Why Bittensor Will Win The Decentralised AI Race
Everything you need to know about Bittensor and $TAO
According to CoinGecko, decentralized AI is expected to be one of the most significant trends this cycle. The key question is which tokens will accrue the most value. In my opinion, $TAO will lead this space and outperform all other AI tokens.
All views and data expressed in this article are based on research conducted at the time of writing. Bittensor is a relatively new and extremely fast-paced project with many independent actors, so please use the most current data when making any investment decisions.
If you find the post useful, consider subscribing to HFA Premium :)
Table of Contents
Bittensor in a nutshell
What is Bittensor
TAO 0.00%↑Tokenomics
What makes Bittensor Unique
Applications on Bittensor
Corcel
Tensorplex
Datura & Sybil
Subnets on Bittensor
Subnet 8 (Taoshi) Proprietary Trading Network
Subnet 6 (Nous Research) Finetuning
Subnet 3 (Myshell) Text to Speech
Subnet 30 & 24 (Wombo & Omega) Image Generation & Scraping
Subnet Stats
Competitors
Ritual
Morpheus
Questions and Common Concerns
Does Bittensor make sense for businesses?
What happens when Google or Open AI starts mining TAO
What happens when most of the $TAO is mined?
What happens when subnets get too big?
What about pre-mined tokens and early holders?
Future of Bittensor
Dynamic TAO
R&D of New AI Models
Bullish & Bearish Keypoints
Alpha & My Strategy
How to participate in the Bittensor ecosystem
Potential Airdrops
When I am selling TAO
Conclusion
Bittensor In A Nutshell
Bittensor is a blockchain network designed to support multiple subnets. Since AI outputs are uniquely non-deterministic, probabilistic, and fuzzy, it's impossible to devise a single algorithm or function that can assess the quality of an AI application comprehensively. Consequently, each subnet supports a specific application with its own validation rules and incentive mechanisms, established by the subnet creators.
Each subnet consists of a set of miners and validators. Users submit their requests for tasks to the validators, who then delegate these tasks to the miners, asking them to provide the best possible output. Validators evaluate this output and rank the quality of work produced by the miners. These evaluations are collectively used as input for the Yuma Consensus mechanism on the blockchain. The results from the Yuma Consensus then dictate how rewards for the subnet miners and validators are distributed.
Miners and validators can participate in multiple subnets, provided they adhere to the subnet-specific rules. Blocks are mined approximately every 12 seconds, with each block distributing 1 TAO as a reward. This reward is shared across 32 subnets and further allocated among the miners, validators, and subnet owners within each subnet.
In summary, Bittensor creates a competitive environment where miners and validators compete in their specialized use cases, striving to deliver the maximum value for a share of the TAO rewards.
$TAO Tokenomics
Maximum Supply: 21 million
Emission Timeline: Halving in roughly every 4 years
Market Cap: $2.5 billion, ranking 43
Fully Diluted Valuation (FDV): $7.8 billion
Current Circulating Supply: 6.68 million TAO tokens or 31.8% of Total Supply
Staked: 85.38%
Subnet Registration Cost: 1356 TAO
At the current inflation schedule, this leads to 7200 new TAOs being issued into circulation every 24 hours. Once half of the supply has been issued, the rate of issuance is halved — with 12 seconds per block that amounts to a halving every 4 years. Every half marker of the remaining amount to be issued creates a new halving event until all 21 million TAO are in circulation.
The primary demand of TAO at the moment is subnet registration, delegation, and speculation but in the future, dynamic TAO would add more utility to the token. Some subnets can also choose to have additional utility for TAO like staking a certain amount in their validators to query the subnet or a fee in TAO for the output.
Under the current inflation schedule, this results in 7,200 new TAOs being issued into circulation every 24 hours. Once half of the total supply has been issued, the issuance rate is halved—this translates to a halving every 4 years, given the 12-second block time. Each half marker of the remaining amount triggers a new halving event until all 21 million TAO are in circulation.
Currently, the primary demand for TAO stems from subnet registration, delegation, and speculation. However, in the future, dynamic TAO (a proposal on Bittensor) is expected to add buy pressure on TAO. Additionally, some subnets may introduce extra functionalities for TAO, such as requiring staking a certain amount in their validators to access the subnet or charging a fee in TAO for outputs. DeFi ecosystem for TAO outside of the mainnet is growing which is expected to add additional utilities for TAO.
