).
However, a certain number of dynamic TAO are recycled daily through key (re)registration.
To become a miner or validator, one must register a key in the network and meet other GPU and computing power requirements. Registration requires recycling TAO, meaning paying a certain amount of TAO to reinvest in the network.
Each time a key is (re)registered, that TAO is removed from the circulating supply and placed back into the protocol's issuance pool, where it can be mined again in the future.
This mechanism delays the planned 4-year halving time, as the recycled TAO is dynamic. When more keys are (re)registered, the cost of recycling TAO increases, or other subnets are launched, the recycled TAO may significantly increase.
Moreover, registration applies not only to newcomers but also to users who have been deregistered for the following reasons:
For miners, their models and inferences are not competitive enough among other miners;
For validators, they fail to continuously set correct weights, maintain issuance, or do not have enough TAO in their keys (self-delegated + shares from other delegators).
These factors themselves will also exacerbate the growth in registration demand.
The number of recycled TAO = Total number of registered (or re-registered) keys across subnets * Average registration (or re-registration) cost.
Therefore, the first halving originally planned to occur 4 years after launch may be delayed to 5 or 6 years, or even longer. This entirely depends on the balance between the issuance and recycling of TAO.
The Bittensor network went live on January 3, 2021. According to token recycling data from taostats, the planned halving date is expected to be delayed to November 2025.
Source: https://taostats.io/tokenomics
What is dTAO?
dTAO is an innovative incentive mechanism proposed by the Opentensor/Bittensor network, aimed at addressing the inefficiencies in resource allocation within decentralized networks. Unlike traditional methods where validators manually vote on resource allocation, dTAO introduces a market-based dynamic adjustment mechanism that directly links resource allocation to subnet network performance, thereby optimizing the fairness and efficiency of reward distribution.
Core Mechanisms
Market-Based Dynamic Resource Allocation
The dynamic TAO allocation mechanism is based on the market performance of subnet tokens. Each subnet in the network has its own independent token, and its relative price determines the distribution ratio of TAO issuance among subnets. As market information changes, this distribution ratio is dynamically adjusted to ensure resources flow to efficient and high-potential subnets.Embedded Liquidity Pool Design
Each subnet is configured with a liquidity pool composed of TAO and subnet tokens (subnet/TAO token pair). Users can exchange TAO for subnet tokens by staking TAO in the liquidity pool. This design incentivizes users to invest in high-performing subnets and indirectly supports the overall development of the network.Fair Token Distribution Mechanism
Subnet tokens are gradually distributed through a "fair launch" model, ensuring that teams need to earn their token shares through long-term contributions and building. This mechanism avoids the risk of tokens being quickly sold off while encouraging teams to focus on technological improvements and ecosystem building.Balancing Roles of Users and Validators
The dynamic TAO resource allocation is influenced not only by the market but also by the joint impact of validators and users. Validators must evaluate the technical capabilities, market potential, and actual performance of teams rigorously, similar to venture capitalists (VCs). Meanwhile, users drive the market value of subnets further by staking TAO and participating in market transactions.
Economic Model Analysis
Current Funding Support
Data shows that currently, subnets in the network can average about $47,000 in rewards daily, corresponding to an annual support of about $17 million. This funding scale far exceeds the median seed round (about $3 million) and Series A financing (about $14 million) for traditional AI startups, providing strong support for the rapid development of subnets.Future Potential
The current annual budget for Bittensor is expected to reach $1.3 billion, comparable to centralized AI research institutions like OpenAI and Anthropic. With the introduction of dynamic TAO, future new issuances of TAO will primarily flow into the liquidity pools of subnet tokens, further promoting the circulation of capital and value within the ecosystem.Long-Term Incentives
The design of dTAO greatly incentivizes teams to continuously improve their technology and applications by linking issuance to market performance. This mechanism also curbs short-term behaviors that seek quick cash-outs through over-the-counter (OTC) trading, laying the foundation for the long-term sustainable development of the network.
Impacts and Significance
Optimization of Resource Allocation
dTAO optimizes resource allocation through market-driven dynamic adjustments, ensuring that high-efficiency and high-growth potential subnets receive more resources. This mechanism not only improves the overall efficiency of the network but also fosters competition and innovation.Building a Decentralized AI Ecosystem
Bittensor is not only a decentralized AI network but also serves as an incubation platform for AI networks through dynamic TAO. The competition and collaboration among subnets further drive the development of a decentralized AI ecosystem.Incentives for Ecosystem Participants
Dynamic TAO balances the interests of users, validators, and teams, ensuring that all participants can contribute to the growth of the network through economic incentive mechanisms.Enhanced Role of Validators
Validators need to play a more significant role in the network. They must evaluate the value and potential of subnets rigorously, ensuring the scientific and rational allocation of network resources.
