Preface
AI development has seen tremendous breakthroughs in recent years due to advancements in data, computational power, and algorithm research, especially with the emergence of OpenAI GPT-4, representing the arrival of foundational large models, enhancing productivity and transforming societal efficiency.
However, the drawbacks of closed-source large models, represented by GPT-4, have also become apparent. That is, centralized models often have restrictions on third-party integrations, which hampers the scalability and interoperability of AI agents based on centralized models.
As a result, open-source large models like Llama and others have become increasingly sought after by researchers. However, open source does not mean transparency; it also faces many challenges.
The main dilemma is that open-source AI development offers no economic incentives for most contributors. Even if some competition rewards arise, they are usually one-time, and subsequent improvement development work still requires working for love, unless a significant scale is reached with a large community of followers, making it more likely to realize income and attract more contributors to continue improvement.
Therefore, the Bittensor AI project seeks 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), validating parties (Validators), and AI project parties (Subnet Creators), making AI research 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, an increase of 10 times!
Recently, the investor of Bittensor and founder of Digital Currency Group established an accelerator called Yuma specifically incubating 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 without scrutiny. 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 possibilities in the decentralized AI arena through research analysis.
What is Bittensor?
Bittensor was founded in 2021 by a team from Toronto, Canada, consisting of 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 already have their own blockchain, can connect to Bittensor's blockchain 'subtensor' and become part of a subnet.
What is a Subnet?
Subnets constitute the core of the Bittensor ecosystem, with each subnet being an independent incentivized competitive market. Anyone can create a subnet, customize the tasks that subnet should perform, and design an incentive mechanism (analogous to a loss function in machine learning, guiding model training toward ideal outcomes). One only needs to pay a registration fee (priced in TAO) to create a subnet and obtain a netuid for that subnet. Note that a subnet creator does not need to undertake the operational tasks within the subnet; instead, they delegate the rights to operate those tasks to others.
Operating tasks for the subnet provide others with another way to participate, namely joining an existing subnet. If joining an existing subnet, there are two ways to participate: as a subnet miner or subnet validator. Apart from paying a registration fee priced in TAO (if a validator, also needs to stake TAO), one just needs to provide a computer with sufficient computational resources and register that computer and their wallet to a subnet while running the subnet miner module or subnet validator module provided by the subnet builder (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: Suppose you decide to become a subnet miner. Subnet validators will assign you tasks to complete. Other miners in the subnet will also receive the same type of tasks. After all subnet miners complete the tasks, they submit their results to the subnet validators.
Subsequently, subnet validators will each evaluate and rank the quality of tasks submitted by subnet miners. As a subnet miner, you will receive rewards (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 receive rewards for ensuring high-quality subnet miners receive better rewards, thus promoting the continuous improvement of the overall quality of the subnet. All these competitive processes are automated according to the incentive mechanism written 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 form a virtuous cycle, with subnet miners continuously improving the required tasks in competition.
On the contrary, poorly designed incentive mechanisms can lead to exploitation and shortcuts, adversely affecting the overall quality of the subnet and hindering the enthusiasm of honest miners.
The specific tasks of each subnet miner depend on the original purpose for establishing that subnet by the subnet builder, which can be relatively variable or more specific. For example, the miner's task in subnet 1 is to respond to text prompts sent by the subnet validator and provide the best prompt completion results, or the task of miners in subnet 47 is 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, such as training decentralized AI, verifiable inference, or providing infrastructure and resources needed by AI, such as a GPU trading market or data labeling services, or helping users identify AIGC deepfake technology in subnets, such as Subnet 34 - BitMind.
Currently, Bittensor has more than 55 subnets, and this number is still continuously increasing!
Source: IOSG Ventures
The role of the Subtensor blockchain
Clearly, the blockchain and the 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 work quality of 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, and 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 continuously runs the YC algorithm independently for each subnet.