What makes Bittensor unique
Out of all the decentralized AI solutions out there, Bittensor is the only functional solution with good user interfaces and independent teams working on top of it that I could find. The subnetworks are live, GitHub is public, core team is doxxed and active. Their stats are visible, the network is trackable, the token was fairly launched, and I have tried some of their apps and they work well.
Despite all the red flags of boldly claiming to be the evolution from Bitcoin and having a 21M token supply to mimic $BTC, I was attracted to the network because of its vision of abstracting the AI layer from the application layer.
You can’t compete with web2 companies like OpenAI by adding a paywall to your website and forcing users to create a new wallet and buy a token they have never heard of.
TAO would never be used as a “gas’ token or a utility token to use AI services which I think is key to attracting non-crypto users. $TAO just creates a positive feedback loop in which miners compete amongst themselves to produce more performant models to grab a larger share of $TAO and facilitate a more positive user experience. Applications that use output from subnets would need to find their business model just like any other real company.
Unlike many crypto projects that spend millions in “ecosystem grants”, “hackathons”, “events” and even pumping their meme coins to get traction, Bittensor has been able to attract independent reputable teams who are not even involved with crypto directly like Nous Research, Wombo, Kaito, Taoshi and others.
Applications on Bittensor
There are 32 subnets each with its own AI use case. Anyone can utilize these subnets and develop applications on top of them. The most user-friendly applications live are -
Corcel
Corcel is the most developed app on the Bittensor ecosystem. It has an amazing UI similar to Chat GPT for text generation and translation. It also has image generation, image search, editing software, and open API access. It uses subnets 18 & 19 to provide an AI response in a clean UI.
People are using Corcel API to develop applications on top of it. There is a community channel in their discord where people discuss various ideas and develop cool applications.
Tensorplex
Tensorplex provides a suite of AI and DeFi products. Tensorplex stream summarises web3 podcasts and extracts key insights from them. You can find many podcast summaries in their web interface. They also have a chat interface like chat GPT that allows you to ask any questions related to web3 and it uses transcripts to form a response. Currently, Tensorplex utilizes subnet 18 to create succinct summaries but in the future, they plan to develop their own model.
Tensorplex LST provides a liquid staking solution for Decentralised AI Network tokens, beginning with Bittensor (TAO) through stTAO. This product is still at it’s early stages and, as of this writing, there are 34 537 TAO (~$13.85 m) staked in the protocol. Bittensor is a stand-alone L1 and not compatible with any large ecosystems like Ethereum, Solana or Cosmos. stTAO can unlock DeFi opportunities for TAO while benefiting from Bittensor staking.
Pendle has recently added support for wTAO and stTAO in its platform. You can get wTAO on Ethereum mainnet by bridging it using Tao Bridge and you can stake it in tensorplex to get stTAO. You can use stTAO in Pendle, the current yield on this product is around 52% APY
Datura & Sybil
Datura is like a decentralized perplexity. It is an LLM-powered search engine that gives you a chat GPT-like text response and provides you with related tweets, web pages, research papers, and various sources in a user-friendly web app. Datura uses subnet 22 for Twitter data which, according to their discord, has scraped over 30M tweets and growing.
I would highly recommend you give it a try, I was able to find unique information and resources while writing this article that I could not find just by Google or even chat GPT.
Another app similar to Datura is Sybil which uses subnet 4 called Targon.
There are many other applications that we would discuss while looking at subnets. Few applications exist outside of the main network like Tpad: A launch pad on Ethereum for Bittensor projects; Kizuna: A chat application similar to Corcel; TaoBridge - A bridge between Ethereum and Bittensor; Tao Stats - A block explorer for Bittensor; It’s AI a AI detection app built on subnet 32 amongst many others.
Subnets
Bittensor has over 32 live subnets and over 150 in testnet and plans to get to 1024 subnets by the end of the year. Each subnet specializes at a specific application. Due to their variance, it is impossible to cover all of them but some of the most developed ones are:
Subnet 8 (Taoshi - Proprietary Trading Network)
Subnet 8 or Taoshi used to be a time series prediction subnet that would reward miners for accurately predicting the price of Bitcoin in the next 5 minutes. Now the subnet has broadened its focus to become a go-to marketplace for any signal or data point that would help anyone make profitable trades.
