The Bittensor AI project attempts to utilize web3 token mining to make open-source AI development more sustainable, verifiable, and efficient.
Introduction
AI development has achieved tremendous breakthroughs in recent years due to advancements in data, computing power, and algorithm research, particularly with the arrival of OpenAI GPT-4, representing the foundational LLM model and driving productivity enhancement and societal efficiency transformation.
However, the disadvantages of centralized models, represented by GPT-4, have also become apparent. Centralized models typically face limitations on third-party integrations, impacting the scalability and interoperability of AI agents based on centralized models.
Therefore, open-source large models like Llama and others have received increasing attention from 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 incentive for most contributors. Even with some competition rewards, these are usually one-time, and subsequent improvement and development work still requires passion unless a large community of followers emerges, providing more opportunities for income and encouraging more contributors to continue improvements.
Thus, the Bittensor AI project attempts to utilize web3 token mining to make open-source AI development more sustainable, verifiable, and efficient. Through Yuma Consensus, resources and research parties (miners), validators, and AI project parties (subnet creators) are aligned, making AI research more transparent and decentralized, allowing anyone to participate in AI contributions and earn deserved rewards.
The performance of tokens in the secondary market confirms people's expectations, with prices rising from over $50 in September 2023 to over $500 in December 2024, achieving a tenfold increase!
Recently, the investor behind Bittensor and the founder of Digital Currency Group established an accelerator called Yuma specifically to incubate subnet projects within the Bittensor ecosystem and serves as CEO, demonstrating confidence in the potential of the Bittensor project.
Source: Coindesk
Of course, the success of any project cannot avoid skepticism. Since Bittensor's inception, there has been a lot of FUD. In this article, we summarize many unanswered questions and attempt to understand Bittensor's future positioning and possibilities in the decentralized AI track 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 one of its subnets.
What is a Subnet?
The subnet constitutes the core of the Bittensor ecosystem. Each subnet of Bittensor is an independent incentive-based competitive market. Anyone can create a subnet, customize the tasks that the subnet will perform, and design an incentive mechanism (in machine learning analogy, the incentive mechanism can be understood as the target loss function that can guide the model training towards the ideal outcome). You only need to pay a registration fee (priced in TAO) to 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 rather delegates the rights to operate those tasks to others.
Operating the tasks of 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 the registration fee priced in TAO (if a validator, they also need to stake TAO), one only needs to provide a computer with sufficient computing resources and register that computer and their wallet in a subnet, while running the miner module or validator module provided by the subnet creator on that computer (both modules are Python code in the Bittensor API).
How does the competitive market of subnets work?
The operation of subnet competition works as follows: assume you decide to become a subnet miner. The subnet validators will assign some tasks for you to complete. Other miners in the subnet will also receive the same type of tasks. After all subnet miners complete their tasks, they will submit the results to the subnet validators.
Subsequently, subnet validators will evaluate and rank the quality of tasks submitted by subnet miners. As a subnet miner, you will receive rewards (priced in TAO) based on the quality of work. Similarly, other subnet miners will also receive corresponding rewards based on their performance. At the same time, subnet validators will receive rewards for ensuring that high-quality subnet miners receive better rewards, thereby promoting the continuous improvement of the overall quality of the subnet. All these competitive processes are automatically completed according to the incentive mechanism written in the code 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 continually improve the required tasks in competition.
Conversely, poorly designed incentive mechanisms may lead to exploits and shortcuts, adversely affecting the overall quality of the subnet and hindering the motivation of fair miners.
The specific tasks of each subnet miner depend on the purpose for which the subnet creator initially established that subnet, which can be quite variable or specific. For example, the miner task of subnet 1 is to respond to text prompts sent by subnet validators and provide the best completion results, or the task of miners in subnet 47 is to provide storage.
Each subnet also has its unique research and commercialization directions, such as attempting to tackle technical challenges in a specific AI field, such as decentralized AI training, verifiable reasoning, or providing essential infrastructure and resources needed by AI, such as GPU trading markets or data labeling services, or helping users identify AIGC deepfake technologies, such as Subnet 34 - BitMind.
Currently, Bittensor has over 55 subnets, and this number continues to grow!
Source: IOSG Ventures
The role of Subtensor blockchain
Clearly, the blockchain and the project's token TAO play a significant role in this series of competitions.
