Author: Prathik Desai Source: tokendispatch Translation: Shan Oppa, Golden Finance

The AI industry now resembles a closed religious system: financing and valuation are conducted behind closed doors. A few leading companies have raised huge amounts of money, recruited top researchers, and rented ultra-large-scale computing power clusters. However, the market can only infer their value by announcing a round of financing within a few months. The so-called “valuation” is often just a number agreed upon by a few people in a room, rather than the true price found in a free-flowing market. By the time ordinary investors can see the price, most of the upside has already been taken up by early participants.

Bittensor’s core proposition is that AI should not be financed in this way. I'm fascinated by the system it's building. Not because it can create better models than OpenAI, Anthropic, and Google, at least not yet, but because it has found a decentralized path to publicly evaluate, finance, and price AI projects before they grow into traditional companies.

This model is completely different from the many decentralization attempts that have appeared in past waves of AI.

Bittensor's subnet system continues to provide support to the team, reward efficient performers, eliminate backward projects, and reprice the entire AI ecosystem in real time. This is unprecedented pricing for AI. I admit that the process of building AI in this model is brutal, but it is also more honest.

In this in-depth analysis, I will take you to break down the operating logic of Bittensor and why it may be more advantageous than any previous attempts at AI pricing.

The Secret Room of AI Pricing

In the first quarter of 2025 alone, AI startups received $73.1 billion in financing, accounting for 58% of total global venture capital investment. Although investment institutions such as GIC and TPG have warned that the valuation multiples of some segmented tracks are too high, there is almost no operating performance to support this valuation.

This model is beneficial to founders, insiders and late-stage investors, but excludes others: resource parties that provide key computing power, developers based on open source models, and early ordinary users cannot share the dividends. Even the rise of open source AI has not changed this situation. Funds are still concentrated in cloud service contracts, deployment layers, enterprise packaging, technical support, security and distribution.

In the entire value creation process, the public participates extensively in making contributions, but the fruits are only picked by a few people. Although this pattern has existed for a long time, the real change comes from the rise of the open source model AI economy.

Red Hat developers pointed out in the report that enterprises are increasingly adopting open source AI models for localized deployment, autonomous controllable and professional tasks, especially in highly regulated industries such as telecommunications and banking. Enterprises need AI deployment solutions that can be used to monitor, automate and scale operations, rather than just referencing an AI model.

Large institutions such as McKinsey also agree with this trend. Its survey shows that more than half of the companies surveyed have fully used open source AI in their technology stack. The survey covered more than 700 technical leaders and senior developers in 41 countries.

Bittensor’s model is based on these industry changes and challenges the current pricing system of AI projects.

Crypto-native investors are going into a frenzy for Bittensor’s native token TAO, with its price doubling in the past month. Others are keen to debate the merits of decentralized AI versus centralized AI. But for me, it’s more important to explore more precise AI pricing methods. The answer given by Bittensor is: let all parties that provide funding, development, verification and use of AI come together in the same market, and price AI based on public indicators.

Bringing AI to the open market

It’s easy to understand Bittensor if you think of it as a network of micro-AI economies rather than a single token.

Each subnet is a specialized market for a specific task in the AI technology stack, which may focus on inference, distributed training, predictive signals, or computing power supply. The subnet creator sets the incentive mechanism and target tasks, the miners perform the tasks, the verifiers score the results, and the stakers can support specific verifiers by staking TAO.

After Bittensor launched the dynamic TAO upgrade in February 2025, the incentive mechanism became more innovative: each subnet has its own token and fund pool. Bittensor is no longer a single generalized AI investment target, but an ecosystem that accommodates many small AI projects.

In the second half of 2025, Bittensor will link reward distribution more to TAO net inflows rather than rigid token prices. In December of the same year, TAO completed its first halving, and the daily circulation dropped to 3,600, further forcing capital to be allocated on the basis of merit, turning the AI ​​market into a survival arena for the fittest.

Web3 researcher and author Jeff accurately summarizes this as "the dynamics of Darwinian AI" and gave a wonderful summary in the 0xJeff newsletter:

The core of Darwinism is natural selection and survival of the fittest. Individuals compete with each other and excellent traits that are conducive to survival are passed on from generation to generation.

Similar logic is implemented at multiple levels of Bittensor:

• Subnet competition: Each subnet competes for the incentive share of 3,600 TAO per day, and the head subnet uses incentives to gain a more lasting living space

• Miner competition: Miners compete to provide the best results. Global participants compete based on key subnet indicators. Top miners can receive 41% of the subnet alpha token rewards

• Validators compete with investors: validators compete to verify miner tasks, and investors compete to bet on the best-performing subnet

What happens if you don’t compete or perform poorly? disuse. The subnet will be removed directly (yes, it will be deleted directly from the system).

This is the core difference from traditional AI models.

In the traditional model, founders pitch the company, raise equity, build a team, build internally, and hope the market recognizes its valuation.