The current mechanism requires miners to simulate a trade i.e. open a position with a certain amount of leverage and close it at a future time. The result of this trade is added to the historical returns for this miner. All miners start at a return of 1. This is their initial portfolio value, the amount does not matter, it is just keeping track of returns.
To evaluate the performance of the miner, Taoshi uses the following scores -
Traditional Return - Historical trades have a dampening effect where older losses or wins have a diminishing effect on your current returns score.
Sharpe Ratio - Normalised return over a miner’s lifetime returns. Promotes consistent returns and provides a metric for the risk level of a miner
Omega Ratio - Weight of gains over losses to evaluate risk reward ratio
Validators are a point of access for this data. They can group different miners based on their metrics to create an index of data or provide financial services based on data provided by miners.
Taoshi will soon launch MoSS (marketplace for signals) where signal providers can sell their unique data like weather, traffic, extraction of natural gas, or anything that would help miners make better predictions
The goal is to bring new miners with different ideas. support miners with different time horizons and strategies, and evaluate existing miners based on past performance.
To learn more about the intricacies of the network, I would recommend these two videos: Overview of Taoshi and More in Depth Guide and their Github. You can check out the dashboard for miner returns and their positions.
There are a few applications like Timesless signals, Taopey vaults, and an upcoming Trade Bot that exposes you to Taoshi in a more user-friendly way.
Taoshi just released their Request network which would make subnet data more accessible and open up more business opportunities for Bittensor validators.
Subnet 6 (Nous Research)
Nous is a group of AI researchers with a goal to build open source, accessible, private, localized, and secure AI free from large corporations that require you to send every query to a data center. This video is a good introduction to Nous.
From Nous: “Traditional benchmarking leans heavily on public datasets which can be easy to game and often lead to superficial score improvements that mask true model capabilities (or lack thereof), misdirecting efforts towards meaningless score chasing” Thus Nous research chose to use Bittensor as an incentive mechanism layer with GPT-4 as a benchmark to evaluate open source models. You can see the performance of various miners on this subnet on their dashboard
Nous recently released Hermes Pro which is ranked as the best model on average in the GPT4All benchmark
Nous will soon release Forge which would allow anyone to produce AI in their local machine with their own hardware. You can also use models made publicly available by the community. You can create your AI and export it as a .forge file and share it with the community. The product is planned for June release.
Subnet 3 (My shell)
Myshell is like an App store for AI models and applications. The product is live with text-to-speech models and chatbots with different personas. You can chat with Walter White, Rick, Elon Musk, and many other characters.
Personally, I did not like the product as it is right now. I asked a few AI personas how they worked and they told me it’s just chat GPT 3 and the personas themselves were not good. I asked the “rent a girlfriend” persona, will you be my girlfriend and the output was “As an AI model, I can’t….” I tried “April with her” and it was just an if and else-based scenario game.
Myshell was a web2 company launched in 2020 and now they are dipping their toes in crypto. They recently launched their subnet on Bittensor intending to incentivize the development of open source text-to-speech models. They currently have two live models Open Voice and Melo TTS. You can use this link to try it out.
I tried generating an Indian accent English voice with random text generated from chat GPT and a voice input of Steve Jobs. I didn’t find their output useful, there are many closed-source models and companies with better models.
Despite my not-so-good experience with myshell, I am excited to see what Subnet 3 produces. Myshell has a reputable team from big-name universities like MIT, and Stanford with years of experience in AI and published research papers.
You can learn more about how Subnet 3 works here, get an overview of myshell through this podcast, and checkout their Github for more in-depth details
Subnet 30 (Wombo) & Subnet 24 (Omega)
Wombo is an image generation AI company that created Dream: an IOS app with over 150M downloads on the app store. It is regarded as one of the best image generators and even the Bittensor critic Eric Wall loves Wombo (based on a Twitter space)
Wombo is a team of 20 engineers and they recently registered as subnet 30 on Bittensor. Out of 20 engineers, 5 are working full-time on subnet 30. Currently, the incentive landscape of this subnet is designed to incentivize the generation of reliable, fast, and high-quality images and eventually, it would be to generate any form of media.
They plan to create another incentive system where miners would compete to get your media the most amount of social media engagement. Miners would link their social accounts to their UID and validators would be able to track engagement performances and reward best performing miners.
