Bittensor (TAO) Bearish Logic: An Income Desert Beneath the Myth of Computing Power

Written by: Pine Analytics

Compiled by: Saoirse, Foresight News

TAO is currently priced at about $275, with a market cap of $2.6 billion and a fully diluted valuation of $5.8 billion. The project has backing from Grayscale (which submitted an ETF listing application to the NYSE in December 2025) and has been publicly endorsed by NVIDIA CEO Jensen Huang. Its token supply narrative is highly attractive: a maximum total of 21 million tokens, with Bitcoin-style halving mechanisms. After the first halving in December 2025, daily issuance drops from 7,200 to 3,600 tokens. Within a year, the number of subnets increases from 32 to 128. Templar’s Covenant-72B training also demonstrates that decentralized compute power can run large language models with benchmark competitiveness.

This report does not deny the facts above. We aim to explore whether the network’s economic model can generate real external revenue supporting its current valuation, and how competitive it is when competing with centralized service providers and self-hosted compute.

Bittensor (TAO) Token Distribution and Allocation

How value flows within the network

Bittensor involves four types of participants:

• Subnet owners build specialized AI markets and receive 18% of TAO issuance rewards;
• Miners perform AI tasks (inference, training, data processing) and receive 41%, totaling about 1,476 tokens daily, with an annual value of approximately $148 million;
• Validators rate miner outputs and receive 41%;
• Stakers lock TAO into subnet liquidity pools in exchange for subnet-specific tokens.

Under the Taoflow model, a subnet’s reward share is determined by net TAO staking inflow; if net inflow is negative, no rewards are issued. The top ten subnets control about 56% of the total issuance.

TAO is a universal token across the network: miners register, validators stake, subnet tokens are purchased, and services are paid using TAO. In theory, subnet activities create structural demand for the underlying token.

Analysis of Chutes (SN64) Subnet and Cost Comparison with Centralized LLaMA 70B Model

Demand Side Status

Transparent supply vs. opaque demand

Bittensor’s supply side is highly transparent: 3,600 TAO are distributed daily according to a hard-coded halving rule, with staking rate (~70%), allocation ratios, and flow data all on-chain.

However, demand is completely opaque. There is no unified dashboard tracking external income by subnet; actual AI service calls (inference, computation, training) occur off-chain and are not recorded on the blockchain. Investors can only infer demand indirectly through staking flows, subnet token prices, and self-reported data from project teams. This opacity is structural, not temporary. The blockchain only records token transfers, not API calls.

Below is the most comprehensive demand-side picture as of March 2026.

Chutes (SN64): Heavily subsidized at low prices

Chutes accounts for 14.4% of total issuance, the highest among all subnets. Developed by Rayon Labs, it offers open-source model inference services without servers, with prices 85% lower than AWS and 10–50% lower than Together AI. Its usage data is dominant within the ecosystem: over 400,000 users (including over 100,000 API users), more than 5 million requests daily, processing a total of 9.1 trillion tokens, with token generation rising from 6.6 billion to 101 billion over three days. It is also a leading inference provider on OpenRouter, with some models outperforming centralized competitors.

But this low price is not due to operational efficiency; it is heavily subsidized.

Based on a 14.4% market share, Chutes earns about 518 TAO daily, with an annual value of roughly $52 million. Its external annual revenue is only about $1.3–2.4 million (the higher figure is self-reported, not independently audited). The subsidy ratio for this subnet is approximately 22:1 to 40:1. For every dollar paid by users, the network must release 22–40 dollars worth of TAO via inflation to subsidize.

Removing subsidies, based on its daily processing volume of about 101 billion tokens, the cost per million tokens is approximately $1.41. In comparison, current centralized market prices are:

• Together.ai’s LLaMA 3.3 70B Turbo: about $0.88 per million tokens;
• DeepSeek V3: approximately $0.40–0.80;
• Smaller models as low as $0.18.

This means that without subsidies, Chutes would be 1.6–3.5 times more expensive than centralized solutions. The so-called 85% cost advantage is completely reversed; its low price is essentially paid for by TAO holders through inflation, not by structural efficiencies of decentralization.

When the next halving occurs (expected late 2026 or 2027), prices will either double, miners will exit, or the subsidy-revenue gap will further widen.

