Author: Kaff
Translation: 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 staked inflow) and profits/losses from Alpha tokens (capital gains from subnet performance).
- Since Taoflow in November 2025, subnets with negative net staking inflow will receive zero releases—either go live or be phased out.
- About 3,600 TAO (roughly $960,000) are distributed daily across all subnets, with the top ten controlling approximately 56% of the share.
- Chutes (SN64) exemplifies market fit: 400,000 users, processed 9.1 trillion tokens, costs 85% less than AWS.
- Templar (SN3) is the most asymmetric investment: fully decentralized training of cutting-edge LLMs, with a market cap around $60 million, compared to OpenAI’s $80 billion.
- TAO = index fund exposure to the entire network; Alpha staking = concentrated investment in specific startups—potentially over 100% annual returns but with real risks.
- Alpha tokens carry no formal yield promises; their value depends entirely on market dynamics and team 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 true, that’s not the full picture.
In reality, Bittensor is 128 independent AI startups competing within a harsh economic system, each with its own token, revenue model, and survival efforts. By March 2026, the total market cap of all subnet tokens is about $1.12 billion, roughly 27% of TAO’s own market cap. Grayscale calls it “Y Combinator for decentralized AI.” But unlike traditional funding decisions by a committee, the market decides.
Understanding this mechanism allows you to evaluate which subnets create real value and which are dying out.
Each subnet is an incentive-driven competitive market producing specific digital goods—AI inference, GPU computation, model training, financial data analysis, and more.
Each subnet is driven by three roles: Subnet Owner, Validator, Miner:
Alpha Token: Equity in the subnet
When you stake TAO into a subnet, your TAO enters an on-chain AMM pool (similar to Uniswap V2). In return, you receive Alpha tokens. The price formula is:
Alpha Price = TAO in Pool ÷ Alpha in Pool
Alpha tokens have a hard cap of 21 million (matching TAO’s supply), and they are automatically compounded and released approximately every 72 minutes (“tempo” = 360 blocks).
2. Two Revenue Streams and Why They Are Overlooked
In Bittensor, there are two completely independent ways to make money:
Revenue Source 1: TAO releases via Taoflow
Since November 2025, Bittensor adopted the Taoflow model, a fundamental shift in token distribution.
Previously, releases were based on token price, which created a loophole: projects could artificially inflate the token price to gain more releases, build a “TAO treasury,” then slowly sell while still earning rewards.
Taoflow fixes this by tracking net staking flow: TAO inflow minus unstaking outflow. The mechanism operates in four steps:
After the first halving of TAO on December 14, 2025, block rewards dropped from 1 TAO to 0.5 TAO per block. Currently, about 3,600 TAO (roughly $960,000 at current prices) are distributed daily across 128 subnets. DCG estimates over $100 million annually flowing into this ecosystem.
Revenue Source 2: Alpha Token PnL (Profit and Loss)
This is the part most TAO holders don’t track.
If a subnet performs well, the price of Alpha (denominated in TAO) will rise. When you unstake, you receive more TAO than you initially invested. This is Alpha PnL = capital gains from holding specific subnet tokens.
Taoflow creates a powerful flywheel:
- Great products → more staking of TAO → positive net flow
- Positive flow → more token releases → deeper liquidity pools
- Deeper pools → lower slippage → attract more capital
- More capital → higher Alpha prices → increased Alpha PnL for existing holders
And vice versa, with equal harshness. Continuous negative flow subnets → zero releases → stakers withdraw → death spiral.
3. Who’s Winning and Why?
Snapshot of leading subnets ranked by release dominance and realized PnL:
- SN3 | Templar: Large-scale LLM pretraining | Release share: 30.39% | rPnL: $6.43M
- SN4 | Targon: AI inference marketplace—hosting and providing AI models for real-time predictions | Release share: 10.39% | rPnL: $12.47M
- SN68 | METANOVA: AI drug discovery—developing reprogramming therapies for mind and body | Release share: 5.95% | rPnL: $0.9M
- SN81 | grail: Verifiable post-processing for LLM training | Release share: 4.8% | rPnL: $109K
- SN75 | Hippius: Decentralized storage and network infrastructure with IP management | Release share: 4.56% | rPnL: $4.48M
Top 10 subnets control about 56% of daily total releases.
