Discovering Taoism - A Journey Into Bittensor - Messari

12/14/23, 11:07 AM Discovering Taoism: A Journey into Bittensor | Messari
Research
Back to Research Reports
Enterprise
Discovering Taoism: A Journey into Bittensor

Sami Kassab
14 hours ago ⋅ 28 min read
Key Insights
Bittensor is the leading decentralized AI project that continuously rewards new, better
machine learning models across a variety of targeted use cases, essentially transforming AI
model creation and discovery into a commoditized process.
Subnets, functioning as self-contained economic markets, play a pivotal role in realizing

Bittensor’s vision of creating an ecosystem capable of generating the resources required
for the production of machine intelligence.
Tokenholders determine the allocation of capital to each subnet, actively shaping the
trajectory of AI development within the network.
Prominent subnets within Bittensor encompass areas such as text, image, and audio
generation, pre-trained model production, web scraping, data storage, cloud computing,
and financial market predictions.
AI, despite a relatively healthy open-source model community, has largely been dominated in a
production sense by large centralized companies like OpenAI, Google, and Anthropic, which all
have governing control over their AI model outputs. Control of centralized models leads to
censorship of outputs and power struggles, as seen in the recent OpenAI leadership drama.
While concentrated control of the models is significant today, there are structural and
technological reasons why decentralized AI presents a more viable route for AI products. Using
crypto coordination and incentive mechanisms enables continuous, granular model discovery
and productionalization for targeted use cases not optimally accounted for out-of-the-box by
the centralized model companies.
Simultaneously, decentralized AI empowers individuals distributed around the globe to

collectively offer their expertise, resources, and intellectual property for the advancement of
https://messari.io/report/discovering-taoism-a-journey-into-bittensor 1/26
shared AI, accelerating the pace of AI development.
Bittensor Overview
Bittensor is the leading decentralized AI project that continuously rewards new, better machine
learning (ML) models across a variety of targeted use cases.
While several decentralized projects at the crypto-AI intersection focus on decentralizing

specific machine learning stack components, such as training, inference, or data collection,
Bittensor takes a different path. Bittensor boldly competes with full-stack AI giants like OpenAI,
Google, and Anthropic, who engage in the entire AI lifecycle — from conceiving new algorithms
and collecting data to training models and hosting them for inference.
Ultimately, Bittensor’s overarching vision is to create an ecosystem capable of generating the

necessary resources required for the production of machine intelligence. To realize this,
Bittensor has built a framework for the creation of digital commodity markets, known as
subnets. These markets can be tailored to incentivize individuals to contribute their expertise,
intellectual property (e.g., models), and digital resources, including compute, storage, and
bandwidth, for a diverse range of AI and non-AI applications.
To grasp the concept of digital commodity markets, think of Bitcoin. Bitcoin operates as a
commodity market for compute, where miners are incentivized to contribute compute power for
SHA-256 hash calculations and are rewarded in proportion to their computational contributions.
Instead of commoditizing SHA-256 hash production, Bittensor commoditizes AI model creation

and discovery. Within Bittensor, each subnet develops its own incentive mechanism tailored to a
specific use case. These use cases include text and image generation, web scraping, data
storage, and pre-trained model production. This approach enables Bittensor to effectively mine
AI discovery and execution resources across many subdomains.
With subnets, Bittensor can address the long tail of AI models, catering to niche industries,
specialized AI applications, and unique problem-solving scenarios often overlooked by
mainstream AI solutions.
Architecturally, Bittensor evolves into a network comprising various self-contained economic

markets, seamlessly united under a single token (TAO) ecosystem, all geared towards the goal
of advancing machine intelligence.
Within this ecosystem, tokenholders shape the network's trajectory by determining the
allocation of capital to each subnet. Their decisions regarding the proportion of network
emissions directed to each subnet actively guides the direction of AI development within
Bittensor.
Origin Story
Bittensor's intricate nature goes far beyond the surface, and delving into its history can provide
a deeper understanding.
January 2021 marked the inception of Bittensor as a text-prompting network, operating similar
to ChatGPT. Within this network, validators would query miners with prompts, initiating a
competition among miners to generate the best possible responses through the use of ML
models.
Validators would then grade miners on their performance using a rubric defined by the network,
known in technical AI jargon as a "reward model." The reward model defined the desired
output and rewarded miners based on their proximity to this target. The miners' rankings were
influenced by the intelligence, speed, and diversity of their responses, prompting fierce
competition among miners as they vied for a greater share of the rewards.
To achieve consensus on miner rankings, Bittensor introduced a mechanism known as the Yuma
Consensus (YC) mechanism, which stands as one of Bittensor's core technological innovations.
Yuma Consensus
The YC mechanism enables validators to express how performant miners are by ranking them
on a scale from 0-1, as determined by the reward model. Validators would then reach consensus
regarding these rankings, and the YC mechanism would translate these rankings into incentives
for miners. The higher a miner's ranking, the more substantial their reward.
Notably, YC operates independently of the consensus mechanism used to secure the