Bittensor disrupts this model by disclosing investments early in the market. In this model, entrepreneurs first launch subnets and subsequently GPU operators contribute computing resources. Afterwards, developers and researchers contribute their work, and investors purchase investment shares through TAO or specific subnet tokens. Finally, customers pay to use the underlying service. Ultimately, the market takes all factors into consideration and prices the project as a whole.

What I love most about it is that it reimagines capital markets for every stakeholder.

Unlike private startups, investors can continue to discover prices without waiting for the next funding round. In fact, Bittensor allows them to overview the entire ecosystem through the TAO platform, or focus on their most promising individual subnets to make more precise investments.

The appeal for developers is that they don’t have to be confined to Anthropico, OpenAI, or other elite hyperscale data centers to participate in the positive impact of AI developments.

It provides entrepreneurs with a capital market around their ideas and access to support even before they develop into a full-fledged company, something never before seen in the venture capital industry. This can be seen in the way capital is aggregated within the network. A handful of subnets now attract a disproportionate amount of TAO capital inflows and outflows, while other subnets lag relatively behind. The top five subnets by market capitalization account for almost one-third of the total market capitalization of the 128 subnets.

For customers, the system provides cheaper, more flexible access to open infrastructure.

In addition to this, the Bittensor model is more appealing to all stakeholders because it not only sounds fairer, but is also more commercially viable.

The market is maturing

The trend is obvious as institutional investors increasingly view Bittensor as compliant and biddable.

In December 2025, Grayscale Bittensor Trust was listed for trading on the OTCQX top over-the-counter market, providing traditional investors with a familiar channel to participate in this unfamiliar but highly demanded asset.

The sign that any emerging market is maturing is having compliant packaging, trading codes, on-screen quotes and access to brokerage accounts, just like the paths experienced by Bitcoin, Ethereum ETFs and Digital Asset Treasury Bonds (DATs). Bittensor may not be as well-known in the crypto market as Bitcoin and Ethereum, but the launch of Grayscale Trust marks that institutional interest has shifted from theory to real products.

Bittensor's work has even attracted recognition from the top minds in the elite industries it promises to disrupt.

When well-known venture capitalist and entrepreneur Chamath Palihapitiya mentioned Bittensor’s distributed training run to Nvidia CEO Jensen Huang, Huang didn’t downplay it as a lowly achievement in the cryptocurrency space. He calls it "a modern version of Folding@home," referring to a decentralized, distributed project that uses the remaining computing power of volunteer computers to simulate protein folding or other complex problems.

This positioning places Bittensor into the long-term historical context of distributed computing, rather than being limited to the token cycle narrative.

The recent achievements of Templar, one of Bittensor’s leading subnets, have further confirmed its capabilities from a technical perspective: its Covenant-72B is a large model with 72 billion parameters, which was trained from scratch based on 1.1 trillion Tokens by more than 20 participants around the world through Bittensor collaboration. In public benchmarks, Covenant-72B achieved an MMLU score of 67.11, higher than LLaMA-2-70B’s 65.63.

To put it bluntly, it still cannot surpass OpenAI or Anthropic, but it is enough to prove that decentralized collaboration can create AI infrastructure with commercial value.

Subnets such as Chutes are clearly positioned as decentralized serverless AI computing power platforms. Bittensor’s official documentation also defines subnets as independent competitive markets for digital commodities such as reasoning and training. This suggests that the market is not pricing a vague AI narrative, but rather is pricing specific modules of the technology stack independently.

The dilemma on the demand side

Bittensor’s supply-side transparency far exceeds that of any other AI market: data such as issuance volume, pledge flow, and capital accumulation subnets are clear at a glance. The real problem is that demand-side information is not transparent.

The blockchain only records token circulation, but does not collect data such as user retention, API usage quality, profit margins, and audited revenue. Even if a subnet appears to be prospering, investors can often only infer business quality from market structure rather than financial statements.

Pine Analytics made sharp criticisms in analyzes such as "Transparent Supply vs. Opaque Demand" and "Chutes (SN64): Low Price Supported by Subsidies": Part of Bittensor's outstanding business performance may still be driven by subsidies, and the essence of subsidies is TAO issuance rewards within the subnet. Pine calculated that the external revenue that can be recognized by the entire network is still minimal compared to the value implied by the TAO price.

The most typical example is Chutes, the largest subnet of Bittensor: it receives an annual TAO issuance subsidy worth US$52 million, with external revenue of only US$2.4 million. Without subsidies, its operating costs would be unaffordable. This is not to deny the Bittensor model, it just shows that the current market is pricing more for the vision of AI rather than pricing the cash flow of AI.

Because of this, I pay special attention to the development process of Bittensor. It has all the signs of ecological maturity. Although it has not ended the debate on "decentralized AI" and is still polishing the most accurate valuation method for AI projects, it has made great strides in uncovering issues that have been deliberately ignored by the private market for a long time and priced beliefs and valuations through the public market.

While private AI giants are asking the world to trust a few people in a room to determine multi-trillion dollar valuations, Bittensor chooses to trust the public markets. I know the latter is not perfect, but I appreciate and recognize the transparency it brings.