Bittensor: AI Algorithm Aggregation Platform
Project Basic Information
Operating Mode
Bittensor is dedicated to creating a platform for aggregating AI algorithms, aiming to foster the development and decentralization of artificial intelligence. Utilizing a design concept similar to Polkadot’s parachains, Bittensor connects different subnets to the mainnet via the Bittensor API to form a complete blockchain network. Miners and validators are the core components of the subnets; miners provide operational and upgrade support for each subnet’s AI models, while validators assess the quality of the miners’ work. The assessment results are used to distribute TAO token rewards.
Subnet creators can earn token rewards by providing various AI-supported services, and users can purchase these services to accelerate their business development. By aggregating different AI models, and since the type of AI is determined by algorithms, Bittensor brings together various AI algorithms to better meet user needs. It also encourages knowledge sharing between subnets to promote collaborative research among different algorithms. Ultimately, the project’s vision is to create more powerful decentralized AI models.
The current operating mode of the Bittensor blockchain at this stage of development is depicted in Figure 1–1, with the network having launched 32 subnets.
Project Team
Ala Shaabana, PhD in Computer Science, a machine learning researcher who has worked at Instacart and VMware.
Jacob Steeves, holds a Bachelor’s degree in Mathematics and Computing, a machine learning researcher who has worked at Google.
From the publicly available information about Bittensor team members, it is evident that the core team has extensive development experience in the AI field, providing solid support for the project’s operations.
Decentralization Tendency
Public information reveals that Bittensor’s development has only been supported by the OpenTensor Foundation and has not accepted any personal or institutional investments, demonstrating the team’s intention to maintain the project’s decentralized development. Notably, the OpenTensor Foundation operates the largest validation pool and is the largest validator on the Bittensor network, with a staking share of 23.41%, which is twice that of the second place.
Bittensor’s protocol governance is divided into three phases:
1. Centralized management by the foundation;
2. Senate mode (current phase): The senate consists of staking pools that hold more than 2% of the TAO token’s circulating market value, totaling 12 seats;
3. Community management;
According to the operations under the Senate mode, control over the project’s development remains concentrated in the hands of the OpenTensor Foundation. Therefore, Bittensor’s progression towards decentralization is relatively slow, and its decentralization tendency is not pronounced. This centralized control structure may impact the project’s deep practice and development in the decentralization philosophy.
Development Strength
Bittensor was founded in 2019 by founder Jacob Steeves and co-founder Ala Shaabana, with the enigmatic developer known by the nickname Yuma Rao assisting in writing the project’s whitepaper. Key events in the project’s development are shown in Table 1–1:
From the official release of key time nodes, it is evident that the team has a solid technical foundation in the field of technology development. However, the economic model of the project, especially the distribution mechanism of TAO tokens, was devised by a mysterious developer named Yuma Rao. Therefore, the TAO token distribution consensus mechanism — Yuma Consensus — was named after him, earning Yuma Rao the nickname “Mr. Nakamoto of TAO tokens” to highlight his central role in the project’s progress.
Innovation Compared to Similar Projects in the Field
The cryptocurrency market’s AI track can be subdivided into three sub-tracks: computing power, algorithms, and models. Bittensor is a pioneer in the algorithm track, with many innovations that differ from other projects:
Business Model Innovation: While many projects focus on computing power aggregation trading, Bittensor innovatively proposed the algorithm aggregation trading model for the first time, marking a breakthrough in this field and providing a new pathway for market development.
Distributed Mixture of Experts (MOE): Each subnet focuses on a specific aspect of data. When new data is introduced, subnets collaborate to form a collective that achieves an optimal answer much faster and of higher quality than traditional single-solution models.
Digital Hive Mind: Bittensor encourages collaborative learning among network nodes to enhance performance and accuracy. Similar to how neurons work together in the human brain, this process involves nodes exchanging data samples and model parameters to form a network that self-optimizes over time to make more accurate predictions.
Project Model
Business Model
The Bittensor economic system consists of three roles: users, validators, and miners.
Miners: Miners host AI models and provide them to the network; they are responsible for maintaining and ensuring the operation of the models. Miners receive rewards based on their contributions to the network. The key success factor for miners is to reduce costs and provide high-quality services.
Validators: Validators are divided into subnet validators and mainnet validators. Subnet validators assess the quality of miners’ products and services, while mainnet validators evaluate the quality of results from each subnet and also act as routing nodes between user demands and miner services. Mainnet validators are held by the subnet validators with the largest stakes; hence, the key success factor for validators is the amount of tokens staked, with larger stakes giving greater influence over the final assessment results.
Users: End users seeking high-quality results and developers building applications using AI services.
The Bittensor business process is shown in Figure 2–1.
Regarding whether users are charged for using services within the network, the official document does not mention it. The future revenue distribution plan for services is also not clearly stated. Thus, the current operation of the network relies entirely on the issuance of TAO tokens, indicating that the project is still in its early stages.
Yuma Consensus
Yuma Consensus is a consensus mechanism used to determine the distribution of newly produced TAO tokens. Its innovation lies in providing a complete evaluation system and rules for assessing the output results of artificial intelligence models and the work quality of model supporters.