blockchain. Currently, Bittensor's substrate-based blockchain, Subtensor, relies on a Proof-of-
Authority consensus mechanism, permitting authorized entities, which currently only includes
the OpenTensor foundation, to validate transactions. While Bittensor won Polkadot's Parachain
Auction in January, the foundation opted to retain Subtensor as a standalone substrate-based
chain.
YC is agnostic to what is being measured, allowing for fuzzy consensus around probabilistic
truths like machine intelligence. YC doesn’t impose rigid criteria or a strict agreement on how
miners should be evaluated. This stands in contrast to Proof-of-Stake, where validators aim to
agree on a definitive, deterministic state.
This inherent flexibility of YC opened doors to applications beyond text-prompting. By granting

developers the freedom to customize an incentive mechanism and express their own subjective
preferences about what a network should reward through YC, Bittensor could expand its scope
to other AI and non-AI use cases.
The Subnet Revolution

On October 2, 2023, the Bittensor Revolution upgrade marked a transformative turning point,
evolving Bittensor from a single digital commodity network into a versatile ecosystem of
numerous specialized incentive-based competition markets known as subnets. These subnets
operate akin to application-specific blockchains, each tailored for distinct purposes and tasks.
Each subnet is overseen by an owner responsible for designing a unique incentive system
tailored to the subnet's objectives, referred to as the validation stack. Within these subnets are a
dedicated set of validators and miners. Subnet validators execute the subnet's validation stack,
directing the focus of miners toward specific tasks or challenges that underpin the creation of
value.
Furthermore, competition thrives within subnets for validators and miners, as there are limited
spots available. Only the most performant, adaptive, and valuable entities continue to operate
and earn emissions.
The demand for subnets was so substantial that all 32 available slots were filled within just two
months of the upgrade. In the future, the subnet slots cap will be increased.
Bittensor now encompasses subnets dedicated to image and audio generation, data scraping,
pre-training, zero knowledge machine learning (ZKML), cloud computing, storage, and more.
Contrary to the initial perception of Bittensor as solely an inference network, it has grown into a
multifaceted ecosystem.
Up to this point, three distinct types of subnets (SNs) have surfaced:
Infrastructure Subnets: These subnets provide services to other subnets, such as pre-
training (SN 9), optimized data transfer (SN 10), storage (SN 7), and security (SN 14).
Growth-Focused Subnets: These subnets prioritize expanding their user bases over
immediate revenue generation. An example is Cortex.t (SN 18), which offers free access to
ChatGPT Premium and rapidly amassed 60,000 users within its first 24 hours.
Revenue-Focused Subnets: These subnets have a clear path to generating revenue. An

example is the Time-Series Prediction Subnet (SN 18), which concentrates on predicting
financial market movements, presenting opportunities for validators to create revenue-
generating applications on top.
Despite their differences in approach and objectives, all subnets are in equal competition for a
share of the network’s emissions.
The Root Network
The Root network, denoted as SN 0, serves as the meta subnet responsible for determining the
allocation of emissions to individual subnets. SN 0 is restricted to 64 validators with the largest
stake. When a new validator with a larger stake seeks to join the subnet, the lowest-staked
validator is replaced.
Validators within SN 0 essentially participate in a voting process to decide how the network’s
emissions should be distributed among the various subnets. If the majority of the validators
values a subnet's contributions, it will receive emissions, embodying free-market economics in
its purest form.
This voting process involves manually assigning a percentage to each chosen subnet, with the
total allocation summing up to 100%. For instance, a validator might express a preference for
distributing 50% of emissions to SN 1, 30% to SN 2, and 20% to SN 7. The final consensus
percentages are determined through the YC mechanism, which takes into account the
preferences of validators in proportion to the tokens they have staked.
As of now, a total of 7,200 TAO tokens are emitted daily. These emissions are distributed as
follows:
Subnet Miners: 41%
Subnet Validators: 41%
Subnet Owners: 18%
This allocation scheme underscores the lucrative opportunities and incentives available to
subnet owners.
Validation Politics
Validators with large stakes wield significant influence over capital allocation in the network.
Their ability to determine the distribution of emissions to various subnets empowers them to
incentivize miners to join specific subnetworks, ultimately impacting the quality of products
developed on that subnet.
Additionally, TAO holders can delegate their tokens to subnet validators, who earn an 18% take
rate of the dividends generated. Delegators seek validators aligned with their network's growth
and development vision, creating competition among validators to attract token holders. Thus,
validation becomes a political endeavor, requiring validators to convince others to stake with
them and align with their views on the direction of the network.
In reality, Bittensor's direction is currently heavily influenced by five validators holding over 60%
of circulating tokens, making critical decisions in capital allocation, regardless of their AI
expertise. While these validators represent a broader community of token holders, Bittensor's
long-term success hinges on the validators' ability to discern which subnets genuinely
contribute value and which do not.
Validating Intelligence
Subnet validators play a crucial role in assessing miner performance by executing the validation
stack provided by the subnet owner. This task bears resemblance to running a validator node for
Layer-1 blockchains, where the technical complexity is relatively low, and the process can be
likened to a "set it and forget it" operation.
Process of Validating
The process of validating on a subnet is as follows:
1. Weight Assignment: Using the validation stack, validators prompt miners and express
how well they are performing by ranking them on a scale from 0-1, a process referred to as
setting weights. This process creates a vector of weights that reflects the validators'
evaluation of each miner.
2. Transmission to Blockchain: Each validator independently transmits these weight vectors

