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

When subsidies decline, what is the future of TAO’s valuation?

Article by: Pine Analytics

Translation: Saoirse, Foresight News

TAO is currently priced around $275, with a market cap of $2.6 billion and a fully diluted valuation of $5.8 billion. The project is endorsed by Grayscale (which submitted an ETF listing application to the NYSE in December 2025) and publicly recognized by NVIDIA CEO Jensen Huang. Additionally, the token supply narrative is highly attractive: a total cap 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 subnetworks increases from 32 to 128, and Templar’s Covenant-72B training demonstrates that decentralized compute power can run large language models with competitive benchmarks.

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

Bittensor (TAO) Token Distribution and Allocation

How Value Flows in the Network

Bittensor involves four types of participants:

• Subnet owners build specialized AI markets, earning 18% of TAO issuance rewards;
• Miners perform AI tasks (inference, training, data processing), earning 41%, approximately 1,476 TAO daily, with an annual value of about $148 million;
• Validators score miner outputs, earning 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: used for miner registration, validator staking, subnet token purchases, and service payments. In theory, subnet activity creates structural demand for the underlying token.

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

Demand Side Status

Transparent supply vs. opaque demand

Bittensor’s supply side is highly transparent: 3,600 TAO are distributed daily according to a programmed halving rule, with staking rate (~70%), distribution 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.

The following 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, pricing 85% lower than AWS and 10–50% lower than Together AI. Its usage data dominates the ecosystem: over 400,000 users (more than 100,000 API users), over 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 subsidized.

Based on a 14.4% 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 at about $0.88 per million tokens;
• DeepSeek V3 at 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 claimed 85% cost advantage is completely reversed; its low price is essentially paid by TAO holders through inflation, not due to structural efficiency from 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 user acquisition subsidies, but Uber, DoorDash, and AWS built switching costs during their subsidy periods: proprietary platforms, driver networks, enterprise ecosystems. Bittensor’s subnet has no barriers: open-source models, standardized interfaces, users can switch providers at zero cost. Once subsidies fade, no lock-in mechanism 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, with an estimated annual revenue of about $10.4 million, corresponding to a valuation of $48 million, and a P/S ratio of about 4.6. It is 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 Covenant-72B training, with a valuation of $98 million, but external revenue is zero. API access for training and enterprise sales is still in progress, and paid products have not yet launched.

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

With a $2.6 billion market cap, the revenue multiple is approximately 175–200x; with 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 rarely maintains over 50x long-term. Bittensor’s valuation multiple is 4–10 times that of aggressive industry benchmarks.

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 can create sustainable value.”

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

Pricing Dilemma: Facing Both Ends

Subnet networks 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 $40–50 daily; tools like vLLM and Ollama make local deployment very easy. NVIDIA’s new generation 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 infrastructure in 2025, with priority hardware quotas, dedicated data centers, and enterprise customer 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 of Microsoft’s AI infrastructure investment. Professional service providers also subsidize low-cost competition with VC funding on open-source models.

Subnet pricing is compressed into a very narrow range, while incurring unique decentralization costs: token friction, validator node expenses, subnet owner shares, network latency, etc.

Barriers to Entry

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 without participating in the TAO ecosystem.

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

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

The community believes that the incentive mechanism itself is a moat, but this relies on continuous large token issuance, which is diminished with each halving.

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 rotations in AI, and long-term options on decentralized AI. These are rational speculative factors, but they are entirely supply-side and sentiment-driven.

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