Following institutional backing and price increases, let's re-examine the true value of Bittensor's 128 subnets.

Author: Kaff

Compiled by: Yuliya, PANews

TL;DR

• Bittensor consists of 128 independent subnets, each operating like a startup with its own token (Alpha), revenue model, and team.

• There are two ways to make money : TAO emissions (protocol subsidies based on staking inflows) and Alpha token gains and losses (capital gains from subnet performance).

• Since Taoflow in November 2025, subnets with negative net staking inflows will receive zero releases and will either be launched or decommissioned.

• Approximately 3,600 TAO (about $960,000) are allocated daily across all subnets, with the top ten controlling about 56% of the share.

• Chutes (SN64) is the most obvious example of market fit: it has 400,000 users, has processed 9.1 trillion tokens, and costs 85% less than AWS.

• Templar (SN3) is the most asymmetrical investment: a fully decentralized training of cutting-edge LLMs with a market value of approximately $60 million, while OpenAI's market value is as high as $800 billion.

• TAO = Exposure to index funds across the entire network, while Alpha staking = investments concentrated in specific startups – with potential annual returns of over 100%, but also accompanied by real risk.

• Alpha tokens do not come with any formal guarantee of returns; their value depends entirely on market dynamics and the team's execution.

1. Subnet structure: Who does what?

When people think of Bittensor, the most common mental model is: it's a decentralized AI project. While that's true, it's not comprehensive enough.

In reality, Bittensor comprises 128 independent AI startups competing in a brutal economic system, each with its own token, revenue model, and struggle to survive. By March 2026, the total market capitalization of all subnet tokens was approximately $1.12 billion, equivalent to 27% of TAO's own market capitalization. Grayscale calls it the "Y Combinator of decentralized AI." However, unlike a committee that decides who receives funding, the market decides.

Once this mechanism is understood, it becomes possible to assess which subnets are creating real value and which are declining.

Each subnet is an incentive-based competitive market that produces specific digital goods—a variety of services such as AI inference, GPU computing, model training, and financial data analysis.

Each subnet consists of three roles: subnet owner, validator, and miner driver.

Alpha Token: Stake in a Subnet

When you stake TAO on a subnet, your TAO will enter an on-chain AMM pool (similar to the mechanism in Uniswap V2). In return, you will receive Alpha tokens. The price formula is:

The hard cap for Alpha tokens is 21 million (corresponding to the supply of TAO), and they are automatically released with compound interest every approximately every 72 minutes (one "tempo" = 360 blocks).

2. Two sources of income and the reasons for their neglect

In Bittensor, there are two completely independent ways to make money:

Revenue Source 1: Obtaining TAO releases through Taoflow

Since November 2025, Bittensor has adopted the Taoflow model, which represents a fundamental shift in how tokens are released and distributed.

Previously, releases were calculated based on the token price. This created a loophole: project teams could artificially inflate the token price to secure releases, creating a "TAO vault," and then slowly dump the tokens while still collecting rewards.

Taoflow addresses this issue by tracking the net staking flow of TAOs: the amount of TAOs staking inflow minus the amount of TAOs unstaking outflow. This mechanism operates in four steps:

Following the first TAO halving on December 14, 2025, the block reward will decrease from 1 TAO to 0.5 TAO per block. Currently, approximately 3,600 TAO (about $960,000 at current prices) are allocated daily to the 128 subnets. DCG estimates that over $100 million flows into the ecosystem annually.

Revenue Source 2: Alpha Token Profit and Loss (PnL)

This is the part that most TAO holders don't track.

If a subnet performs well, the price of Alpha (denominated in TAO) will rise. When you unstake, you will receive more TAO than you initially invested. This is Alpha profit/loss = capital gain from holding tokens from a specific subnet.

Taoflow creates a powerful flywheel effect:

• Excellent products → More people pledge TAO → Positive net inflow

• Positive flow of liquidity → More tokens released → Deeper liquidity pools

• Deeper liquidity → lower slippage → attracting more capital

• More capital → higher Alpha prices → increased Alpha profits and losses for existing holders

Conversely, the same applies, and it's equally brutal. Subnets with consistently negative traffic → zero release → stakers continuously withdrawing funds → death spiral.

3. Which subnets are winning and why?

Below is a snapshot of the leading subnets ranked by released dominance and realized profit/loss (rPnL):

• SN3 | templar : Large-scale LLM pre-training | Release share: 30.39% | rPnL: $6.43 million

• SN4 | Targon : AI Inference Market – Hosting and Providing AI Models for Real-Time Prediction | Released Share: 10.39% | rPnL: $12.47 Million

• SN68 | METANOVA : AI-powered drug discovery company developing therapies to reprogram the mind and body | Share released: 5.95% | rPnL: $900,000

• SN81 | grail : Validated LLM training post-processing | Release share: 4.8% | rPnL: $109,000

• SN75 | Hippius : Decentralized storage and networking infrastructure with IP management | Released share: 4.56% | rPnL: $4.48 million

The top 10 subnets control approximately 56% of the total daily release.