to the blockchain, and these transmissions can occur at different intervals, typically around
every 100-200 blocks.
3. Weight Matrix Formation: Onchain, a weight matrix is constructed from the vectors
provided by validators. This weight matrix is then utilized in the YC mechanism.
4. YC mechanism: The YC mechanism, a stake-weighted consensus system, determines the

consensus weight matrix used to distribute rewards for miners.
In order to discourage validators from attempting to manipulate weights or dishonestly favor

their own miners, YC includes a variable known as Vtrust, which serves as a validator's trust
score. Vtrust evaluates a validator's reliability and effectiveness by assessing the alignment of
their weight assignments with the consensus of the network. Validators with a higher Vtrust
score directly translates to increased rewards for the validator.
Validation Stack
Validation stacks within Bittensor are meticulously tailored to the unique requirements of each
subnet. To demonstrate, let's explore two distinct subnets as illustrations:
1. Text-Prompting Subnet (SN 1): In this subnet, the validation stack sends text prompts to
miners and then uses a reward model, which consists of several machine learning models,
to access miner responses. Consequently, validators in SN 1 are mandated to possess

high-end GPUs to effectively carry out their validation tasks.
2. Cloud Compute Subnet (SN 27): On the contrary, the validation stack for the cloud
compute subnet focuses on evaluating the computational capabilities of miners. Validators
achieve this by presenting miners with intricate hashing challenges, purposefully designed
to evaluate the processing power and reliability of miners' systems. Notably, this validation
process places lower hardware demands on the validator.
A critical aspect of Bittensor is that the validation stack software is executed entirely offchain by
the subnet validators, enabling it to be data heavy and compute intensive. The only onchain
aspect of the validation process pertains to the weights derived by validators, which are
relatively lightweight. This means that the number of subnets that can emerge on Bittensor is
primarily constrained by the cumulative data size of all subnet weights.
A recurring issue within various subnets revolves around validators replicating the weight
distributions of larger stake validators. This practice enables validators to sidestep the expenses
associated with running validation stacks. Fortunately, the foundation is actively working on a
solution to rectify this situation in the near future.
Validator Operations
Validators can participate in validation across multiple subnets without the need to split their
stake among them, including SN 0. When validating on multiple subnets, validators commit
their entire stake to each individual subnet, eliminating the requirement to allocate, for instance,
50% for SN 1 and 50% for SN 27.
This arrangement incentivizes validators to validate on all subnets with emissions, primarily
because they would miss out on potential rewards if they chose not to validate a subnet. This
decision can diminish the yield for their delegates, potentially causing dissatisfaction and
prompting them to seek an alternative validator to delegate their tokens to.
However, as the number of subnets grows, it may become infeasible for validators to validate on
every subnet. At this time, 15 subnets have emissions, with the majority of validators validating
on every subnet as the costs of renting the necessary GPUs and virtual private servers is around
$5,000 a month, according to a top five validator.
Mining Intelligence
Within subnets, the miners’ job is to decipher the incentives set by the validation stack and carry
out various actions accordingly.
Similar to validators, miners don't perform their tasks on the blockchain itself. Instead, they
complete their jobs offchain and directly communicate the results of their work to the validators.
For instance, in SN 1, miners run machine learning models offchain to respond to validator
queries.
Notably, validators are only interested in the outputs, not the specifics of the models or
hardware used by miners. This unique feature allows miners to keep their models private, which
is often not possible in other decentralized inference projects.
However, this also implies that miners must secure their own GPUs and computing
infrastructure. They have the option to use a cloud provider or rely on a dedicated hardware
setup.
Infinite Adaptation
Being a miner demands a higher level of technical expertise and active involvement compared
to being a validator. As validation stacks evolve, miners must adapt accordingly. In the case of
SN 1, top-performing miners go the extra mile by crafting custom programs, potentially fine-
tuning models, and updating them as the validation stack changes.
With monetary rewards at stake, miners are incentivized to explore ways to gain an edge within
the system, often leading to the discovery of loopholes. For instance, on subnet 1, miners
devised methods to exploit the validation stack. Instead of sending authentic responses to
validators by querying an ML model, they cleverly submitted specific artificial responses that
