Bittensor (TAO) Bearish Thesis: A Revenue Desert Under the Computing Power Myth

Original Compilation: Saoirse, Foresight News

TAO’s current price is approximately $275, with a market capitalization of $2.6 billion and a fully diluted valuation of $5.8 billion. The project has received institutional backing from Grayscale (which filed an application for a NYSE-listed ETF in December 2025) and public recognition from NVIDIA CEO Jensen Huang. Its token supply narrative is also highly attractive: a total cap of 21 million tokens with a Bitcoin-style halving mechanism. After the first halving in December 2025, the daily issuance dropped from 7,200 to 3,600 tokens. The number of subnets grew from 32 to 128 within a year, and Templar’s Covenant-72B training has proven that decentralized computing power can produce large language models with baseline competitiveness.

This report does not deny the above facts. What we aim to explore is: can the network’s economic model generate real external revenue sufficient to support its current valuation scale, and what is its actual competitiveness when pitted against centralized service providers and self-hosted computing power?

Bittensor (TAO) Token Issuance Allocation Ratio

How Network Value Flows

Bittensor has four types of participants:

• Subnet owners build specialized AI pasars and receive 18% of the subnet’s TAO issuance rewards;
• Miners perform AI tasks (inference, training, data processing) and receive 41%, totaling approximately 1,476 TAO daily, with an annualized value of about $148 million;
• Validators score miner outputs and receive 41%;
• Stakers deposit TAO into subnet liquidity pools to obtain subnet-specific tokens.

Under the Taoflow model, a subnet’s reward share is determined by the net inflow of TAO staked; negative net inflow results in no rewards. The top ten subnets control approximately 56% of the network’s total issuance.

TAO is the universal network token: it is required for miner registration, validator staking, purchasing subnet tokens, and service payments. Theoretically, subnet activity should create structural demand for the underlying token.

Comparative Analysis of Bittensor Subnet Chutes (SN64) vs. Centralized Service Provider LLaMA 70B Model Inference Costs

Current State of Demand

Transparent Supply vs. Opaque Demand

Bittensor’s supply side is highly transparent: 3,600 TAO are distributed programmatically each day, the halving rules are hard-coded, and staking rates (~70%), allocation ratios, and liquidity data are all on-chain.

However, the demand side is completely opaque. There is no unified dashboard tracking external revenue by subnet. Actual usage of AI services (inference, computation, training) occurs off-chain and is not recorded on the blockchain. Investors can only infer demand through indirect indicators like staking flow, subnet token prices, and self-reported data from projects. This opacity is structural, not temporary. The blockchain records token transfers, not API calls.

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

Chutes (SN64): Low Prices Rely Entirely on Subsidies

Chutes commands 14.4% of the network’s total issuance, the highest among all subnets. Developed by Rayon Labs, it provides serverless inference services for open-source models, quoting prices 85% lower than AWS and 10%–50% lower than Together AI. Its usage data is unparalleled within the ecosystem: over 400,000 users (over 100,000 API users), more than 5 million daily requests, cumulative processing of 9.1 trillion tokens, and a three-day average token generation soaring from 6.6 billion to 101 billion. It is also a leading inference provider on OpenRouter, with some models outperforming centralized competitors.

However, this low price does not stem from operational efficiency but from subsidies.

Based on its 14.4% share, Chutes receives approximately 518 TAO daily, with an annualized value of about $52 million. Its external annual revenue is only about $1.3–2.4 million (the higher figure is self-reported by the team, unaudited). The protocol’s subsidy ratio for this subnet is approximately 22:1 to 40:1. For every $1 paid by users, the network must release $22–40 worth of TAO through inflation as a subsidy.

Removing the subsidy and reverse-calculating based on its daily processing of about 101 billion tokens, the cost price would be approximately $1.41 per million tokens. Current centralized market prices are:

• Together.ai’s LLaMA 3.3 70B Turbo: ~$0.88 / million tokens;
• DeepSeek V3: ~$0.40–0.80;
• Smaller models can go as low as $0.18.

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

When the next halving arrives (expected late 2026 or 2027), either prices will double, miners will leave, or the gap between subsidies and revenue will widen further.

Some may draw parallels to early internet subsidies for user acquisition, but companies like Uber, DoorDash, and AWS built switching costs during their subsidy phases: proprietary platforms, driver networks, enterprise ecosystems. Bittensor subnets have no such barriers: models are open-source, interfaces are standardized, and users can switch providers at zero cost. Once subsidies recede, there are no lock-in mechanisms to retain users.

Rayon Labs also operates SN56 and SN19, collectively controlling about 23.7% of the network’s total issuance. Neither has disclosed external revenue. A single team effectively controls nearly a quarter of the network’s incentive distribution.

