Preface
With the progress of data, computing power, and algorithm research in recent years, AI development has made great breakthroughs, especially the emergence of OpenAI GPT-4. The arrival of the basic LLM large model represented by it has promoted the improvement of productivity and the transformation of social efficiency.
However, the disadvantages of these closed-source large models represented by GPT-4 have also been revealed, namely, centralized models usually have restrictions on third-party integrations, which undermines the scalability and interoperability of AI agents based on centralized models.
Therefore, large open source models such as the Llama series have been sought after by more and more researchers, but open source does not mean transparency, and it also faces many challenges.
The main dilemma is that open-source AI development offers no economic incentives for most contributors. Even with some competition rewards, these are often one-off, and subsequent improvement work still requires altruism, unless at a certain scale with a significant community following, there will be more possibilities for realizing revenue and attracting more contributors to continue improvements.
Therefore, the Bittensor AI project attempts to utilize web3 token mining to make open-source AI development more sustainable, verifiable, and efficiently operational. Through Yuma Consensus, it aligns resources with research parties (miners), validation parties (validators), and AI project parties (subnet creators), 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 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, Bittensor's investor and founder of Digital Currency Group established an accelerator named Yuma, dedicated to incubating subnet projects within the Bittensor ecosystem, and serves as CEO, demonstrating confidence and potential in the Bittensor project.
Source: Coindesk
Of course, no project's success comes 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 potential 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 part of a subnet.
What is a subnet?
Subnets constitute the core of the Bittensor ecosystem, with each subnet being an independent incentive-based competitive market. Anyone can create a subnet, customize the tasks to be executed, and design the incentive mechanism (in machine learning analogy, the incentive mechanism can be understood as the target loss function that guides model training towards ideal results). One only needs to pay a registration fee (priced in TAO) to create a subnet and obtain a subnet's netuid. Note that a subnet creator does not need to undertake the operational tasks within the subnet but delegates the rights to operate the tasks to others.
The tasks operated by the subnet provide another way for others 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 the registration fee priced in TAO (if a validator, they also need to stake TAO), they only need to provide a computer with sufficient computing resources and register that computer and their wallet to 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: Suppose you decide to become a subnet miner. The subnet validator will assign some tasks for you to complete. Other miners in the subnet will receive the same type of tasks. Once all subnet miners complete their tasks, they submit the results to the subnet validator.
Subsequently, subnet validators will each assess and rank the quality of tasks submitted by subnet miners. As a subnet miner, you will receive rewards based on the quality of your work (priced in TAO). 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 automated according to the incentive mechanisms written by the subnet creator.
Source: Steps on how Subnet Creator defines Incentive Mechanism
The incentive mechanism is ultimately a judgment of the performance of subnet miners. When the incentive mechanism is well-calibrated, a virtuous cycle can form, with subnet miners continuously improving the required tasks in competition.
Conversely, poorly designed incentive mechanisms can lead to exploitations and shortcuts, negatively impacting the overall quality of the subnet and hindering the enthusiasm of fair miners.
The specific tasks of each subnet miner depend on the original purpose for which the subnet creator established the subnet, which can be quite variable or 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, while in subnet 47, the miner's task 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, like training decentralized AI, verifiable inference, or providing some infrastructure and resources required 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 increase!
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 distribution of rewards 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 arrive at Subtensor at different times, but the YC algorithm on Subtensor waits for all rankings to arrive, usually every 12 seconds, and calculates rewards based on the input from all validators' rankings. These rewards (priced in TAO) are 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: first, a weight vector maintained by each subnet validator, where each element represents the weight assigned to subnet miners. This weight indicates the ranking of all subnet miners based on the historical performance of the subnet validator. The second factor is the staking amount of each validator and miner. The on-chain Yuma consensus will use this weight vector and staking amount to calculate rewards and allocate them among subnet validators and subnet miners.
The Bittensor API also serves to transmit and connect validator opinions on the subnet with the Yuma consensus on the Subtensor blockchain. Moreover, validators within the same subnet will only connect to miners in that same subnet, and 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 desired subnet, obtaining a UID slot in that subnet, which represents the verification rights for that subnet. Note that subnet validators can simultaneously hold multiple UID slots and validate for multiple subnets, but do not need to increase the staking amount; one staking of TAO can choose multiple UID slots to validate for multiple subnets (similar to the restaking concept).
Therefore, to obtain the maximum reward, validating stakers will 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 the true verification permission 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 (one must have at least 1000 TAO to set weights in a subnet). Each validator will try to build a good reputation and performance record to attract more delegated TAO staking to increase their staking amount and become one of the top 64 validators in that subnet.