they knew would get them rewarded highly, all without the need to actually run an ML model.
Consequently, the validation stack had to be updated to address this issue.
Interestingly, this ongoing cat-and-mouse game between miners and validators benefits the
subnets. As miners discover new ways to outsmart the validation stack, it falls upon the subnet
owner to patch vulnerabilities and update the stack. This iterative process continues until a
virtually impervious validation mechanism emerges, one that resists adversarial tactics so
effectively that even well-financed competing R&D teams find it impossible to deceive (credit to
Timo).
In essence, the validation stack takes on the role of a continuous bug bounty, driving relentless
improvement in the subnet’s product and security.
The Dynamics of Mining
Similar to DePIN, Bittensor subnets operate as digital resource marketplaces, enabling entities
to capitalize on their underutilized resource, such as compute power, storage, and bandwidth
and intellectual property, including ML models. This versatility extends to unconventional assets
like an OpenAI API key which can be monetized on SN 18.
Just as Bitcoin miners worldwide seek out stranded and underutilized energy sources,
aggressively striving to reduce their operational costs, Bittensor's competitive markets
encourage similar dynamics. Entities with access to cheap power, substantial computational
resources, or even the human capital of a dedicated group of ML researchers gain a significant
edge within the network. As competition intensifies, less performant and inefficient miners may
find it economically unsustainable to continue operations, mirroring the cyclical nature observed
among Bitcoin miners.
Displayed above are the miner incentive distributions for two subnets: SN 1 (left) and SN 18
(right). In subnet 1, miners who have strategically discerned the preferred outputs of the
validation stack occupy the upper echelons of the earnings hierarchy. Conversely, in subnet 18,
where miners are encouraged to provide OpenAI API keys, the highest earners are those
operating Tier 5 keys, with a descending gradient of rewards corresponding to lower-tier API
keys. These distribution curves serve as a visual representation of how the YC mechanism
transforms miner performance into a dynamic incentive landscape.
Building Applications on Bittensor
Bittensor is primarily designed for businesses and applications to leverage its capabilities
through APIs, not for end-consumer usage. Its core objective is to streamline the developer
experience by eliminating the necessity of engaging with anything crypto-related.
Within Bittensor, validators facilitate external access to the network because they are the sole
entities capable of querying the miners and accessing the network's intelligence. Consequently,
there are two primary avenues for building a business on Bittensor: becoming a subnet validator
or utilizing an API provided by a subnet validator.
The dominant method of access to the network will be via subnet validator APIs. Payment for
API access is straightforward and occurs offchain, directly between the developer and the
validator. Validators have the flexibility to accept various forms of payment, including fiat
currencies. Essentially, using Bittensor can be as straightforward as a developer paying for a
traditional Web2 API.
When a validator operates on every subnet, it enables the use of a single API to access the
entire Bittensor network. This streamlined approach significantly simplifies the developer
experience. For instance, consider building a real-time voice translation video call application. It
requires functionalities like transcription, translation, speech-to-text, and storage, each
potentially available through a dedicated subnet.
One of the most popular APIs operated by a validator for accessing the Bittensor network is
BitAPAI.
Validator Insights and Competition
Validators within the Bittensor network have varying degrees of knowledge and capabilities.
Some may construct routing models and algorithms that provide a deeper understanding of the
miner landscape, allowing for more effective work allocation. This increased efficiency translates
into better applications and services for developers.
For instance, in the text-prompting subnet, a validator can create a comprehensive map that
categorizes miners based on their expertise, such as identifying those best suited for academic,
financial, or entertainment-related requests.
Validators can also develop programs to enhance user responses by querying multiple miners
and selecting the most suitable response, rather than defaulting to the highest-ranked miner.
This diversity in validator bandwidth across the network gives rise to competition at the
validator level. While currently, all applications are free, it is anticipated that revenue models will
be integrated in the future.
Scalability Challenges in Meeting Growing Demand
In its current state, Bittensor is likely to face challenges as it scales in two key areas:
1. Emission setting for subnets – The manual process of emission allocation presents
scalability concerns as Bittensor aims to accommodate a growing number of subnets.
2. Miner sustainability – As demand for Bittensor’s services grows, the absence of a