Targon, Templar, and Other Subnets

Targon (SN4) is the highest-revenue subnet, operated by Manifold Labs, offering confidential GPU computing services to enterprises. Estimated annual revenue is about $10.4 million, corresponding to a valuation of $48 million, with a price-to-sales ratio of about 4.6x, making it the most solid valuation within the ecosystem. However, the $10.4 million figure is a forecast cited by multiple reports, not an audited number.

Templar (SN3) completed the Covenant-72B training, with a market cap of $98 million, but has zero external revenue. Its training API and enterprise sales are still in progress, with no paid product launched yet.

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

Overall Picture

The total confirmed annual external revenue across the entire network is only about $3–15 million. The annualized subsidy for Chutes alone (~$52 million) exceeds the upper bound of the entire network’s external revenue.

Based on a $2.6 billion market cap, its revenue multiple is approximately 175–200x; based on the $5.8 billion fully diluted valuation, it’s close to 400x. In contrast, centralized AI compute companies have recently raised at valuations of only 15–25x forward revenue, and high-growth SaaS companies rarely sustain multiples above 50x long-term. Bittensor’s valuation multiples are 4–10 times higher than those of aggressive industry benchmarks.

The vast gap between valuation and demand fundamentals indicates that the market prices TAO almost entirely based on 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 entirely separate from the logic of “Bittensor as an AI service network creating sustainable value.”

Comparison of Hyperscale Cloud Provider AI Capital Expenditure vs. Bittensor (TAO) Annual Subsidy Scale

Pricing Dilemma: Squeezed from Both Sides

Subnets face pressure from both ends:

• Upper Bound: Self-Hosting Cap

All models on the platform are open-source, with publicly available weights. Running a 70B model on a single H100 costs only $40–50 per day in total costs. Alats like vLLM and Ollama make local deployment extremely simple. NVIDIA’s next-generation chips will further significantly reduce inference costs. Institutions with sufficient usage will find self-deployment cheaper.

• Lower Bound: Cloud Giant Pressure

Microsoft, Google, Amazon, and Meta’s combined AI capital expenditure in 2025 exceeded $200 billion. They have hardware priority allocations, dedicated data centers, enterprise customer relationships, and can subsidize AI with cash flows from other businesses. Bittensor’s entire annual incentive budget (~$360 million) is less than Microsoft’s weekly AI infrastructure spending. Professional service providers also compete on price using VC subsidies on open-source models.

Subnet pricing is compressed into an extremely narrow band, while also bearing decentralization-specific costs: token friction, validator node overhead, subnet owner shares, network latency, etc.

The Moat Problem

Even if a subnet creates a valuable service, the underlying model and methodology are inherently public: Covenant-72B uses the Apache license, and its technical paper is publicly published. Any competitor can replicate it without participating in the TAO ecosystem.

Traditional moats (proprietary technology, network effects, switching costs, brand) do not apply:

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

The community believes the incentive mechanism itself is the moat, but this relies on sustained, large-scale token issuance, and each halving continuously shrinks the incentive budget.

What is TAO Actually Trading On?

At a $2.6 billion market cap, TAO’s price does not reflect demand fundamentals; $3–15 million in annual revenue cannot support this valuation under any traditional framework. The market is trading on: Bitcoin-style scarcity, Grayscale ETF expectations, AI sector rotation, and the long-term option value of decentralized AI. These are all valid speculative factors, but they stem entirely from the supply side and market sentiment.

If you hold TAO based on scarcity and narrative, you may still profit even with weak demand. But if you believe Bittensor will become a truly large-scale AI service network, there is currently no evidence, and it faces structural resistance that is difficult to overcome. Investors should clearly distinguish their investment thesis.

Artikel ini bersumber dari internet: Bittensor (TAO) Bearish Thesis: A Revenue Desert Under the Computing Power Myth

Related: Intelligent Computing Convergence: The Deep Integration Architecture, Paradigm Evolution, and Application Landscape of AI and the Cryptocurrency Industry

In the third decade of the 21st century, the convergence of Artificial Intelligence (AI) and Cryptocurrency (Crypto) is no longer merely the combination of two buzzwords; it represents a profound revolution in the technological paradigm. As the global cryptocurrency market capitalization officially surpassed the $4 trillion milestone in 2025, the industry has completed its transition from an experimental niche market to a vital component of the modern economy. One of the core driving forces behind this transformation is the deep integration of AI—an immensely powerful decision-making and processing layer—with blockchain, which serves as a transparent, immutable execution and settlement layer. This combination is addressing the respective pain points of both fields: AI is at a critical juncture in its evolution from centralized monopolies towards a decentralized, transparent era of “Open…