The launch of dTAO marks a significant advancement in the resource allocation mechanism of decentralized networks. Through market-based dynamic adjustments, embedded liquidity pool designs, and fair issuance models, dTAO achieves efficient and equitable resource distribution. Additionally, as an AI network incubation platform, it not only empowers the development of subnets but also provides new development paths for the future of decentralized AI networks.
Applications of Agents on Bittensor
Many people say that Bittensor is an AI token represented by VC elites, which has fallen behind the flourishing application era of various agent developer frameworks today. With the recent rise of AI Agents and the total market capitalization of AI Agent-related tokens surpassing $10 billion, especially projects represented by the Virtuals ecosystem dominating with a market cap of $5 billion (including various utility investment and research analysis Agents, such as $AIXBT, $VADER, $SEKOIA, etc.), Bittensor seems to many to be left behind.
However, in reality, Bittensor still possesses many "alphas." What many do not realize is that the success of Virtuals/ai16z in the consumer AI Agent space complements the efforts of Bittensor subnets in decentralized AI infrastructure.
As the TVL (Total Value Locked) and influence of Agents expand, the need for robust training and inference infrastructure becomes increasingly important.
Currently, Virtuals and Bittensor have collaborated on many ecological fronts.
Many consumer-facing virtuals protocol agents are supported by Bittensor subnets, leveraging TAO's computing power and data ecosystem to create new possibilities, such as
$TAOCAT
- TAOCAT is an AI agent created by Masa within the Virtuals ecosystem, primarily serving as a staunch defender of TAO, actively participating in discussions on X and voicing the influence of TAO.
- TAOCAT utilizes the real-time data infrastructure of subnet 42 Masa and the advanced LLM provided by Bittensor subnet 19 to compete for TAO token distribution in the Agent Arena on Bittensor subnet 59, creating a new paradigm for tokenized AI value capture, where any user's interaction on X will become training data for TAO Cat.
Other projects supported by Bittensor subnets include:
- $AION: The first Agent capable of predicting outcomes and participating in prediction market betting, with copy-trading functionality coming soon.
- $SERAPH: The first project focused on validating infrastructure, aimed at certifying the upcoming wave of AI Agents sweeping our digital world.
The collaboration between Virtuals and Bittensor demonstrates that significant practical value can be created on the Bittensor infrastructure. With the official launch of AgenTAO (SN62), this will mark an important milestone for automated software engineering Agents on Bittensor, with all Bittensor subnets gradually being developed by Agents on Bittensor. In the future, we will see more application-oriented AI Agents emerging from the Bittensor ecosystem!
Source: taogod
Conclusion
The future of Bittensor is exciting, with many research and investment institutions focusing on the Bittensor ecosystem beginning to emerge, similar to the Ethereum network. This includes the recent shout-out from the founder of DCG, podcasts, blogs, and OSS Capital, which is dedicated to researching the Bittensor ecosystem while also being a subnet research organization. A network of connections similar to the PayPal mafia has formed around Bittensor, with Contango, Canonical, Delphi Labs, and DCG recently holding a gathering where many experts from the Crypto x AI space began to converge on and support Bittensor. Therefore, it is not without reason that Bittensor was able to surpass Virtuals in Kaito's mindshare recently.
Source: BitMind Bittensor Subnet 34
In April 2025, Bittensor will host a The Endgame Summit conference and hackathon in Austin, Texas, with over 300 participants, specifically focusing on bringing more subnets, validators, and miners into the Bittensor ecosystem and expanding its territory.
Endgame Summit
Ultimately, whether centralized or decentralized AI projects, the final standard will return to the product itself. Currently, the Bittensor ecosystem has emerged and is flourishing.
Source: Outpost AI Research
Recently, the founder of Bittensor summarized the main achievements of various subnets over the past year on his personal X:
Therefore, let us continue to look forward to Bittensor and see what products and use cases can emerge from Bittensor in the future, becoming the preferred place for people seeking specific AI solutions!