mechanism linking miner revenue to subnet usage poses significant scaling challenges.
Manual Emission Setting
Currently, Bittensor is limited to 32 subnets, but in the future, this cap will gradually increase,
potentially accommodating hundreds or thousands of subnets. While this expansion holds
promise for network growth, it also introduces a significant challenge related to the manual
process of emission allocation.
In the current operational model, the responsibility of determining emissions for each subnet
rests with the top 64 subnet validators. These validators play a crucial role in assessing the value
proposition of individual subnets and deciding whether they deserve a portion of the emissions
pool.
For instance, with 32 subnets, validators can reasonably analyze each subnet to evaluate its
contributions and potential value. However, as the number of subnets increases, manual
evaluation becomes increasingly complex and time-consuming.
The absence of a mechanism within Bittensor to quantitatively measure the usage and impact of
each subnet exacerbates this challenge. Without such automated assessment tools, the burden
on validators to individually review and allocate emissions to numerous subnets becomes

onerous.
This situation raises scalability concerns. If the network continues to rely on manual evaluations,
there may come a point where validators cannot efficiently and effectively assess the value and
impact of each subnet. Consequently, this could lead to delays in emission allocation decisions
or even inaccuracies in determining each subnet's fair share of rewards.
Miner Sustainability Challenges
Another significant concern within Bittensor pertains to the sustainability of miners. At the heart
of this issue lies the absence of a mechanism that correlates subnet usage with miner revenue.
Currently, a subnet miner's income is determined by two primary factors: the total emissions
allocated to a subnet and the prevailing price of TAO. In essence, the revenue pool available to
miners resembles a fixed pie, and any fluctuations in emissions or the value of TAO can have a
direct impact on miner profitability.
This situation raises a critical question: What happens when the network experiences a surge in
usage driven by a popular application on a subnet, prompting an increase in their operational
costs? Without a mechanism for the network to respond by increasing emissions, miners may
find themselves facing unprofitability.
To draw a parallel, consider storage providers in the Filecoin network. Currently, storage
providers offer their services at no cost, relying on block subsidies to cover their operational
expenses. However, if these subsidies no longer suffice, they have the option to introduce fees
for their services. Bittensor miners, on the other hand, lack the ability to charge fees.
Currently, with a TAO price of $300, subnet miners collectively have access to a daily pool of
$885,000. However, when considering a specific subnet like SN 1, which receives 15% of the
network's emissions, the subnet miner pool is allocated a daily share of $132,800. This
allocation sets the upper limit on their potential expenditures within the subnet.
A potential solution to this challenge could involve:
Granting subnet owners the ability to dynamically adjust the distribution of emissions
between miners and validators. Given that validators have additional revenue streams
from offering network access, reducing their share of emissions and directing more to
miners could help ensure their continued profitability.
Implementing a revenue-sharing mechanism, whereby validators or subnet owners

distribute a portion of their earnings back to miners.
Subnets and Application Landscape

There are too many subnets and applications to cover, but this section will highlight some
noteworthy ones.
Text-Prompting (SN 1)
Objective:
Operated by the OpenTensor foundation, the text-prompting subnet is dedicated to building a

system that can generate superior outputs when compared to other large language models
(LLMs). This is accomplished by merging a variety of open-source and proprietary models,
essentially functioning as an LLM aggregator.
Applications:
ReplyTensor, a Twitter and Telegram reply bot, is currently active on SN 1; however, the subnet
has yet to achieve significant success in delivering highly dependable responses.
Targon (SN 4)
Objective:
Targon is focused on creating a next-generation search engine using multi-modal LLMs for high
throughput search using a system that can comprehend and analyze content in real time.
By crawling the web, watching videos, listening to podcasts, and engaging in online
discussions, Targon aspires to build a system that genuinely understands information as it
encounters it, revolutionizing the way AI systems interpret and process data.
Applications:
Targon introduced Sybil, an application reminiscent of Perplexity.ai, an AI-driven search engine.