Wombo has allocated 10% of its subnet earnings to create Bittensor-related content such as videos, memes, tweets, and articles.
Omega or subnet 24 is a data collection subnet incentivizing miners to scrape high-quality YouTube video data that can help generate models that control someone’s computer or help in audio-video synchronization model generation. Miners are using Moon Dream and Lava to describe what’s happening in a video. Most basic miners are just using titles and video descriptions to describe video content. So far they have labeled over 15 million videos!
Omega recently released their MVP version of the data visualization tool to help query the huge dataset they are building and assess the quality of data so people can train models on it.
Omega and Wombo are working closely to create a multi-modal any-to-any model. A model that can take any form of media like text, sight, or audio as input to perform an action, generate text, audio, or any form of output. They plan to create a Mac application with browsing capabilities that would record your screen to learn about your workflow and generate a personalized AI agent for you. The screen-recorded data would be fed to a global AI model in an anonymized fashion to create a super-intelligent global model and users who opt in would be rewarded with TAO. The most basic version of this tool is live right now, based on my attempts it’s not very good at the moment but this is just the beginning!
Subnet Stats
This is what the current emissions to various subnets look like. Subnets 19 & 18 have the highest emissions. Subnet 19 provides high-quality image generation and 18 provides text generation both being used by Corcel. Sn 21 is filetao which is like filecoin and has 91PB or 91,324,617 GB capacity!
If you plan to delegate to a subnet or plan to take part as a miner or validator, you should check out taostats which lists miner registration costs, and delegation rewards, and join the discord.
Competitors
I couldn’t find a direct competitor to Bittensor which has live products and users. Below are some adjacent Crypto x AI projects that are most promising.
Ritual
Ritual is aiming to allow smart contracts to access AI inference where the output generation would occur off-chain and smart contracts would just feed the output to perform various on-chain tasks.
I think at the launch, Ritual could be a bigger deal than Bittensor at least initially due to their ties to EigenLayer, their backers, and the Ethereum ecosystem. It has a lot of headwinds for it if we think in terms of valuation.
Ritual would facilitate financial agents for crypto natives and Bittensor would create consumer applications
Bittensor has a chance to surpass Ritual. I think so because it’s focusing on creating consumer-grade applications. Products on top of it don’t have any crypto element to it at the surface level. No crypto project so far has been able to reach mass adoption to any meaningful level and Bittensor has a chance to do it.
You can run a ritual node to speculate on a potential airdrop. Here is a step-by-step guide on how to set it up. I don’t plan on buying a Ritual token if it launches (might change my mind if their tokenomics are good) because it would likely launch at a high valuation with soonish VC unlocks. I am speculating on getting an airdrop and would just hold the token to add it to my Crypto x AI portfolio.
Morpheus
Morpheus is similar to ritual with a goal to create valuable smart agents that can perform on-chain tasks on the user’s behalf.
Morpheus has a unique tokenomics structure with supply distributed to various stakeholders like capital providers, code contributors, app builders, and infrastructure providers each getting 25% of the emissions. The project is 100% fair launched with no VC funding and more trackable token distribution.
You can track capital providers, their stETH contribution, and the amount of MOR emissions they are receiving on chain. There is a contribution Github page to track MOR emissions to code contributors. Compute contributions are not live yet but you can track them here when they go live. The community page was not working when I tried to access it.
Given the transparency of the project, it is likely to do well and has the potential to surpass Bittensor in FDV (because less than 1/42 of the supply will be tradeable at launch) if its smart agents can perform well on-chain. MOR token would go live on May 8 2024 and it would be interesting to see the market sentiment for it.
There are many AI x Crypto projects launching soon like Gensyn, Nillion, Rivalz, Prime Intellect, and many others that present an interesting opportunity in the Crypto x AI narrative.
Questions and Common Concerns
What happens when the price of TAO goes down or there is high demand for AI output?
When the dollar value of TAO decreases (at some point it must, nothing goes up forever) or subnetworks receive a lot of requests for their output, they would have to do more work for lesser and lesser rewards. This might diminish the quality of AI output.
What about the quality of output for businesses?
