Ambitious AI Lego, making algorithms composable

Market trend shifts, with multiple sectors reactivating.

Aside from the Bitcoin ecosystem under the spotlight, the AI track has been a continuous hotspot this year, serving as a stage for frequent “shitcoins.”

Beyond tokens like FET, RNDR, and OCEAN that are currently being hyped, a token called TAO has surged threefold in the past month. Its underlying project, Bittensor, has rarely been deeply analyzed in the Chinese-speaking market.

Meanwhile, the other side reacts much faster than us.

The rapid price surge has also alerted keen investors to the opportunity. On Thursday, the Bittensor community announced that well-known crypto VCs Pantera and Collab Currency have become TAO holders and will provide more support for the project’s ecosystem development.

Venture capitalists are adept at capturing evolving trends and driving their development.

What makes TAO, which is favored and rapidly rising in price, so extraordinary? What are the notable differences in its narrative, product, and token economics compared to mainstream AI projects?

In this issue, we will conduct a comprehensive analysis of Bittensor, including its industry background, project goals, technical composition, and token valuation, to assist your judgment and decision-making.

But first, understand the investment logic of Crypto + AI Any token’s rise is supported by fundamental investment logic and broader industry narratives. Before studying TAO, it’s helpful to review the overall AI industry landscape.

AI Boom Amid Bond Bubble Tokens related to AI concepts are very hot, but the absence of crypto doesn’t diminish AI’s independent popularity.

Data from CB Insights shows that in 2023, interest in generative AI has grown significantly, with total funding for AI-related companies and projects skyrocketing to $14 billion; last year, this figure was only $2.5 billion.

Source: CB INSIGHTS

Therefore, whether it’s TAO, RNDR, or FET, the underlying driving forces are far more complex than just ChatGPT or Nvidia’s hype.

Industry veteran Arthur Hayes recently outlined a possible or ongoing scenario — a collective AI funding boom driven by a bond bubble.

Estimates suggest that, led by the US, the total government debt that will need rollover and issuance over the next three years due to fiscal deficits could reach $33.58 trillion.

Government bonds are issued with promises to pay back principal and interest at maturity. If bond yields are high, it means most funds go into buying government bonds, absorbing capital from the private sector (or government public sector), which inevitably squeezes other investment and financing opportunities, such as funding for private enterprises or a sluggish stock market.

Thus, Arthur believes the Federal Reserve will inevitably print money to buy its own issued debt, reducing the impact on the private sector; this is expected to lead to a significant increase in the global money supply by 2026 — even surpassing the COVID period.

Where will this excess money flow?

“Money will flow into new tech companies promising crazy returns when mature. Every fiat liquidity bubble has a new form of technology to attract investors and large capital.”

In the 1990s, it was the internet bubble; after the 2008 financial crisis, it was online advertising and social media; and this time, it’s AI.

This may also be one of the deeper reasons why generative AI has attracted so much investment this year. The technology behind GPT is well-known, but from a broader perspective, it’s just the brightest gem in the flood of capital, with a clear trend of collective capital inflow into AI.

Crypto + AI, Narrative Segmentation Money is flowing in, so the next question is: what to invest in? Let’s further examine the investment logic of Crypto + AI.

It’s a common saying that AI essentially is an advanced productivity tool, whose rapid development depends on three core elements: data, algorithms, and computing power; cryptocurrencies and blockchain are more about production relations, promoting changes through incentives, coordination, and organizational forms affecting these three elements.

Tokens that can enhance these three elements have the potential to generate hype.

Without discussing feasibility for now, in previous projects, we’ve seen two main narratives:

• Crypto + Data: AI requires massive amounts of data for training. Blockchain can incentivize data providers to contribute data or utilize decentralized storage to support democratized, decentralized data training needs.

In this narrative, benefiting cryptocurrencies could be decentralized storage infrastructure like Filecoin, which Arthur strongly recommends.

• Crypto + Computing Power: AI models require powerful computation. Major tech firms or some resource providers have this capability, but there’s also room for the long tail market—dispersed computing resources (personal GPUs/devices) can contribute computing power for incentives in crypto.

In this narrative, cryptocurrencies like RNDR and other projects that contribute computing power benefit.

As for algorithms, another logic applies:

• Crypto + Algorithms: Unlike resource-intensive data and compute, algorithms are highly technical and are the secret sauce and barrier for AI companies’ iterative development. It’s difficult to create better algorithms from scratch solely through crypto incentives; contribution, coordination, and incentive logic don’t work well here.

(Note: An AI model is the result of algorithm training; strictly speaking, algorithms and models have a sequential relationship. But for simplicity, the author mixes the two in the following descriptions.)