The distribution of TAO tokens involves three steps:
- Mainnet validators score the 32 subnets, and through Yuma Consensus, the number of tokens each subnet can distribute is calculated.
- The rewards allocated to each subnet are distributed as follows: 18% to the mainnet creator, 41% to the validators of that subnet, and 41% to the miners of that subnet. The distribution amount for each miner and validator is decided by Yuma Consensus.
- On the subnet, subnet validators score the quality of services provided by miners. Based on this, Yuma Consensus calculates the number of tokens each miner can receive, using the amount of TAO staked by each subnet validator as a weight. The rewards each subnet validator receives are based on how close their evaluation results are to the final consensus. The closer the results, the higher the proportion of rewards; otherwise, it’s lower, reflecting Yuma Consensus’ mechanism to prevent validators from acting maliciously.
Token Model
Token Distribution
The total supply of TAO is 21 million, according to the white paper: there is no pre-mining by the project team, and every circulating token must be earned by actively participating in the network.
As of now, 6.64 million TAO tokens have been circulated, accounting for 31.64% of the maximum supply, as shown in Figure 2–2.
According to the whitepaper, the Bittensor network undergoes a halving every 10.5 million blocks, with 64 halvings planned over approximately 45 years, completing in 256 years. Specific halving times are shown in Figure 2–3. The reward per block is 1 TAO, distributed approximately every 12 seconds, totaling about 7,200 TAO tokens each day. These rewards are distributed to miners and validators.
Staking Model
To become a subnet validator, TAO tokens must be staked. Only the largest 64 subnet validators on any specific subnet are considered licensed validators, and only licensed subnet validators are eligible to receive rewards.
Ordinary users can stake their TAO tokens with subnet validators to enjoy TAO token rewards.
85.3% of the circulating supply is already staked, indicating the following:
- The Opentensor Foundation mined a significant amount of tokens at the launch stage, which were staked in the official mining pool, attracting more TAO staking. This is why it consistently ranks first in staking volume.
- New entrants are optimistic about the future price prospects of TAO tokens and have staked their acquired tokens.
Thus, the TAO token staking model is simple but reflects the optimistic tendency of ecosystem participants.
TAO Value Determination
In the Bittensor network, there is no scenario of centralized or regular token burning, so the total supply of tokens will remain inflationary, but the inflation rate will decrease over time, similar to the inflation mechanism of Bitcoin.
The staking ratio of circulation becomes a primary reference factor for assessing the token price, calculated as: staking ratio = total staked tokens / total issued tokens.
- A staking ratio around 85% indicates a stable outlook for TAO;
- A staking ratio above 90% suggests an optimistic outlook for TAO;
- A staking ratio below 80% indicates a pessimistic outlook for TAO.
In summary, the current investment value of TAO lies in staking rewards, while the long-term value is in users purchasing services with TAO, leading to higher TAO prices, increased income for miners and validators, attracting better AI models (miners and validators), improving service quality, and thereby attracting more users, creating a growth flywheel.
Token Price Performance
According to Coingecko statistics, since TAO began trading in April 2023, its price has risen tenfold (from $50 to $500), primarily traded on Gate, MEXC, and Binance, with a daily trading volume of $16 million. Its circulating market cap is approximately $450 million (issued token value — staked token value), with a turnover rate of less than 5%, which is relatively low. This indicates that most market investors are not familiar with TAO, lacking trading momentum.
Project Risks
- Centralization Risk: As mentioned earlier, the Opentensor Foundation controls 23.4% of the staking volume, a proportion that continues to rise. Thus, Bittensor’s level of decentralization still needs improvement.
- Flaws in Yuma Consensus: The mainnet validators are the strongest among the 64 subnet validators, and the objectivity of model evaluation results does not entirely depend on the true assessment of TAO holders. This could lead to subnet bias among validators, resulting in users receiving suboptimal evaluation results due to validators aiming for more rewards.
- Subnet Number Limitation: Competing with major companies like OpenAI requires more than just 32 simple AI models. However, due to the limitation on the number of validators, it’s not feasible to add more subnets or models, an area where the project still needs enhancement.
Recent Developments
Based on discussions from the official Discord, the following proposals might be implemented in the future:
- Establish a Dynamic TAO mechanism, transferring the power to evaluate subnet work quality from a few validators to all TAO holders.
- Discuss removing the limit on the number of mainnet validators, increasing the subnet cap to 1024, thus lowering the barriers for new models to join Bittensor.
If these proposals are successfully implemented, Bittensor’s level of decentralization would significantly increase, and with a richer variety of models, its competitive edge in business would also strengthen.
Summary
Currently, investors’ interest in AI concept assets is unprecedentedly high. Computing power and algorithms are two critical areas in need of breakthroughs in AI development. As the pioneer of the algorithm aggregation concept, Bittensor has established a consensus evaluation system for model quality through the Yuma Consensus, gathering various types of AI algorithms to better meet user needs. Moreover, Bittensor encourages knowledge sharing between different subnets and promotes collaborative research between different algorithms, aiming to build a stronger decentralized AI model. This project is likely to foster innovation in the AI sector, pushing the industry forward and is worth close attention.
Website: Frontier Lab
Follow Frontier Lab on X
Telegram channel: https://t.me/frontierlab
