Bittensor subnets: The best opportunities for investing in TAO

Context

In February 2024, the Bittensor protocol underwent one of the most significant upgrades in its history: Dynamic TAO (dTAO). This reform transformed the network's governance by allowing TAO holders to freely choose which subnets they want to delegate their tokens to and participate in voting and reward allocation for the best subnets.

Historically, Bittensor's governance was centralized around the Root Subnet (Subnet 0), with only 64 validators responsible for distributing rewards. This system revealed several limitations: risk of manipulation, concentration of power, barriers to innovation, and a lack of transparency in the evaluation process.

It was within this context that the dTAO update emerged. Now, subnets have their own tokens (called alpha tokens) and compete to attract TAO from investors. This paradigm shift has two major effects:

• It incentivizes teams to build robust projects with clear value propositions and effective communication;
• It gives TAO holders an active role in the network's resource allocation, transforming Bittensor into a decentralized, market-driven incubator.

Each user can now gain exposure to subnets by staking TAO tokens in them and receiving a subnet token in return, whose price fluctuates based on demand. This mechanism turns staking into a speculative bet. But how can one identify the most promising subnets in a sea of often technical, opaque, or very early-stage projects?

→ Read our full presentation of Bittensor (TAO) :

---

How to Evaluate a Subnet?

Since the dTAO update, the number of subnets on Bittensor has exploded. Currently, over 80 subnets are active, with projections exceeding 200 by year-end. With this exponential growth, it's essential to adopt a rigorous framework to evaluate a subnet's quality and potential.

Several criteria can guide this analysis:

1. Project Fundamentals

• Founding team: A public, experienced team with backgrounds in AI, Web3, or reputable institutions is a strong signal. Subnets backed by solid entities (Macrocosmos, Rayon Labs, Inference Labs, etc.) have shown the ability to deliver quickly.
• Value proposition: The product or service should solve a real problem. Projects with a clear, easily understandable mission have an edge in gaining market attention (e.g., deepfake detection, 3D generation, distributed training...).
• Addressable market: Subnets targeting fast-growing verticals (compute, AI agents, AI x DePIN, healthcare, etc.) often show higher potential.

2. Narrative Attractiveness

In the post-dTAO environment, it's no longer just technical validators allocating network resources, but the market itself. This means that storytelling, mindshare, and project clarity are just as critical as the technology—much like the rest of the crypto market.

• Clear narrative: Projects that can explain what they do in one sentence have a strong advantage.
• Presence on X and Discord: Active subnets that share updates, performance data, or integrations naturally attract more stakers.
• Design and UX: A well-designed dashboard, app, or even a visual can spark enthusiasm.

Example: Chutes gained visibility from a single well-shared comparison visual, while Tensorplex saw its token price skyrocket after an investment announcement from CZ's fund, even before the platform went live.

3. Real Traction

If the goal of the network is to foster AI solutions with real-world usage, then user traction and monetization are fundamental indicators:

• Number of users/downloads (e.g., Dippy, 4M+ users)
• Revenue generated (e.g., Celium, $1M+ in 5 months)
• Token utility within a clear economic model (buybacks, usage, etc.)

However, these criteria need nuance. Some projects like Metanova or Safescan are still in the research phase, and their value will become apparent over time. Their contribution to Bittensor lies more in advancing research than in immediate revenue.

4. Other Signals

Additional indicators can indirectly gauge a project's credibility:

• Subnet mindshare within the community (tracked via tools like Swordscan).
• Support from recognized investors or validators (e.g., Yuma, DCG's incubator branch; Crucible Labs, a top validator).
• Attention from major validators (OpenTensor, Mentat Minds...), helping identify trending subnets.
• Emission ranking (based on staker voting), which—if the market functions well—should favor the most promising subnets with higher staker interest and emissions.

For deeper insights, Bittensor’s Discord often offers more valuable information than Twitter or official websites.