However, you can incentivize the selection of better algorithms from existing ones, preventing everyone from using the same solution. Similar to oracle projects that incentivize competition to select better data sources.

Currently, there are no particularly prominent projects in this sub-narrative, but Bittensor is one of them — it neither directly contributes data nor compute power but uses blockchain networks and incentive mechanisms to schedule and select among different algorithms, creating a free-market, knowledge-sharing ecosystem for AI models.

Understanding Bittensor’s Narrative: AI Lego, Making Algorithms Composable Sounds complicated?

For easier understanding, a rough summary of Bittensor: We don’t produce algorithms; we are just carriers of high-quality algorithms.

Why carry algorithms? The current AI ecosystem reveals the problem.

Players in the AI track have isolated algorithms and models. Due to commercial competition, you can’t let different companies’ algorithms learn from each other to improve; this means, from the supply side, competition is zero-sum: if one AI wins the market, others are out.

Source: Bittensor official website

This is fine for the winners.

But Bittensor believes this is detrimental to overall AI progress and algorithm innovation efficiency. Isolated models and AI services that only pick winners mean that developing new models often requires starting from scratch.

Suppose Model A is proficient in Spanish, and Model B in coding. When a user needs AI to explain code with Spanish comments, the best output would come from combining both algorithms, but currently, that’s not feasible.

Additionally, since third-party app integration requires AI model owners’ permission, limited functionality also means limited value, and the collective potential of AI isn’t fully unleashed.

Therefore, Bittensor’s main goal is to enable different AI algorithms and models to collaborate, learn, and combine, forming more powerful models that better serve developers and users.

This idea and approach resemble what we saw a few years ago during DeFi Summer — financial Lego.

Financial components like stablecoins, lending, and liquidity mining are open-source and permissionless, allowing users to combine them freely, like Lego blocks, to create new products and services.

Similarly, AI algorithms specializing in image, text, or audio processing can be combined to serve different tasks, forming AI Lego.

Thus, for Bittensor, the project itself doesn’t perform computation or provide data for on-chain machine learning but mobilizes off-chain AI models to collaborate.

Theoretically, by assembling AI Lego blocks, Bittensor can expand its AI capabilities faster and more efficiently than isolated models.

But whether AI model providers will buy into this, how to do business expansion, and whether it can be practically implemented remain to be seen.

Mining and Incentives for AI Model “Oracles” Enabling different AIs to collaborate is a big goal, but how to achieve it?

Bittensor’s answer is to build a blockchain network that coordinates and operates through mining incentives.

It adopts Polkadot’s parachain (application chain) design, essentially having its own chain dedicated to AI model collaboration, with its own token $TAO for incentives.

To understand how this chain operates, at least three questions need to be clarified:

1. What roles are on this chain?

2. What do these roles do? How are they connected?

3. Which behaviors of these roles are incentivized by the token?

Roles and functions on the chain:

• Miners: Can be understood as providers of various AI algorithms and models worldwide. They host AI models and offer them to the Bittensor network; different types of models form subnets, e.g., models specialized in images or sound.

• Validators: Evaluators within the Bittensor network. They assess the quality and effectiveness of AI models, rank them based on performance for specific tasks, and help consumers find the best solutions.

(Note: Currently, validators seem to be affiliated with the project’s own institutions, which may lack full decentralization. As the network develops, other organizations might be involved as validators.)

• Nominees: Token holders who delegate their TAO to specific validators to show support, or switch support to different validators. Similar to staking tokens in DeFi to earn rewards.

• Users: Final users of AI models provided by Bittensor. They can be individuals or developers seeking AI models for applications.

Connections among roles:

Users need better AI models; validators filter and select better models for different purposes; miners provide their models; nominees choose which validators to support.

In essence, it’s an open AI supply and demand chain: some provide models, others evaluate, and users get results from the best models.

Source: ReveloIntel

The diagram above illustrates this simply: users input their needs, validators route these needs to miners in the Bittensor network; miners produce answers, validators evaluate quality, and finally, results are returned to users.

What does the TAO token incentivize?

• Validators: The more accurately and consistently they evaluate and select AI models, the more rewards they earn. To be validators, they need to stake a certain amount of TAO tokens.

• Miners: Respond to user needs by providing their models and earn TAO tokens based on contribution.

• Nominees: Delegate their TAO to validators, similar to liquidity staking rewards.

• Users: Pay TAO tokens to initiate tasks, effectively making purchases.

Ideally, different AI models in this network will collaborate, and different tasks will likely perform better with different models; since these tasks are visible on-chain, models can learn from each other and adapt accordingly.