The YC consensus algorithm primarily 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. This weight reflects the historical performance of the subnet miners based on that subnet validator's assessment. The second factor is the staking amount of each validator and miner. The Yuma consensus on-chain will use this weight vector and staking amounts to calculate rewards and distribute them among subnet validators and 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; 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 miner, one must first register and stake. Registration means registering a key in the desired subnet, obtaining a UID slot in that subnet, which represents the right to validate that subnet. Note that subnet validators can hold multiple UID slots simultaneously and validate for multiple subnets without needing to increase the staking amount; one staking of TAO can choose multiple UID slots to validate for multiple subnets (similar to the concept of restaking).
Therefore, to obtain the most reward, staking validators tend to choose to provide validation 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 validation permissions for that subnet. This reduces the risk of validators acting maliciously, as the staking amount becomes a high threshold and increases the cost of wrongdoing (must have at least 1000 TAO to set weights in the subnet). Each validator will try to build a good reputation and performance record to attract more delegated TAO staking to increase the staking amount, becoming 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: total limit of 21 million, halved every 4 years.
Bittensor started with a fair launch, without pre-mined TAO tokens or an ICO. Currently, the network generates 7,200 TAO daily, with each block generating 1 TAO approximately every 12 seconds. The total supply limit is set at 21 million, following a programmatic issuance plan similar to Bitcoin.
However, Bittensor introduces a unique mechanism where the issuance rate is halved once half of the total supply has been distributed. 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, this curve is actively dynamic due to its recovery mechanism and is not completely fixed like Bitcoin.
Recovery mechanism:
The daily token issuance for the current cycle is 7,200 TAO (the same as Bitcoin's issuance during its first cycle from January 2009 to November 2012).
However, a certain amount of dynamic TAO is recovered daily through key (re)registration.
To become a miner or validator, one must register a key in the network and meet other GPU and computational requirements. Registration requires recovering TAO, i.e., paying a certain amount of TAO to reinvest in the network.
Each key (re)registration will remove that TAO from the circulating 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 4-year halving time because the recovered TAO is dynamic. When more keys are (re)registered, the cost of TAO recovery increases, or other subnets are launched, the recovered TAO may increase significantly.
Moreover, registration is not only applicable to newcomers but also to those 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 failed to consistently set the correct weights, maintain issuance, or have enough TAO (self-delegation + shares from other delegators) in their keys.
These factors themselves will also exacerbate the growth of registration demand.
The number of recovered 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. It entirely depends on the balance between the issuance and recovery of TAO.
The Bittensor network went live on January 3, 2021. According to token recovery 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 within decentralized networks. Unlike traditional methods where resource allocation is manually decided by validators through voting, dTAO introduces a market-based dynamic adjustment mechanism, linking resource allocation directly to subnet network performance, thereby optimizing the fairness and efficiency of reward distribution.
Core mechanism
Market-based dynamic resource allocation
The dynamic TAO distribution mechanism is based on the market performance of subnet tokens. Each subnet in the network has its independent token, and its relative price determines the distribution ratio of TAO issuance among subnets. As market information changes, this distribution ratio will be dynamically adjusted to ensure resources flow to efficient and promising subnets.
Embedded liquidity pool design
Each subnet configures a liquidity pool consisting 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 share through long-term contributions and building efforts. This mechanism avoids the risk of tokens being quickly dumped while encouraging teams to focus on technological improvements and ecosystem building.
Balancing the roles of users and validators
Dynamic TAO's resource allocation is determined not only by the market but also jointly influenced by validators and users. Validators need to evaluate the team's technical capabilities, market potential, and actual performance rigorously, like venture capitalists (VCs). Users further drive the market value formation of the subnet by staking TAO and participating in market transactions.
Economic model analysis
Current funding support
Data shows that currently, subnets in the network can earn an average of about $47,000 in rewards daily, corresponding to approximately $17 million in annual support. This scale of funding far exceeds the median of traditional AI startup seed rounds (about $3 million) and Series A funding (about $14 million), 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 such as OpenAI and Anthropic. With the launch of dynamic TAO, future new issuance of TAO will mainly flow into the liquidity pool 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 with market performance. This mechanism also suppresses short-term behaviors that quickly cash out through over-the-counter (OTC) transactions, laying the foundation for the long-term sustainable development of the network.