Case Study: Chutes , a Model of Product-Market Fit (PMF)

Chutes, built by Rayon Labs, is a decentralized serverless AI inference marketplace and a Web3 alternative to the OpenAI API and AWS for model deployment.

Chutes' outstanding feature is:

• 9.1 trillion tokens have been processed since the end of 2024.

• It has 400,000+ users (of which 100,000+ access via API).

• AI model deployment costs are 85% cheaper than AWS.

• Supported models: DeepSeek, Mistral, LLaMA, and dozens of others.

• Platform revenue is automatically staked → Alpha tokens are bought back → Organic demand flywheel

During a surge in February 2026, Chutes attracted over 2,740 TAOs in just 9 hours. The Alpha token reached a high of $99.94 (0.225 TAOs), with a FDV of 2.05 million TAOs (approximately $518 million at the peak of TAO's price).

Rayon Labs also operates SN56 (Gradients - model training) and SN19 (Nineteen - high-frequency inference), which together accounted for more than 23% of the total releases at peak times.

Case Study: Templar (SN3), the Bet with the Greatest Asymmetric Reward Potential in a Subnet Ecosystem

On March 10, 2026, Templar (SN3) completed Covenant-72B, a model with 72 billion parameters, which is known as the largest decentralized pre-training run in history.

4. Underlying Mechanisms: Registration, Yuma Consensus, and Competitive Pressure

Registration, a competitive arena

Not everyone can open a subnet. Registration uses a dynamic burn pricing mechanism: the cost doubles every time a new subnet is registered; and when no new subnets are registered, the cost is linearly halved over 28,800 blocks (approximately 4 days).

When all 128 slots are full, a new subnet must replace the worst-performing existing subnet (measured by the lowest EMA price). Newly registered subnets will receive a four-month grace period before being eligible for deregistration. The network is projected to expand to 256 subnets by 2026.

Yuma consensus, automated independent auditing

Within each subnet, the Yuma consensus transforms the validators' subjective evaluations into objective reward distributions:

• Validators submit weight vectors, scoring each miner they have evaluated.

• The blockchain calculates the median (kappa=0.5) for each miner based on their staking weight.

• Weights above the median will be pruned—to prevent collusion and overvaluation.

• Validators use a commit-reveal mechanism—the weights are committed and sealed, and then revealed only after a set number of blocks—to prevent weight duplication.

• Validators who can identify high-quality miners early and maintain consistent evaluations will build stronger binding positions and earn a larger share of bonuses.

The result is that no subnet owner can unilaterally change who receives the reward. This is the fundamental difference between Bittensor and typical "AI projects" in the crypto space.

5. Investment Framework: TAO = Index Fund, Alpha Staking = Betting on Startups

Auditless Research aptly summarized this: "TAO is actually more like an option token—a call option that can point to the release of any seemingly undervalued Alpha token."

How to interpret a subnet like looking at a balance sheet

• Release amount = Protocol subsidies - Network revenue, similar to government grants or accelerator funding.

• Alpha Profit/Loss = Market Value Signal - The market is pricing in the true value of this subnet.

• Net pledged flow = Revenue growth indicator - Positive flow = "Product selling well", Negative flow = Lost customers

• Subnet Ownership = Founder Qualities - Communication Frequency, Delivery Speed, and Product Roadmap are the key metrics to track.

• Number of validators = Board quality - More independent validators = Less likely the score is to be manipulated

The signal from the institution is increasing.

This is no longer just a story about retail investors:

• DCG holds more than 500,000 TAO tokens (approximately 2.4% of the total supply).

• Polychain Capital holds approximately $200 million in TAO.

• The Grayscale GTAO Trust was listed on the New York Stock Exchange (NYSE) on January 6, 2026.

• Stillcore Capital (co-founded by Jason Calacanis) has launched a fund specifically for subnet tokens.

6. Conclusion: Introductory Framework

Bittensor has created a unique structure in the crypto space: 128 AI companies compete with each other, sharing approximately 3,600 TAO tokens (about $960,000) per day, with capital allocation entirely determined by the behavior of stakers.

When evaluating a subnet, ask these five questions:

• Product: What does this subnet provide? Is there a real need for it?

• Traffic: Is the net pledged traffic positive or negative? What is the trend over 30 days?

• Team: Is the subnet owner maintaining ongoing communication and delivery?

• Flywheel: Does revenue create organic demand for Alpha, or is it pure speculation?

• Exit: Is the liquidity pool large enough to allow exits without significant slippage?

TAO offers you broad exposure to the entire ecosystem. Alpha staking, on the other hand, provides you with concentrated bets on specific “startups”—along with all the upside potential and downside risks that come with them.