Once subnet validators and subnet miners (running miners do not need to stake) register their keys to the subnet, they can start 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: a total cap of 21 million, halving every four years.
Bittensor started with a fair launch, with no 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 where once half of the total supply is distributed, the issuance rate is halved. 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 concept of Bitcoin, due to its recovery mechanism, this curve is actively dynamic and 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 from January 2009 to November 2012 during its first cycle).
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 computing power requirements. Registration requires recovering 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 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 because the recovered TAO is dynamic. When more keys are (re)registered, the cost of TAO recovery increases, or other subnets are released, the recovered TAO may significantly increase.
Moreover, registration applies not only to newcomers but also to users who have been deregistered for various reasons:
For miners, their models and reasoning are not competitive enough among other miners;
For validators, they fail to consistently set the correct weights, maintain issuance, or do not have enough TAO (self-delegation + shares from other delegators) in their keys.
These factors themselves will also exacerbate the growth in registration demand.
The number of TAO recovered = the total number of registered (or re-registered) keys for each subnet * average registration (or re-registration) cost)
Therefore, the first halving originally planned for four years after launch may be delayed to five or six years, or even longer. This 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 solving the inefficiencies in resource allocation in decentralized networks. Unlike the traditional method where resource allocation is decided by manual voting from validators, 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 mechanism
Market-based dynamic resource allocation
The dynamic allocation mechanism of TAO is built 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 adjusts dynamically 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 staked in the liquidity pool for subnet tokens. This design incentivizes users to invest in well-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 tokens being quickly sold off while encouraging the team to focus on technological improvement and ecosystem building.
Balancing user and validator roles
The dynamic allocation of TAO resources is determined not only by the market but also influenced by validators and users. Validators need to conduct strict evaluations of the team's technical capabilities, market potential, and actual performance, like venture capitalists (VCs). Users drive the market value formation of subnets by staking TAO and participating in market transactions.
Economic model analysis
Current funding support
Data shows that the average daily reward for subnets in the current network is about $47,000, corresponding to an annual support of about $17 million. This scale of funding far exceeds the median of traditional AI startup seed rounds (about $3 million) and Series A financing (about $14 million), 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 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 to market performance. This mechanism also suppresses short-term behaviors that quickly cash out through OTC transactions, laying the foundation for the long-term sustainability 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 the overall efficiency of the network but also promotes 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 cooperation 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 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 rational distribution of network resources.
The launch of dTAO marks a significant advancement in the resource allocation mechanism of decentralized networks. 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 a new development path for the future of decentralized AI networks.
Agents application on Bittensor
Many people say that Bittensor is an AI coin represented by the VC elite, now lagging behind the flourishing application era of major agents' 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 with a market cap of $5 billion (including various practical investment and research-oriented agents like $AIXBT, $VADER, $SEKOIA, etc.), Bittensor seems to be falling behind in the eyes of many.
However, in reality, Bittensor still possesses many 'Alphas'. Many people do not realize 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, robust training and inference infrastructure becomes increasingly important.
Currently, Virtuals has collaborated with Bittensor 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.
$TAOCAT
TAOCAT is an AI agent built by Masa within 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. 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 bets, with copy-trading functionality coming soon.
$SERAPH: The first project focused on verification infrastructure, aimed at certifying the AI agent wave that will sweep 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 mark an important milestone for automated software engineering agents on Bittensor, with all Bittensor subnets gradually developed by agents on Bittensor. In the future, we will see more application-end AI agents 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 the recent endorsement from the founder of DCG, podcasts, blogs, and OSS Capital focusing on Bittensor investments, dedicated to researching the Bittensor ecosystem while being a subnet research organization. A network similar to the PayPal mafia has formed around Bittensor, with recent gatherings involving Contango, Canonical, Delphi Labs, and DCG, where many experts from the crypto and AI fields have begun to lean towards and support Bittensor. Thus, it was not without reason that Bittensor was able to surpass Virtuals in mindshare recently.
Source: BitMind Bittensor Subet 34
In April 2025, Bittensor will host a conference and hackathon in Austin, Texas, with over 300 attendees, focusing on introducing more subnets, validators, and miners into the Bittensor ecosystem and expanding its territory.
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 emerged and is flourishing.
Source: Outpost AI Research
Recently, Bittensor's founder 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 maintain our expectations for Bittensor and see what products and use cases can emerge from Bittensor in the future, becoming the preferred destination for people seeking specific AI 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.