While Sybil was initially launched and gained substantial attention, it is offline due to
overwhelming demand. It is currently undergoing maintenance and is expected to be back
online in the near future.
Time Series Prediction (SN 8)
Objective:
The Time Series Prediction subnet establishes an ecosystem where miners contribute their ML
models to predict the future of various markets, beginning with predictions for Bitcoin
(BTCUSD) on the 5-minute chart.
Applications:
The Taoshi dashboard provides a visual representation of the subnet's prediction accuracy,
which is continually improving. Additionally, it offers insights into the predictions made by
miners regarding Bitcoin's price movements.
Pre-training (SN 9)
Objective:
The pre-training subnet provides incentives for miners to develop pre-trained models using the
GPT-2 structure based on the Falcon Refine Web dataset. The subnet operates as an ongoing
benchmark, rewarding miners for achieving the lowest losses, indicating the most accurate
responses. Miners engage in a competitive leaderboard format, vying to optimize their model
weights, with the top-performing miner earning the highest emissions until surpassed.
Subnet owners are actively exploring the integration of federated training techniques into the
network.
Applications:
The subnet dashboard provides insights into the progression of model losses achieved by
miners over time. Remarkably, within just a few weeks of the subnet's launch, the network
consistently attained loss levels below 4, surpassing the performance of OpenAI's GPT2 100M
base model trained back in 2018.
Map-Reduce (SN 10)
One of the primary challenges in distributed model training is the substantial communication
overhead, which involves the transfer of data such as model weights and gradients between
entities. The Map-Reduce subnet is designed to alleviate this communication burden not only
for distributed training but also for various other applications.
Consequently, this subnet holds the potential to support the pre-training subnet in its efforts to
implement federated learning.
Cortex.t (SN 18)
Objective:
Cortex.t is designed to deliver reliable, high-quality text and image responses through OpenAI
API usage. This allows validators to offer GPT-4 Turbo and Dall-E 3 within their APIs for
developers.
Furthermore, Cortex.t logs all the intelligence outputs, creating a valuable dataset for training
models using synthetic data. By utilizing multiple API keys to avoid rate limits, Cortex.t aims to
train models capable of replicating the performance of OpenAI’s models.
Applications:
Corcel – A free chatbot and image generator that gained rapid popularity, amassing
60,000 users within its first 24 hours of launch.
Chat with Hal – An AI assistant accessible through various social networks and messenger
apps, including Discord, Twitter, Nostr, WhatsApp, Telegram, and more.
Cloud Compute (SN 27)
The Cloud Compute subnet Integrates various cloud providers such as Runpod, Lambda, and
AWS into a cohesive unit, creating a unified environment that allows high-level cloud platforms
to offer seamless compute composability across different underlying platforms. This subnet is
designed to be utilized not only by other subnets within the network but also by external users
seeking access to its resources.
Zktensor (SN 28)
Zktensor is building the ZKML layer for Bittensor, which is designed to generate zero-knowledge
proofs to verify that outputs were correctly generated using the right ML model. Additionally,
Zktensor is actively developing TaoPunk, an onchain generative art NFT project based on
Zktensor.
TAO Token Model