→ Read our full analysis of Bittensor's Dynamic TAO (dTAO) update :

---

Subnet Mapping

To bring more clarity to the over 80 subnets now live on the Bittensor network, we've created a Notion board compiled by our contributor Victor. It features detailed entries for each subnet, with short descriptions, key links (website, X, GitHub), application status, and most importantly, a tier-based ranking from S (most promising) to D (least promising). This ranking is updated regularly based on traction, real progress, and community sentiment.

In this section, we present a curated selection of 25 subnets that we currently consider the most promising to follow. This is not financial advice, but a starting point for refining your own research.

• Apex (SN1)

Run by Macrocosmos, Apex is one of the most advanced subnets in AI model inference. It incentivizes miners to build workflows that optimize conversational agents. The app is live with multiple models available.

• Omron (SN2)

A subnet focused on zero-knowledge proof of inference (zkML). Led by Inference Labs, one of the most technically proficient teams. Already integrated into several projects. Massive potential for decentralized AI.

• Templar (SN3)

In a transitional phase, Templar is a pioneer in distributed AI model training. Personally recommended by Const (Bittensor founder), making it worth watching despite a limited web presence.

• Targon (SN4)

Developed by Manifold Labs, Targon aims to make inference faster and cheaper. Outperforms Web2 solutions on some benchmarks and offers an accessible playground with multiple open-source models.

• OpenKaito (SN5)

Backed by Kaito, a major player in InfoFi. Still early, building around text embedding models. Worth watching, especially with upcoming Yaps integrations.

• Infinite Game (SN6)

A creative concept based on event prediction. Building an AI agent (Aion) to guide decision-making on platforms like Polymarket. Strong narrative, execution still to be proven.

• Proprietary Trading Network (SN8)

Developed by Taoshi with a quantitative trading focus. Miners compete to deliver the best trading signals. Glitch Financial app is in beta. Execution looks serious.

• Pre-training (SN9)

Another Macrocosmos subnet focused on foundation model training. Goal: create top-tier open-source pre-trained models for fine-tuning via the network.

• Dippy Roleplay (SN11)

A consumer-facing subnet focused on roleplay AI. High traction: 4M+ users, top 3 in Germany’s App Store, token buybacks. Great product execution and strong commercial strategy.

• Data Universe (SN13)

Also from Macrocosmos, this subnet focuses on dataset collection and sharing. Functional app already scrapes and visualizes data. A key link in the network’s AI pipeline.

• 404 GEN (SN17)

Formerly a Web2 company (Atlas), now focused on 3D content generation via AI. Smooth migration to Bittensor. One-click 3D generation app live. Clear use case in gaming.

• Nineteen (SN19)

First subnet from Rayon Labs. Optimizes inference on models like Llama 3.1. Outperforms Web2 competitors with lower costs and open API. Solid app, methodical approach.

• Protein Folding (SN25)

A scientific project by Macrocosmos simulating protein folding for biomedical research. Over 160,000 folded proteins. University collaborations. High-potential DeSci use case.

• It’s AI (SN32)

Detects AI-generated content. Functional app analyzes documents and text. Accessible narrative, rapidly growing market. Strong potential for mass adoption.

• Bitmind (SN34)

Specialist in deepfake detection. Trusted team with strong community roots. Browser extension already live. Clear value proposition, well-executed.

• Fine-tuning (SN37)

The final component of Macrocosmos’s AI pipeline. Hosts fine-tuning competitions to optimize models for specific use cases. Technical and gamified approach.

• Masa (SN42)

Web3-native project specializing in real-time data scraping (X, Telegram, Discord). Subnet 42 logged over 320M data points—more than X’s own API. Strong use case, integrations in progress.

• Score Vision (SN44)

Applies computer vision to football. Uses match footage for analysis, betting, or AI agent training. Working with CrunchDAO. Unique product.