Some compare this to early internet subsidies for customer acquisition, but Uber, DoorDash, and AWS built switching costs during their subsidy periods: proprietary platforms, driver networks, enterprise ecosystems. Bittensor’s subnets have no barriers: models are open-source, interfaces standardized, and users can switch providers at zero cost. Once subsidies end, no lock-in mechanisms can retain users.

Rayon Labs also operates SN56 and SN19, controlling about 23.7% of total issuance, with undisclosed external income. A single team nearly controls a quarter of the network’s incentive distribution.

Targon, Templar, and Other Subnets

Targon (SN4) is the highest revenue subnet, operated by Manifold Labs, providing confidential GPU compute services for enterprises. Estimated annual revenue is about $10.4 million, with a valuation of $48 million, and a price-to-sales ratio of approximately 4.6, making it the most solidly valued within the ecosystem. However, this $10.4 million is a forecast cited by multiple reports, not an audited figure.

Templar (SN3) completed training of Covenant-72B, with a market cap of $98 million, but external income is zero. API training and enterprise sales are ongoing, with no paid products launched yet.

The remaining 120+ subnets either have no public revenue or are still in early product stages, mainly surviving on token issuance subsidies.

Overall Overview

The total confirmed annual external demand-side revenue across the network is only about $3–15 million. Chutes alone’s annual subsidy (~$52 million) exceeds the entire network’s external income upper limit.

Based on a $2.6 billion market cap, the revenue multiple is roughly 175–200x; at a fully diluted valuation of $5.8 billion, nearly 400x. In contrast, centralized AI compute companies’ recent valuations are only 15–25x forward revenue, and high-growth SaaS companies rarely sustain multiples above 50x. Bittensor’s valuation multiple is 4–10 times that of aggressive industry targets.

This huge gap between valuation and demand fundamentals indicates that TAO’s market price is almost entirely driven by supply-side scarcity (halving, staking lock-up), institutional catalysts (Grayscale ETF, exchange listing expectations), and AI sector sentiment, rather than real economic output. These are indeed price drivers, but they are fundamentally different from the idea that “Bittensor as an AI service network creates sustainable value.”

Comparison of Large-Scale Cloud AI Capital Expenditure and Bittensor (TAO) Annual Subsidy Scale

Pricing Dilemma: Two-Sided Pressure

Subnets face dual pressures:

Top: Self-hosted cap

All models on the platform are open-source, weights are public. Running a 70B model on a single H100 costs only about $40–50 daily; tools like vLLM and Ollama make local deployment very easy. NVIDIA’s new chips will further reduce inference costs. For institutions with sufficient volume, self-deployment is cheaper.

Bottom: Cloud giants’ squeeze

Microsoft, Google, Amazon, and Meta will spend over $200 billion on AI in 2025, with priority hardware quotas, dedicated data centers, and enterprise relationships, plus the ability to subsidize AI with cash flows from other businesses. Bittensor’s annual incentive budget (~$360 million) is less than a week’s worth of Microsoft’s AI infrastructure investment. Professional service providers also subsidize low-cost competition with open-source models via VC funding.

Subnets’ pricing is compressed into a very narrow range, while they also bear unique decentralization costs: token friction, validator node expenses, subnet owner shares, network latency, etc.

Moat Issues

Even if a subnet offers valuable services, the underlying models and methods are inherently open: Covenant-72B uses the Apache license, and technical papers are publicly available. Any competitor can directly replicate outside the TAO ecosystem.

Traditional moats (proprietary tech, network effects, switching costs, branding) do not hold:

• Technology is open-source;
• Network effects belong to TAO, not individual subnets;
• Model weights are identical, and user switching costs are zero.

Community believes that the incentive mechanism itself is a moat, but this depends on continuous large token issuance, which halves periodically, shrinking the incentive budget.

What is TAO actually trading?

With a $2.6 billion market cap, TAO’s price does not reflect demand fundamentals; annual revenue of $3–15 million cannot support such valuation under any traditional framework. The market is trading on: Bitcoin-like scarcity, Grayscale ETF expectations, sector rotation in AI, and long-term options on decentralized AI. These are rational speculative factors but are driven entirely by supply-side and market sentiment.

If you hold TAO based on scarcity and narrative, you might profit even if demand weakens; but if you believe Bittensor will become a truly scalable AI service network, there is no evidence currently, and structural barriers are significant. Investors should clearly distinguish their investment logic.