Source: ReveloIntel

A better analogy is: Bittensor is like an AI “oracle.” In DeFi, oracles feed the best prices to applications; Bittensor feeds the best models to AI demanders.

Participation as validator or miner involves technical coding and interfaces, which are not detailed here. Interested readers can visit the official documentation for more.

$TAO Token: How to Value It? Token Economics

According to official documents, Bittensor launched in 2021 with a “fair launch” (no pre-mined tokens), called TAO.

TAO’s total supply is 21 million (a nod to BTC), with a four-year halving cycle, rewarding every 1.05 million blocks, with halving every block reward. There will be 64 halving events in total, with the latest halving cycle occurring in August 2025.

It’s somewhat sci-fi that, based on this halving schedule, it would take 256 years for all tokens to be mined.

Currently, a TAO is issued every 12 seconds. Roughly, 7,200 TAO are produced daily, split equally between miners and validators.

The fair launch means no VC rounds, private sales, ICO/IEO/IDO, or foundation reserves — purely a mining token.

Each reward cycle, TAO is distributed among validators and miners.

However, notable institutional investors like DCG, GSR, Polychain, and Firstmask are also involved.

A reasonable inference is that, since most validators are affiliated with Bittensor’s official institutions, mined tokens can be retained and then distributed to market makers for liquidity.

These large institutions can also act as validators or miners, mining TAO.

As mentioned at the start, crypto VCs like Pantera have recently become TAO holders. So, Bittensor is a fair launch, but not entirely free of VC involvement.

In this new market cycle, the “VC sell to secondary” token issuance model is less favored. TAO’s “fair start, then attract capital” approach has, objectively, been as fair as possible.

Market Performance and Valuation

Looking solely at TAO’s market performance, its price has increased over five times from this year’s lows.

But other AI projects have also performed well. For example, RNDR has roughly quintupled since the start of the year.

Therefore, analyzing token value solely based on absolute price increase isn’t very meaningful.

Compared to other popular AI projects, TAO’s market cap is second only to RNDR. Due to its four-year halving release schedule, its market cap to fully diluted valuation ratio is among the lowest, indicating relatively low circulating supply but high per-unit price.

Original chart: User @Moomsxxx, TAO price as of press time calculated by the author.

Low circulating supply sometimes makes it easier to push prices higher. Assuming the price remains stable (around $160)), with daily issuance of 7,200 TAO tokens, total market sell pressure would be about $1.15 million. Given current market activity and trading volume (TAO’s daily volume is around $5 million), absorbing this sell pressure is feasible.

Looking beyond TAO itself, token valuation makes more sense when compared to similar existing projects.

As previously discussed, Bittensor’s focus on crypto + algorithms/models means it cannot be directly compared to projects like RNDR, which provide foundational compute power.

According to Nansen’s AI track research report, Bittensor falls into the “Model Training” category, with competitors like Gensyn and Together, the former supported by a16z.

However, neither of these projects currently has a publicly available token, so comparing TAO’s market cap with theirs isn’t feasible.

Source: Nansen Research

Omnichain Capital co-founder David Attermann proposed a more aggressive comparison in a blog in May — directly benchmarking Bittensor against OpenAI.

Interestingly, David explicitly stated he does not hold TAO tokens, to keep his analysis objective.

Since both core businesses involve training models for user deployment — one is a closed-source company, the other coordinates global AI models — both aim to improve AI usability.

Considering OpenAI’s previous private market valuation from Microsoft (close to $30 billion), and TAO’s current FDV around $3.6 billion, TAO still has roughly 8x upside potential.

I don’t fully agree with this valuation comparison method. The fundamentals, growth pace, and market focus of Web3 and Web2 projects differ significantly. Using valuation multiples as the sole basis for potential upside is only a reference; much depends on TAO’s own positive factors and capital market sentiment.

Conclusion In summary, TAO/Bittensor offers an alternative narrative outside the well-known AI-themed crypto projects: it doesn’t involve directly producing productivity resources (computing or data), but relies purely on the mobilization of production relations to enable AI models to collaborate, compete, and optimize.

This narrative has some appeal, but key factors like AI model integration, validator centralization, and quality assessment are not easily solved by a white paper — AI itself is simple, but commercial games are complex. Convincing more participants to join with token rewards, persuading tech companies to collaborate with other AI models, remains a matter of perspective.

Beyond fundamentals, the token’s price increase indicates the market’s collective enthusiasm for AI. Given that Bittensor currently has no comparable competitors in its niche, TAO could benefit from the broader AI hype. However, due to the lack of suitable valuation benchmarks, whether it’s worth holding long-term remains uncertain.

Monitoring project updates and sudden trading volume changes may be a more practical approach.