Bittensor incentivizes high-quality output, but as mentioned above, there might be periods when the quality of the whole network diminishes just because of price action or high demand. AI in general is non-deterministic but the Bittensor model adds another level of uncertainty. A lot of businesses like Wombo, Corcel, Timeless are using output from Bittensor but the scalability and reliability is still unclear.
A counter-argument to the points above would be -
The Bitcoin hash rate has been steadily going up even during the bear market. More efficient miners have always replaced the less profitable/inefficient miners.
During uncertain times, more serious miners who believe in the Bittensor network might double down to extract as much TAO as possible. Application developers who use subnets would eventually need to pay for output and make deals with validators. Even in the worst-case scenario when TAO isn’t doing well, I think miners will choose to be in the network because, unlike Bitcoin where hardware has no other use case than mining Bitcoin, participants here don’t have anything to lose by participating in the network.
What happens when rewards are good enough to attract closed-source models?
This is a less likely scenario but what if Google, Open AI, or Meta decides to run a miner in each subnet when TAO rewards are profitable enough? These miners would make the other miners irrelevant. We know that closed-source models are 1-2 generations ahead of open-source models and training these models requires a huge corpus of data and computing only accessible to few.
Regarding this point, I think nature follows a power law distribution. Even if Open AI or Google never takes an interest in mining, there might be few miners that grow too big. The good thing is that TAO is distributed across subnets and then gets trickled down to subnet participants. It will be hard for any miner to specialize across many subnets that perform different tasks.
What happens when most of the $TAO is mined?
$TAO emissions are the major source of revenue for miners and validators in the network. If the price of $TAO does not keep going up, there won’t be any incentive for miners to do any work. And nothing goes up forever.
But that does not mean it’s all over. Bitcoin to this very day, runs mostly on $BTC emissions. Miners solve useless hash puzzles and invented new hardware just to get more $BTC. And yes, Bitcoin has somewhat become a store of value but it was not always like that.
Bitcoin, according to its whitepaper, was meant to be a “Peer-to-Peer Electronic Cash System” which to this very day, it’s not at scale. It then became a “payments network” for lightning or Taproot-based assets (never worked out) and now it’s a store of value.
$TAO may become money, a store of value, or an AI agent’s currency only used by subnets to share value or it may evolve itself to have a native fee model and change economic incentives. Nobody knows yet. And these things are hard to predict.
What happens when subnets get too big?
As the subnets on Bittensor grow, they will have a power law distribution. Since subnets have specific rules, miners and validators set would tend to get unique because the competition to earn TAO would get harder and it would get hard for miners to distribute their resources to compete in other subnets. What's stopping this group, since they have already specialized in a task, from opting out of the TAO ecosystem and starting their own chain for higher valuation and own token?
Any subnet can opt out of the Bittensor network similar to how Dydx moved from Starknet. The only reasons for a big subnet to stick to Bittensor are narrative, TAO rewards, and other subnets benefiting the whole ecosystem and community. Since all subnets contribute to the success of Bittensor, their actions might help increase the value of token so subnets might want to collect more TAO rather than bootstrapping a new token from scratch.
What about pre-mined tokens and early holders
For TAO there are many miners and early buyers. Binance listing finally brought more volume to the token allowing these early comers to take some profits. Moreover, 20% of the supply is held by the top 10 wallets and these early miners would have to sell at some point to cover mining costs or daily operations.
According to TaoStats -
There was an initial iteration of the network name Kusangi which was started on the 3rd January 2021 and then halted in the middle of May so that some issues could be addressed. The blockchain and all 546,113 previously mined TAO were migrated to Nakamoto which was started on 2nd November 2021 from block 0. The Finney network network was officially launched on the 20th March 2023.”
This means that people have been mining TAO before subnets existed and based on my discord discussions, there was only one subnet initially which was for text generation which is now SN0. Bittensor was supposed to be built on top of Polkadot and they even bought a parachain slot. Both founders donated 100k TAO each to the foundation to facilitate the parachains on Kusama and Polkadot.
The blockexplorer hosted at bittensor.com only shows transactions for Finney network. Transactions for Kusangi and Nakamoto are hosted at TaoStats
This news was a shock to me and I was considering selling my allocation but after a lot of thought I think this should not be very concerning if we think of the long-term viability of the project and here’s why -
According to estimates , Satoshi mined over 1M BTC which is 4.7% of all BTC supply which has not hampered the ability of the network to grow. ETH had a presale where over 60M ETH or 50% of tokens based on current supply were sold, moreover, the chain was rolled back after the DAO hack. Most new layer 1/2s like Solana and others are VC funded with up to 50% of supply concentrated to few entities.