Case Study: Chutes — a PMF exemplar
Built by Rayon Labs, Chutes is a decentralized serverless AI inference marketplace, a Web3 alternative to OpenAI API and AWS.
Highlights:
- Processed 9.1 trillion tokens since late 2024
- 400,000+ users (including 100,000+ via API)
- Model deployment costs 85% less than AWS
- Supports dozens of models: DeepSeek, Mistral, LLaMA, etc.
- Platform revenue automatically stakes → buyback Alpha → organic demand flywheel
In a February 2026 surge, Chutes attracted over 2,740 TAO in just 9 hours. Alpha peaked at $99.94 (0.225 TAO), FDV hit 205K TAO (~$51.8M at peak TAO price).
Rayon Labs also runs SN56 (Gradients—model training) and SN19 (Nineteen—high-frequency inference), which together accounted for over 23% of total releases at peak.
Case Study: Templar (SN3)—most asymmetric reward bet in subnet ecosystem
On March 10, 2026, Templar (SN3) completed Covenant-72B, a model with 72 billion parameters, claimed as the largest decentralized pretraining run in history.
4. Underlying Mechanics: Registration, Yuma Consensus, and Competition
Registration: a competitive arena
Not everyone can open a subnet. Registration uses a dynamic burn pricing mechanism: each new subnet registration doubles the cost; when no new subnets register, the cost linearly halves over 28,800 blocks (~4 days).
When all 128 slots are filled, new subnets must replace the worst-performing existing ones (measured by lowest EMA price). New registrants get a 4-month grace period before potential deregistration. The network is expected to expand to 256 subnets by 2026.
Yuma Consensus: automated independent auditing
Within each subnet, Yuma consensus converts validators’ subjective assessments into objective reward distribution:
- Validators submit weight vectors, scoring each miner they evaluate
- The blockchain computes the median of these weights (kappa=0.5)
- Weights above median are clipped—preventing collusion and overestimation
- Validators use a commit-reveal scheme—submitting sealed weights, revealing after a set number of blocks—to prevent copying
- Validators who identify high-quality miners early and maintain consistent assessments build stronger bonds and earn larger dividends
Result: no single subnet owner can unilaterally change who gets rewarded. This is a fundamental difference from typical “AI projects” in crypto.
5. Investment Framework: TAO = index fund, Alpha staking = startup bet
Auditless Research succinctly summarizes: “TAO is more like an options token—a call option—pointing to any undervalued Alpha token release.”
Interpreting subnets like a balance sheet
- Release volume = protocol subsidy – network revenue, akin to government grants or accelerator funding
- Alpha PnL = market cap signal—market pricing the true value of the subnet
- Net staking flow = revenue growth indicator—positive flow = “product selling well,” negative flow = customer churn
- Subnet owner = founder quality—communication frequency, delivery speed, product roadmap to monitor
- Validator count = board quality—more independent validators reduce the risk of score manipulation
Institutional signals are strengthening
This is no longer just retail:
- DCG holds over 500,000 TAO (~2.4% of total supply)
- Polychain Capital holds about $200M in TAO
- Grayscale GTAO Trust listed on NYSE on Jan 6, 2026
- Stillcore Capital (co-founded by Jason Calacanis) launched a dedicated subnet token fund
6. Conclusion: Entry Framework
Bittensor has created a unique structure in crypto: 128 AI companies competing, sharing about 3,600 TAO daily (~$960K), with capital allocation entirely driven by stakers’ actions.
When evaluating a subnet, ask these five questions:
- Product: What does this subnet offer? Is there real demand?
- Flow: Is net staking flow positive or negative? What’s the 30-day trend?
- Team: Are the subnet owners actively communicating and delivering?
- Flywheel: Does revenue organically create demand for Alpha, or is it purely speculative?
- Exit: Is the liquidity pool large enough to exit without severe slippage?
TAO gives you broad exposure to the entire ecosystem. Alpha staking offers concentrated bets on specific “startups”—with all the upside and downside risks that entails.
