Article source: IOSG
Preface
AI development has made tremendous breakthroughs in recent years due to advances in data, computing power, and algorithm research, particularly with the emergence of OpenAI GPT-4, representing the advent of foundational LLM large models, driving productivity improvements and transforming social efficiency.
However, the shortcomings of these closed-source large models represented by GPT-4 have also come to light, as centralized models often face limitations with third-party integrations that hinder 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 popular 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 when some competition rewards appear, they are usually one-time; subsequent improvement and development work still requires goodwill unless a significant scale is reached and a large community of followers emerges, increasing the potential for income and attracting more contributors for continued improvement.
Therefore, the Bittensor AI project 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, and AI project parties (Subnet Creators), making AI research more transparent and decentralized, allowing anyone to participate in AI contributions and earn the rewards they deserve.
The performance of tokens in the secondary market also validates people's expectations, with prices rising from over $50 in September 2023 to over $500 in December 2024, achieving a tenfold increase!
Recently, Bittensor investor and founder of Digital Currency Group established an accelerator called Yuma, dedicated to incubating subnet projects within the Bittensor ecosystem, and serves as CEO, demonstrating his confidence and potential in the Bittensor project.
Source: Coindesk
Of course, no project's success can be without its doubts. Since its inception, Bittensor has faced considerable FUD, and in this article, we summarize many questions that have not yet received complete answers, trying to understand Bittensor's future positioning and potential in the decentralized AI track.
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 already have their own blockchain, can connect to Bittensor's blockchain 'subtensor' and become part of a subnet.
What is a Subnet?
Subnets are at the core of the Bittensor ecosystem, and each subnet is an independent incentive-based competitive market. Anyone can create a subnet, customize the tasks to be performed, and design the incentive mechanism (in machine learning analogy, the incentive mechanism can be understood as the target loss function, guiding model training towards ideal results). By paying a registration fee (valued in TAO) to create a subnet, one can obtain a netuid for a subnet. Note that a subnet creator does not need to assume the operational tasks within the subnet but delegates the right to operate those tasks to others.
Operating tasks for this subnet provide 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 as a subnet validator. Apart from paying a registration fee in TAO (validators also need to stake TAO), one only needs to provide a computer with sufficient computing resources, register that computer and their wallet in a subnet, and run the subnet miner module or subnet validator module provided by the subnet creator (both modules are Python code in the Bittensor API).
How does the competitive market for subnets work?
The operation of subnet competition is as follows: suppose you decide to become a subnet miner. Subnet validators will assign some tasks for you to complete. Other miners in the subnet will also receive tasks of the same type. After all subnet miners complete their tasks, they submit the results to the subnet validators.
Subsequently, subnet validators will each assess 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 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 promoting the continuous improvement of the overall quality of the subnet. All these competitive processes are completed automatically according to the incentive mechanism coded by the subnet creator.
Source: Steps on how Subnet Creator defines Incentive Mechanism
The incentive mechanism ultimately serves as a judgment on the performance of subnet miners. When the incentive mechanism is well calibrated, it can create a virtuous cycle where subnet miners continuously improve the required tasks through competition.
Conversely, poorly designed incentive mechanisms may lead to exploitation and shortcuts, adversely affecting the overall quality of subnets and undermining the enthusiasm of fair miners.
The specific work of each subnet miner depends on the initial purpose for which the subnet creator established that subnet, which can be quite variable or specific. For example, the miner's task in subnet 1 might be to respond to text prompts sent by the subnet validator and provide the best completion results, or the miner's task in subnet 47 might be to provide storage.
Each subnet also has its unique research and commercialization directions, such as attempting to tackle technical challenges in certain AI fields, like decentralized AI training, verifiable inference, or providing some necessary infrastructure and resources for AI, such as GPU trading markets or data labeling services, or helping users identify AIGC deepfake technology subnets, 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 the project token TAO play significant roles 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 validators. An algorithm called Yuma Consensus (YC) continuously runs on the Subtensor blockchain. Each subnet validator ranks the quality of work for all subnet miners, and the rankings from each validator are collectively submitted as input for the YC algorithm. Generally, the rankings from different subnet validators arrive at Subtensor at different times, but the YC algorithm on Subtensor waits until all rankings arrive; typically, every 12 seconds, the YC algorithm calculates rewards based on the input from all validators' rankings. These rewards (valued in TAO) will be deposited into the wallets of subnet miners and validators. The Subtensor blockchain continuously runs the YC algorithm independently for each subnet.
The YC consensus algorithm considers two main factors: first, a weight vector maintained by each subnet validator, where each element represents the weight assigned to subnet miners; this weight ranks subnet miners based on the historical performance of that subnet validator. The second factor is the staking amounts of each validator and miner. The Yuma consensus on the chain will use this weight vector and staking amounts to calculate rewards and distribute them among subnet validators and miners.
The Bittensor API will transmit and connect the opinions of validators within subnets 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
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 right to validate in that subnet. Note that subnet validators can hold multiple UID slots simultaneously and validate for multiple subnets without needing to increase the amount staked; staking once allows them to choose multiple UID slots for validation across multiple subnets (similar to the concept of restaking).
Therefore, to obtain the most reward, stakers 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 legitimate validation permissions for that subnet. This reduces the risk of validator misconduct since the amount staked becomes a high barrier, increasing the cost of misconduct (at least 1000 TAO is needed 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 staking amount and become one of the top 64 validators in that subnet.
Mining can begin once subnet validators and miners (who do not need to stake) register their keys with the subnet.
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, halved every four years.