This somewhat unfair (to some extent) distribution, has not hindered the ability of these projects to attract eyeballs and serious traditional investors. So in my opinion, as long as some value is generated on Bittesnor and independent teams continue to build on it, TAO should have demand for it. However, price swings are to be expected as these tokens get into more hands.
Everyone has already seen these hiccup births and the lifecycle of all the decentralised networks so people also have a higher risk appetite and different expectations.
Future of Bittensor
Bittensor plans to support around 1024 subnets by end of this year and further decentralize it’s network through dynamic TAO.
Dynamic TAO
Dynamic TAO is a proposal on the Bittnesor network to introduce a token and liquidity pool for each subnet. This token would only exist within the liquidity pool and is not tradable. The primary goal is to tie a subnet’s value and the emission it receives to the subnet token. To understand the importance of this proposal, we need to look at how Bittensor currently works.
From const’s post - Bittensor calculates the distribution of TAO such that each block is determined via the ‘Root Network’: a voting pool composed of Bittensor’s top 64 validators (measured by the sum of TAO that has been delegated to them). The more Root Network emissions a subnet receives, the more commodity it can extract, whilst the subnets that do not receive sufficient emission are pruned out.
The current mechanism is centralized and there is no mechanism to allow the free market to dictate the reward distribution of TAO. Dynamic TAO will create a Uniswap V2-like liquidity pool with TAO and subnet token pair. Miners would put in TAO and effectively buy the subnet token to register, they would receive rewards in the subnet token and can take out liquid TAO from the pool at any time. People can also speculate on the success of a subnet with this proposal.
You can learn more about DTAO through a video by Opentensor Foundation
R&D of New AI Models
Many believe that world-class AI models can only be created by tech behemoths. Now I am not an AI researcher but Nearcyan says otherwise.
I have read many articles and tweets criticizing Bittensor for not being able to innovate on any model creation. This couldn’t be further from the truth. As of my knowledge, Bittensor has not yet created any groundbreaking, state-of-the-art model or created a large model like Llama 3. However, it has improved upon and developed many useful models and in the future may yield a state-of-the-art model.
Cerebrus and Opentensor launched a 3B model with 7B performance and it was ranked highest in its category at the time.
Myshell has created Open Voice and Melo TTS text-to-speech models.
Taoshi miners have created many well-performing models, with Tarvis having a 24-0 (winning trades - losing trades) success rate so far.
Nous has developed Hermes 2 which ranks as the best model in GPT4All metric
Nous recently released Hermes 2 pro on Llama 3 which surpasses Llama 3 8B Instruct on AGIEval, GPT4All Suite, TruthfulQA, and BigBench.
Subnet 17 (Three Gen) recently launched a paper for text or image-to-3D object generation called Dream Gaussian
Virtual Protocol (a testnet on Bittensor) launched an Audio to animation model
As competition intensifies with the increasing demand for TAO, I anticipate that highly performant and efficient models will emerge from the Bittensor ecosystem. However, I am skeptical that the ecosystem could produce a model on par with GPT-4 Turbo or any large model soon, given the economic constraints and the current reward structure of the network. If TAO follows a trajectory similar to Bitcoin's, we might see the development of larger models or even specialized hardware, much like ASICs were developed for Bitcoin mining.
Bullish Keypoints
AI x Crypto Narrative
Competent independent teams building on top of it
Fair launch token with most of the supply being staked
Very strong community
Real-world applications are being developed and already live
Dynamic TAO would increase the demand for TAO and add a speculative aspect directly on chain
Bearish Keypoints
There will be a huge sell-off at some point because of miners and early adopters
The viability of the network to be useful for businesses is still unclear
The supply would almost double by next year
Competitors and better alternatives in terms of ROI
Centralization concerns remain until Dynamic TAO is launched and works
Alpha & My Strategy
To make money in the Bittensor ecosystem, you can participate as the following:
Delegator
You can delegate your tokens to a validator for around 15% yield. Based on our research, Roundtable 21 has the most perks at the moment. They do TAO airdrops (from TAO they receive as validator) to anyone staking over 1TAO to their validator. RoundTable 21 is also a partner of Taoshi. In the future, if there ever is a Taoshi token airdrop (highly unlikely), you might qualify for it.