Bittensor initiated its journey with a fair launch that included no pre-mined TAO tokens or ICOs.
Currently, the network generates 7,200 TAO tokens daily, issuing 1 TAO per block, with blocks
being created approximately every 12 seconds. The total token supply cap is set at 21 million,
closely following a programmatic emission schedule akin to Bitcoin.
However, Bittensor introduces a distinctive approach, where the issuance rate halves once half
of the total supply has been distributed. This halving occurs approximately every 4 years and
continues at half markers of the remaining token issuance until all 21 million TAO tokens are in
circulation.
The specific point at which these halving events occur is determined by the total token
issuance, rather than block numbers due to Bittensor’s unique token recycle feature. Tokens that
are recycled are effectively burned back into the unissued supply, effectively postponing the
halving date.
The Bittensor network went live on January 3, 2021, and based on token recycle data from
taostats, the planned halving date is projected to be in September 2025. The delay in the
halving event suggests that roughly 1.7 million TAO have been recycled since the network's
inception. Consequently, TAO inflation may reach 0% well before the scheduled halvings.
Tokens are mandatorily recycled for various actions within the network:
1. Miner and Validation Registration – Subnet validators and miners must pay a variable
registration fee based on demand, with competition for these limited slots driving up the
recycling requirement.
2. Transactions – A fee of 145 RAO (0.000000145 TAO) is recycled for each transaction.
Work Token Model
Bittensor’s token model resembles the common work token model, which requires service
providers to stake the native token in order to perform work for the network. On Bittensor,
validators must stake TAO to access the network’s intelligence and develop revenue-generating
applications and APIs.
In the conventional work token model, the amount of tokens staked is proportional to the
amount of work service providers can perform. This relationship creates a dynamic where
service providers earn income (in the native token) based on the amount of tokens they stake.
As the demand for the service increases, more revenue flows to service providers. In a system
with a fixed token supply, service providers will rationally pay a higher price per token for the
right to earn from the revenue stream.
The TAO token model has similar dynamics. Validators with larger stakes receive priority from
miners in terms of having their requests serviced. As the demand for Bittensor’s intelligence
grows and more applications are built on the network, increasing the revenue opportunity for
validators, they will compete for TAO to offer superior APIs. Thus, the token model creates a
linkage between the revenue opportunity and price of TAO.
Moreover, validators will seek more TAO to wield greater influence over emission allocations to
subnets on the Root network.
TAO Valuation
Year-to-date, TAO has surged 11x, propelling its circulating market cap to $1.8 billion and its
fully diluted market cap to $6.3 billion. This re-rating can be largely attributed to the substantial
surge in investor interest in AI in the last year. Faced with limited opportunities for public AI
investments, a growing number of investors began exploring the crypto space as a means to
gain exposure.
With TAO’s price correlated to the network’s revenue opportunity, as described in the token
model section, Bittensor’s potential value can be assessed using a top-down revenue analysis,
exploring various scenarios. However, three constant assumptions are made:
1. The machine learning as a service market reaches $320 billion by 2030.
2. Subnet owners succeed in developing validation stacks that produce products that
outperform or are at parity with open-source alternatives.
3. Developers build revenue-generating applications on top of subnets.
The emphasis here is on the last two assumptions, as their realization remains uncertain.
However, most investors are making an optimistic bet that incentive-based competitive markets
will attract top talent, ultimately transforming these assumptions into reality. In the case that
TAO holds above $300, there would be at least $320 million in annual rewards for miners to
compete over.
Revenue Multiples
Revenue multiples vary significantly across AI companies and crypto networks. For AI
companies, the median revenue multiple in Q4 2022 was 2.5. However, leading AI companies
like OpenAI and Anthropic command much higher multiples:
OpenAI – 66x
Anthropic – 100x
Hugging Face – 90x
Stability AI – 66x
Comparatively, revenue multiples for crypto networks also display considerable diversity:
Ethereum – 122x
Filecoin – 537x
Akash – 539x
Lido – 3x
So, we’ll look at three potential scenarios with varying revenue multiples and market share rates
to project the fully diluted valuation (FDV) of Bittensor. It's important to note that the market
share percentages are based on simplified assumptions, using OpenAI's 4% share of the 2023
ML-as-service market as a comparative benchmark.
Bear Case
$3 billion FDV in 2030 (-52%)
Assumptions:
Bittensor does not achieve the scalability required to support mass user applications and
is primarily utilized by niche decentralized applications
Bittensor trades at a 2x revenue multiple
Bittensor captures 0.5% of the ML-as-a-service market
Neutral Case
$26 billion FDV in 2030 (3x)
Assumptions:
Subnets empower the long tail of AI models, catering to niche industries, specialized AI
applications, and unique problem-solving scenarios often underserved by mainstream AI
solutions
Bittensor is able to scale with numerous revenue-generating applications built on top
Bittensor captures 2% of the ML-as-a-service market
Bullish Case
$154 billion FDV in 2030 (23x)
Assumptions:
Thousands of subnets emerge
Bittensor intelligence products surpass OpenAI’s capabilities
Bittensor successfully executes on federated training to create new models
Bittensor captures 8% of the ML-as-a-service market
Final Thoughts
In a way reminiscent of the early days of Bitcoin, where the concept of a decentralized currency
faced skepticism and uncertainty, decentralized AI now stands at a similar crossroads. As the
risks of centralized AI development and usage become increasingly evident – including model
censorship, power struggles over AI control, and the dominance of a few key players – the case
for decentralized AI is becoming more compelling.
Just as Bitcoin emerged as a counterbalance to fiat currency in response to shifting geopolitical

and macroeconomic landscapes, Bittensor has the potential to serve as a counterbalance to
entities like OpenAI, offering a decentralized and more democratic approach to AI
development and utilization.
--
Thanks to Dustin, timo, Nick, mogmachine, Keith, MrNiche, and Opentensor for their engaging
discussions and willingness to answer questions.
Let us know what you loved about the report, what may be missing, or share any other feedback
by filling out this short form. All responses are subject to our Privacy Policy and Terms of
Service.
All content was produced independently by the author(s) and does not necessarily reflect the
opinions of Messari, Inc. Author(s) may hold cryptocurrencies named in this report. This report is
meant for informational purposes only. It is not meant to serve as investment advice. You should
conduct your own research, and consult an independent financial, tax, or legal advisor before
making any investment decisions. Nothing contained in this report is a recommendation or
suggestion, directly or indirectly, to buy, sell, make, or hold any investment, loan, commodity, or
security, or to undertake any investment or trading strategy with respect to any investment,
loan, commodity, security, or any issuer. This report should not be construed as an offer to sell
or the solicitation of an offer to buy any security or commodity. Messari does not guarantee the
sequence, accuracy, completeness, or timeliness of any information provided in this report.