First, the Subtensor blockchain records all key activities of the 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 work quality of all subnet miners, and the rankings of all subnet validators will collectively be submitted as input to the YC algorithm. Generally, the rankings from different subnet validators will arrive at Subtensor at different times, but the YC algorithm on Subtensor waits for all rankings to arrive. Typically every 12 seconds, the YC algorithm calculates rewards based on the input rankings of all validators. These rewards (priced in TAO) will be deposited into the wallets of subnet miners and subnet validators. The Subtensor blockchain continuously runs 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 the subnet miners, indicating the ranking of subnet miners based on the historical performance of that subnet validator. The second factor is the staking amount of each validator and miner. The Yuma consensus on the chain will use this weight vector and staking amount to calculate rewards and distribute them among subnet validators and subnet miners.
The Bittensor API will play the role of transmitting and connecting the opinions of validators on the subnet and the Yuma consensus on the Subtensor blockchain. Additionally, validators within the same subnet will only connect to miners within the same subnet, while validators and miners from different subnets will not communicate or connect with each other.
Source: Bittenso
The 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 chosen subnet to obtain a UID slot in that subnet, which represents the verification rights for that subnet. Note that subnet validators can hold multiple UID slots simultaneously and validate for multiple subnets, but do not need to increase the staking amount; staking one amount of TAO allows selection of multiple UID slots for validating multiple subnets (similar to the concept of restaking).
Therefore, to maximize rewards, staking validators tend to choose to provide verification services for all subnets. However, not all staked validators have the right to actually provide staking services. Only the top 64 validators ranked by staking amount in a subnet are considered to have true verification rights for that subnet. This reduces the risk of validator misconduct, as the staking amount becomes a high threshold and increases the cost of misconduct (you must have at least 1000 TAO to set weights in the subnet). Each validator attempts to build a good reputation and performance record to attract more TAO for delegated staking to increase their staking amount and become one of the top 64 validators in that subnet.
Once subnet validators and subnet miners (operating miners 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, with halving every 4 years.
Bittensor started with a fair launch approach, without pre-mined TAO tokens or ICO. Currently, the network generates 7,200 TAO daily, with each block generating 1 TAO 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 whereby once half of the total supply is distributed, the issuance rate will halve. This halving occurs approximately every 4 years and continues at every half boundary of the remaining tokens until all 21 million TAO tokens are in circulation.
Although TAO adopts the issuance curve and philosophy of Bitcoin, due to its redemption mechanism, this curve is actively dynamic and not as fixed as Bitcoin.
Redemption mechanism:
The daily token issuance volume for the current cycle is 7,200 TAO (the same issuance volume as Bitcoin from January 2009 to November 2012 in its first cycle).
However, a certain number of dynamic TAO are redeemed through key (re)registrations every day.
To become a miner or validator, one must register a key in the network and meet other GPU and computing power requirements. Registration requires the redemption of TAO, which means paying a certain amount of TAO to reinvest in the network.
Every key (re)registration will remove that TAO from circulation supply and reintroduce it into the protocol's issuance pool, where it can be mined again in the future.
This mechanism has delayed the planned halving time from 4 years, as the redeemed TAO is dynamic. When more keys are (re)registered, the redemption cost of TAO increases, or other subnets are released, the amount of redeemed TAO may increase significantly.
Moreover, registration is not only applicable to newcomers but also to users who have been deregistered for the following reasons:
For miners, their models and reasoning may not be competitive enough among other miners;
For validators, they fail to continuously set the correct weights, maintain issuance, or do not have enough TAO in their keys (self-delegated + other delegators' shares).
These factors themselves will also exacerbate the growth of registration demand.
The amount of redeemed TAO = the total number of registered (or re-registered) keys across all subnets * average registration (or re-registration) cost.
Thus, the originally planned first halving after 4 years may be postponed to 5 or 6 years, or even longer. This entirely depends on the balance between TAO issuance and redemption.
The Bittensor network was launched on January 3, 2021. According to token redemption data from taostats, the planned halving date is expected to be delayed until 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 in decentralized networks. Unlike traditional resource allocation determined by manual voting by validators, dTAO introduces a market-based dynamic adjustment mechanism that directly links resource allocation to the performance of subnet networks, optimizing the fairness and efficiency of reward distribution.
Core mechanism
Market-based dynamic resource allocation
The dynamic allocation mechanism of TAO 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. This distribution ratio will dynamically adjust with changes in market information, ensuring that resources flow to efficient and promising subnets.
Embedded liquidity pool design
Each subnet is configured with a liquidity pool consisting of TAO and subnet tokens (subnet/TAO token pair). Users can exchange TAO for subnet tokens by staking them 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 the team must gradually acquire token shares through long-term contributions and construction. This mechanism avoids the risk of rapid token sell-offs while encouraging the team to focus on technical improvements and ecosystem building.