Other than Roundtable 21, we have delegated to Bittensor Guru, Foundry, and Tensorplex to diversify.
Miner
You can run models provided by subnets or improve on them to get juicy 41% subnet yields. If you would like to participate as a miner, have a look at taostats for requirements of various subnets and join their discords. Each subnet has a non-refundable miner registration fee, limited slots, and you must perform well to receive rewards.
Validator
Becoming a Validator is hard. You would need over 10k TAO staked to your validator. However, you don’t need to put these funds on your own. You can create a community through social media platforms to convince others to stake with you. Validators also earn 41% emissions
Subnet Owner
This is probably the toughest role in the ecosystem. You need to write code to create an incentive system for miners and validators. Convince miners and validators to participate in your subnet. Arrange subnet registration fee which is around 1k TAO as of this writing. Subnet owners make 18%.
Application Developer
Anyone can use data generated by subnets on Bittensor. Currently, this service is free for most subnets and is funded by the network itself. You can create valuable services and build a business on top of Bittensor for free which could normally cost hundreds of thousands of dollars.
Potential Ecosystem Airdrops
A thriving ecosystem of DApps is growing around Bittensor. Tensorplex has launched a few DeFi products like an LST service using stTAO and a yield-bearing strategy with Pendle, using those services would be a good way to farm a potential Tensorplex airdrop.
Taopey can potentially do an airdrop to its vault users in Perpy Finance.
Tao Bridge can release an ERC-20 and airdrop to its bridge users
Myshell currently has a points program. One of the team members of Myshell confirmed in a bittensor podcast that they would eventually do an airdrop. Myshell is also becoming a layer 2. AI x Crypto narrative with a layer 2 and open source models can make Myshell airdrop worth a good amount.
Selling TAO
It is important to reap the benefits of your harvest or your crops will wither away, the same applies to crypto. While we are bullish on TAO, we expect to see a huge TAO sell-off at some point in the medium term maybe before the bitcoin cycle tops. Bittensor only has 135k active addresses and most people don’t know about it. Just a few thousand addresses had TAO for around 2 years and these parties have accumulated millions of TAO! They will take profit at some point. I don’t think the recent 50% drawdown of TAO was a significant event given every altcoin saw similar or worse price action.
Currently, the market cap of all of Crypto x AI is less than Doge's market cap! Thus I don’t think it is anywhere near the Crypto x AI top. We expect Crypto x AI to top at least as much as Open AI (90B) and ideally, a few multiples of that given the narrative. Keep in mind Cardano alone topped at 94B last cycle!
Conclusion
Despite some concerns around the centralization and viability of the Bittensor network, I am positive about the project overall because they have consistently attracted top talent and developed real AI models, they have a good community and the first working product in the decentralized AI space. Most early-stage crypto products have some sort of centralization, foundation, centralized set of developers, etc to bootstrap the network which is essential so the project does not lose focus.
If we compare Bittensor to Bitcoin, Bittensor has a much better incentive for miners. Becoming a miner and validator on Bittensor is very lucrative because relatively, emissions at the moment are high, and competition is low. Unlike Bitcoin, the mining equipment required is GPU to create AI models that are general-purpose and useful outside of just mining so miners don’t have a huge risk.
I have staked some TAO and it now represents around 15% of my portfolio. I think the project has the potential to be top 15 projects by marketcap in this cycle because of the narrative around AI and its competitive advantage of being a fairly launched coin. At the moment, I don’t plan on buying more TAO. I think there might be ERC20 token launches by subnets or we can wait for dynamic TAO and invest in subnets. New projects like Morpheus, and Gensyn are on the horizon which have a good chance of gaining AI x Crypto market share.
Unless your time horizon is over 5 years and you have a high conviction in TAO, I would recommend incrementally taking profits since miners and validators are earning a lot of emissions and have incentives to sell.
Resources & References
Bittesnor Wiki - Detailed Resource List
Bittensor Community Discord with Projects
GreyThorn’s overview on Bittensor
Bittensor Community Projects Discord
Const (Founder or Bittensor) Twitter
Mog Machine (Founder of Corcel) Twitter
What is Bittensor - A twitter spaces for beginners