Please see our Terms of Service for more information.
No part of this report may be (a) copied, photocopied, duplicated in any form by any means or
(b) redistributed without the prior written consent of Messari®.
Sami Kassab
We believe crypto is the technology of free people, free thinking, and free markets.
50 Reports Old_Samster
Terms of Service
Sami Kassab is an Privacy Policy SOC 2 Compliance Status
Enterprise Support
Research Analyst focusing primarily on Web3 Infrastructure and Bitcoin. Sami
previously spent 5 years as an Aerospace Engineer designing aircraft engines and missile & defense systems.
Mentioned in this report
TAO $332.09 +4.61%
Read more
Research Reports
Senate Banking Committee House Financial Services Committee

Enterprise Policy Dec 12, 2023 Enterprise Policy Dec 5, 2023
Analyzing Solana's Growth: Airdrop Dissecting Derivatives Data

Impact and User Engagement Enterprise Decentralized Exchanges
Enterprise Layer-1 DeFi Nov 30, 2023 Nov 29, 2023
Stablecoin Brief: December 13th, Consumer Sector Brief: December

2023 11, 2023
Pro Stablecoins Dec 13, 2023 Pro NFTs Web3 Dec 11, 2023
Are PFPs making a comeback? The Week in Crypto Policy: E&C

Pro NFTs Dec 8, 2023 Advances 2 Bills; More AML/CFT
Pro Policy Dec 8, 2023
Read more
Based on your watchlists
Create a new watchlist

Discovering Taoism - A Journey Into Bittensor - Messari

Uploaded by

Copyright:

Available Formats

Discovering Taoism - A Journey Into Bittensor - Messari

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Discovering Taoism - A Journey Into Bittensor - Messari

Uploaded by

Copyright:

Available Formats

12/14/23, 11:07 AM Discovering Taoism: A Journey into Bittensor | Messari

Back to Research Reports

Discovering Taoism: A Journey into Bittensor

14 hours ago ⋅ 28 min read

Subnets, functioning as self-contained economic markets, play a pivotal role in realizing

Simultaneously, decentralized AI empowers individuals distributed around the globe to

shared AI, accelerating the pace of AI development.

While several decentralized projects at the crypto-AI intersection focus on decentralizing

Ultimately, Bittensor’s overarching vision is to create an ecosystem capable of generating the

Instead of commoditizing SHA-256 hash production, Bittensor commoditizes AI model creation

Architecturally, Bittensor evolves into a network comprising various self-contained economic

Notably, YC operates independently of the consensus mechanism used to secure the

This inherent flexibility of YC opened doors to applications beyond text-prompting. By granting

The Subnet Revolution

Up to this point, three distinct types of subnets (SNs) have surfaced:

Revenue-Focused Subnets: These subnets have a clear path to generating revenue. An

The Root Network

Subnet Miners: 41%

Subnet Validators: 41%

Subnet Owners: 18%

The process of validating on a subnet is as follows:

2. Transmission to Blockchain: Each validator independently transmits these weight vectors

4. YC mechanism: The YC mechanism, a stake-weighted consensus system, determines the

In order to discourage validators from attempting to manipulate weights or dishonestly favor

to access miner responses. Consequently, validators in SN 1 are mandated to possess

The Dynamics of Mining

Building Applications on Bittensor

Validator Insights and Competition

Scalability Challenges in Meeting Growing Demand

2. Miner sustainability – As demand for Bittensor’s services grows, the absence of a

Manual Emission Setting

on validators to individually review and allocate emissions to numerous subnets becomes

Miner Sustainability Challenges

A potential solution to this challenge could involve:

Implementing a revenue-sharing mechanism, whereby validators or subnet owners

Subnets and Application Landscape

Operated by the OpenTensor foundation, the text-prompting subnet is dedicated to building a

Targon introduced Sybil, an application reminiscent of Perplexity.ai, an AI-driven search engine.

Time Series Prediction (SN 8)

Map-Reduce (SN 10)

Cortex.t (SN 18)

Cloud Compute (SN 27)

Zktensor (SN 28)

TAO Token Model

Work Token Model

1. The machine learning as a service market reaches $320 billion by 2030.

3. Developers build revenue-generating applications on top of subnets.

Hugging Face – 90x

$3 billion FDV in 2030 (-52%)

Bittensor trades at a 2x revenue multiple

Bittensor captures 0.5% of the ML-as-a-service market

$26 billion FDV in 2030 (3x)

Bittensor is able to scale with numerous revenue-generating applications built on top

Bittensor trades at a 4x revenue multiple

Bittensor captures 2% of the ML-as-a-service market

$154 billion FDV in 2030 (23x)

Thousands of subnets emerge

Bittensor intelligence products surpass OpenAI’s capabilities

Bittensor successfully executes on federated training to create new models