Impact and significance
Resource allocation optimization
dTAO adjusts resource allocation dynamically based on market dynamics, ensuring that efficient and high-growth potential subnets receive more resources. This mechanism not only improves overall network efficiency 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 between subnets further promote the development of the 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.
Validator role enhancement
Validators need to play a more critical role in the network. They rigorously assess the value and potential of subnets in a manner similar to venture capital, ensuring the scientific and rational allocation of network resources.
The launch of dTAO marks a significant advancement in decentralized network resource allocation mechanisms. Through market-driven dynamic adjustments, embedded liquidity pool design, and fair issuance models, dTAO achieves efficient and fair resource allocation. Additionally, as an AI network incubation platform, it not only empowers the development of subnets but also provides new development paths for decentralized AI networks.
Agents application on Bittensor
Many people say that Bittensor is the AI coin represented by VC elites and has fallen behind in today's flourishing application era of various agent development frameworks. With the recent surge in AI Agents and the total market value of AI agent-related tokens exceeding $10 billion, especially projects represented by the Virtuals ecosystem dominating the total market cap with a market value of $5 billion (including various utility investment and research analysis type Agents, such as $AIXBT, $VADER, $SEKOIA, etc.), Bittensor seems to be left behind in many people's eyes.
However, in reality, Bittensor still possesses many 'Alphas'. What many people 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, powerful training and inference infrastructure becomes even more crucial.
Currently, Virtuals and Bittensor have collaborated on many ecological fronts.
Many consumer-facing virtuals protocol agents are supported by Bittensor subnets, leveraging TAO's computational power and data ecosystem to create new possibilities, for example,
$TAOCAT
TAOCAT is an AI agent built by Masa in the Virtuals ecosystem, primarily serving as a staunch defender of TAO, 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 allocation 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 set to launch 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 proves that tremendous practical value can be created on top of the Bittensor infrastructure. With the official launch of AgenTAO (SN62), this will become a significant 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 AI agents on the application side emerging from the Bittensor ecosystem!
Source: taogod
Conclusion
The future of Bittensor is exciting, with many research and investment institutions emerging around the Bittensor ecosystem, similar to the Ethereum network. This includes calls from the founder of DCG, podcasts, blogs, and OSS Capital, which is dedicated to studying the Bittensor ecosystem while being a research organization for a subnet. A network of contacts similar to a 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 field have begun to align with and support Bittensor. Therefore, recently, Bittensor's ability to surpass Virtuals in Kaito's mindshare is not without reason.
Source: BitMind Bittensor Subnet 34
In April 2025, Bittensor will hold a The Endgame Summit conference and hackathon in Austin, Texas, with over 300 participants, focusing on bringing more subnets, validators, and miners into the Bittensor ecosystem and expanding its reach.
Endgame Summit
Of course, whether centralized AI projects or decentralized AI projects, the ultimate standard will return to the product itself. Currently,
The Bittensor ecosystem has already emerged and is blossoming.
Source: Outpost AI Research
Recently, the founder of Bittensor summarized the major achievements of various subnets over the past year on his personal X account:
Source: https://x.com/const_reborn/status/1873359385373909008
Therefore, let us continue to maintain our expectations for Bittensor and see what products and use cases will emerge from Bittensor in the future, making it the preferred choice for people looking for specific AI problem-solving solutions!
Appendix
1,https://tengsthoughts.com/p/bittensor-doomed?utm_source=www.chainofthought.xyz&utm_medium=referral&utm_campaign=research-bittensor-flawed
2,https://blog.bittensor.com/tao-token-economy-explained-17a3a90cd44e
3,https://docs.bittensor.com/subnets/register-validate-mine#:~:text=To%20participate%20as%20a%20subnet,UID%20slot%20in%20that%20subnet.&text=You%20do%20not%20have%20to,validate%20on%20the%20Bittensor%20network.
4,https://docs.bittensor.com/subnets/checklist-for-validating-mining#:~:text=Keep%20in%20mind%20that%20to,must%20also%20stake%20enough%20TAO.