• Synth (SN50)

Built by Mode Network, Synth uses synthetic data to forecast price movements—currently focused on BTC. Strong narrative, execution underway.

• Celium (SN51)

GPU marketplace on Bittensor. Over $1M in revenue in 5 months. Endorsed by Const and OpenTensor Foundation. A prime example of DePIN: useful and profitable.

• Dojo (SN52)

Tensorplex subnet for human data collection. Platform not yet live, but recent investment from CZ’s fund sent the token soaring. Execution to be monitored.

• Gradients (SN56)

Second Rayon Labs subnet, focused on no-code AI. Lets users train models with zero technical skills. In beta, but UI is very intuitive. Could broaden Bittensor’s user base.

• Chutes (SN64)

Third subnet from Rayon Labs, providing GPUs for general-purpose inference. Over 12M tokens processed daily. Strong integration with Rayon’s AI agent. Solid tech, good storytelling.

• Nova (SN68)

Created by Metanova, focused on drug discovery. Simulates chemical compounds to identify therapeutic molecules. Strong DeSci narrative. Semi-anonymous team but actively building.

• Safescan (SN76)

Recently revived project aiming to detect cancer using AI. Ambitious, still in R&D. Long-term subnet, but with one of the most powerful narratives if successful.

---

How to Invest in a Subnet

The introduction of alpha tokens has turned staking on Bittensor into a genuine speculative market. Each subnet now has its own token, whose price currently fluctuates based on various parameters. This mechanism allows for direct exposure to the perceived performance of the subnet, but also introduces significant volatility and risk.

Today, most subnet tokens are highly illiquid, with low market caps, aggressive emissions, and a high degree of speculation. It’s common to see some tokens gain +300% in a single day, only to drop -90% shortly after. Investing in a subnet means betting on a team, a vision, an execution—but also on its ability to market itself effectively in a competitive and immature environment.

In this context, several staking strategies are available for TAO holders.

1. The Conservative Approach: Staking on the Root Subnet (SN0)

This is the simplest and least risky strategy: staking your TAO on subnet 0 provides a stable yield without exposure to the volatility of subnet tokens. The yield decreases over time, but it remains attractive—especially compared to the risks elsewhere on the network. It’s the default option for those who want to participate in the ecosystem without analyzing each project individually.

2. Delegation via Mentat Minds

For those who want to optimize their yield without dedicating too much time, Mentat Minds offers a simple, non-custodial, and secure solution, made possible by Substrate infrastructure. Three strategies are currently available:

• Optimized Root: staking only on SN0, automatically selecting the best-performing validators.
• Protected Alpha: same principle, but dividends are reinvested in top 15 subnet tokens. The initial capital is never exposed—only the yield is.
• Alpha 5: a more aggressive strategy where 100% of TAO is exposed to subnet tokens from the top 5. High potential returns, but also a risk of capital loss if selected tokens drop.

Mentat Minds allows you to delegate TAO at any time, with no unstaking period and no loss of control over your wallet thanks to the proxy system. It is currently one of the most robust solutions for TAO holders.

To join Mentat Minds, here's an affiliate link proposed by Victor.crypto, feel free to use it to support his work.

3. The Active Approach: Manually Selecting Subnets

This approach consists of analyzing subnets individually to choose where to stake your TAO. It is the most time-consuming method but also potentially the most rewarding—provided you have strong risk management and can read subtle signals effectively.

Since subnet tokens face high selling pressure (due to emissions), only projects capable of generating real traction—or attracting significant community attention—manage to hold up over time. It is crucial to diversify, monitor subnet developments (Discord, X, roadmap), and be ready to adjust allocations.

Some tips if you choose this approach:

• Invest only a small portion of your TAO in the most promising subnets.
• Avoid overly opaque subnets or those whose tokens have already done a 5x without any tangible announcement.
• Favor projects with real user traction or demonstrable short-term utility.
• Stay active on Discord: valuable insights often circulate there well before X or official announcements.