Balance of roles between users and validators
The dynamic allocation of TAO resources is determined not only by the market but also influenced by both validators and users. Validators perform strict evaluations of the team's technical capabilities, market potential, and actual performance like venture capitalists (VCs). Meanwhile, users further drive the market value formation of the subnet through staking TAO and participating in market transactions.
Economic model analysis
Current funding support
Data shows that the average daily rewards for subnets in the network are about $47,000, corresponding to an annual support of about $17 million. This funding scale is far higher than 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
Currently, Bittensor's annual budget is expected to reach $1.3 billion, comparable to centralized AI research institutions like OpenAI and Anthropic. With the introduction of dynamic TAO, the future new issuance of TAO will mainly 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 links the issuance amount to market performance, greatly incentivizing teams to continuously improve their technologies and applications. This mechanism also curbs short-term behaviors of quickly cashing out through OTC trades, laying a foundation for the long-term sustainable development of the network.
Impact and significance
Resource allocation optimization
dTAO dynamically adjusts resource allocation through the market, ensuring that subnets with high utility and growth potential receive more resources. This mechanism not only improves the overall efficiency of the network but also promotes competition and innovation.
Decentralized AI ecosystem construction
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 promote the development of the decentralized AI ecosystem.
Incentives for ecosystem participants
Dynamic TAO balances the interests of users, validators, and the team, ensuring that all participants can contribute to the growth of the network through economic incentives.
Enhancement of validator roles
Validators need to play a more important role in the network. They rigorously evaluate the value and potential of subnets in a venture capital-like manner to ensure the scientific and rational distribution of network resources.
The introduction of dTAO marks a significant advance in decentralized network resource allocation mechanisms. Through market-driven dynamic adjustments, embedded liquidity pool designs, and fair issuance models, dTAO achieves efficiency and fairness in resource allocation. Additionally, as an AI network incubation platform, it not only empowers the development of subnets but also provides new pathways for the future of decentralized AI networks.
Agents applications on Bittensor
Many people say Bittensor is the AI coin represented by the VC elite, which has fallen behind the flourishing application era of various agent developer frameworks. With the recent rise of AI Agents and the total market value of AI Agent-related tokens exceeding $10 billion, especially projects represented by the Virtuals ecosystem dominating with a market cap of $5 billion (including various utility investments and research analysis-type Agents, such as $AIXBT, $VADER, $SEKOIA, etc.), Bittensor seems to be lagging behind in the eyes of many.
However, in reality, Bittensor still has many 'Alphas.' What many people do not realize is that the success of Virtuals/ai16z in the consumer AI Agent domain is complementary to the efforts of Bittensor subnets in decentralized AI infrastructure.
As the TVL (Total Value Locked) of Agents expands, strong training and reasoning infrastructure becomes increasingly important.
Currently, Virtuals have collaborated extensively with Bittensor in the ecosystem.
Many consumer-facing virtuals protocol agents are supported by Bittensor subnets, creating new possibilities using the computing power and data ecosystem of TAO.
$TAOCAT
TAOCAT is an AI agent created by Masa within the Virtuals ecosystem, primarily serving as a staunch defender of TAO and actively participating in discussions on X to voice 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 of Bittensor subnet 59, creating a new paradigm for tokenized AI value capture, where any user 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 set to launch soon.
$SERAPH: The first project focused on verification infrastructure, aimed at certifying the upcoming wave of AI Agents sweeping our digital world.
The collaboration between Virtuals and Bittensor proves that enormous practical value can be created on the Bittensor infrastructure. With the official launch of AgenTAO (SN62), this will become an important milestone for automated software engineering agents on Bittensor, and all Bittensor subnets will gradually be developed by Agents on Bittensor. In the future, we will see more application-side AI Agents emerging from the Bittensor ecosystem!
Source: taogod
Conclusion
The future of Bittensor is exciting, with many research and investment institutions beginning to emerge focusing on the Bittensor ecosystem, 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 and is also a subnet research organization. A network of connections related to Bittensor, akin to the PayPal mafia, has formed with gatherings recently involving Contango, Canonical, Delphi Labs, and DCG, where many experts from the Crypto x AI field have begun to align with and support Bittensor. Therefore, it is not without reason that Bittensor recently surpassed Virtuals in Kaito's mindshare.
Source: BitMind Bittensor Subnet 34
In April 2025, Bittensor will host The Endgame Summit and hackathon in Austin, Texas, focused on introducing more subnets, validators, and miners to the Bittensor ecosystem and expanding its territory.
Endgame Summit
Of course, whether centralized AI projects 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 major 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 will emerge from Bittensor in the future, becoming the preferred place for people to seek solutions to specific AI problems!