Bittensor started in a fair launch manner, with no pre-mined TAO tokens or ICO. Currently, the network generates 7,200 TAO daily, with 1 TAO produced 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 introduced a unique mechanism where the issuance rate is halved once half of the total supply is distributed. This halving occurs approximately every four years and continues at each 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 reclamation mechanism, this curve is dynamically active and not entirely fixed like Bitcoin.
Recycling 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 number of dynamic TAO are reclaimed daily through key (re)registration.
To become a miner or validator, one must register a key on the network and meet other GPU and computing power requirements. Registration requires reclaiming TAO, which means paying a certain amount of TAO to reinvest in the network.
Every key (re)registration removes that TAO from circulating supply and reintroduces it into the protocol's issuance pool, where it can be mined again in the future.
This mechanism delays the planned four-year halving time, as reclaimed TAO is dynamic. When more keys are (re)registered, the cost of TAO reclamation increases, or other subnets are launched, the reclaimed TAO may significantly increase.
Moreover, registration applies not only to new entrants but also to those users who have been deregistered for the following reasons:
For miners, their models and inference are not competitive enough among other miners;
For validators, they fail to maintain the correct weight, sustain issuance, or do not have enough TAO in their keys (self-delegation + shares from other delegators).
These factors themselves will also exacerbate the growth in registration demand.
The number of reclaimed 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 four years after the launch may be delayed to five or six years, or even longer. This completely depends on the balance between the issuance and reclamation of TAO.
The Bittensor network launched on January 3, 2021, and according to token reclamation 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 inefficiencies in resource allocation within decentralized networks. Unlike traditional methods where validators manually vote on resource distribution, 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 allocation mechanism of TAO 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 the subnets. As market information changes, this distribution ratio is dynamically adjusted to ensure resources flow to efficient and potential subnets.
Embedded liquidity pool design
Each subnet configures a liquidity pool consisting of TAO and subnet tokens (subnet/TAO token pairs). Users can exchange TAO for subnet tokens by staking into the liquidity pool. This design incentivizes users to invest in high-performing subnets, indirectly supporting 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 construction. This mechanism avoids the risk of tokens being sold off quickly and encourages teams to focus on technical improvements and ecosystem building.
Balancing the roles of users and validators
Dynamic TAO resource allocation is determined not only by the market but also influenced by both validators and users. Validators must assess the technical capabilities, market potential, and actual performance of teams rigorously like venture capitalists (VCs). Users further push the market value formation of subnets by staking TAO and participating in market transactions.
Economic model analysis
Current funding support
Data shows that currently, subnets in the network can obtain an average of about $47,000 in rewards daily, corresponding to an annual support of approximately $17 million. This funding scale far exceeds the median of traditional AI startups' seed rounds (around $3 million) and Series A funding (around $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 like OpenAI and Anthropic. With the launch of dynamic TAO, future issuance 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 suppresses short-term behaviors that quickly cash out through over-the-counter (OTC) trading, laying the foundation for the network's long-term sustainable development.
Impact and significance
Resource allocation optimization
dTAO adjusts resource allocation dynamically through market forces, 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 fosters 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 teams, ensuring through economic incentives that all participants can contribute to the network's growth.
Validator role enhancement
Validators must play a more important role in the network. They rigorously assess the value and potential of subnets in a manner similar to venture capital, ensuring the scientific and reasonable allocation of network resources.
The launch of dTAO marks a significant advancement in the decentralized network resource allocation mechanism. Through market-driven dynamic adjustments, embedded liquidity pool designs, and a fair issuance model, dTAO achieves efficiency and fairness in resource allocation. Additionally, as an AI network incubation platform, it empowers the development of subnets and provides a new development path for the future of decentralized AI networks.
Agents applications on Bittensor
Many people say that Bittensor represents the AI coin of VC elites and has fallen behind in the flourishing application era of various agent developer frameworks. With the recent rise of AI agents and the total market cap 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 investment and research analysis type agents like $AIXBT, $VADER, $SEKOIA, etc.), Bittensor seems to be left behind in the eyes of many.
However, in reality, Bittensor still has many 'Alphas'. What many do not realize is that the success of Virtuals/ai16z in the consumer AI agent field 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 increasingly important.
Currently, Virtuals has collaborated with Bittensor on many ecological aspects.
Many consumer-oriented virtual protocol agents are supported by Bittensor subnets, leveraging the computational power and data ecosystem of TAO to create new possibilities, for example.
$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 Agent Arena on Bittensor subnet 59, creating a new paradigm for tokenized AI value capture, where any user interaction on X becomes 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 enormous practical value can be created on top of the Bittensor infrastructure. With the official launch of AgenTAO (SN62), this will be 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-oriented AI agents emerge from the Bittensor ecosystem!
Source: taogod
Conclusion
The future of Bittensor is exciting, with many research and investment institutions emerging specifically around the Bittensor ecosystem, similar to the Ethereum network. This includes the calls of the DCG founder, podcasts, blogs, and OSS Capital, which focuses on 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 recent gatherings including Contango, Canonical, Delphi Labs, and DCG, where many experts from the Crypto x AI fields have begun to gravitate towards and support Bittensor. Thus, it is not unreasonable that Bittensor was able to surpass Virtuals in Kaito's mindshare recently.
Source: BitMind Bittensor Subnet 34
In April 2025, Austin, Texas, Bittensor will host The Endgame Summit conference and hackathon with over 300 participants, focusing on introducing more subnets, validators, and miners to the Bittensor ecosystem and expanding its reach.
Endgame Summit
Of course, whether it is 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 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:
Source: https://x.com/const_reborn/status/1873359385373909008
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!
Appendix
1,https://tengsthoughts.com/p/bittensor-doomed?utm_source=www.chainofthought.xyz&ut...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.
