Bittensor TAO 2026: After Subnet Expansion, Why Is Decentralized AI Heating Up Again?

In May 2026, decentralized AI is moving from the narrative-building phase into the infrastructure verification phase. In this track, Bittensor and its native token TAO have consistently occupied an unavoidable core position.

According to Gate market data, as of May 25, 2026, TAO is priced at $273.7, down about 1.93% over the past 24 hours. Its market cap is approximately $2.626 billion, and overall market sentiment is in a neutral range. Over the past 90 days, TAO has risen by 63.01% in total, rebounding from a low of $165.1 to a high of $379.2, but it is still down about 36.94% compared with a year ago’s high of $538.9. This price action reflects a market perception filled with tension: optimists view it as the “Bitcoin of AI,” while skeptics question whether its massive market cap is truly backed by real income that matches it.

Dual Narratives: Subnet Breakout and a Governance Crisis

In the first half of 2026, the Bittensor ecosystem saw highly dramatic ups and downs.

On one hand, ecosystem expansion accelerated. The number of Bittensor subnets has grown to 129, gradually forming a structure with top-heavy concentration and layering. According to industry data tracking institutions, as of the late March 2026 period, the total TAO staked across all subnets in the Bittensor ecosystem surged from about $74,400 a year earlier to over $620 million, an increase of 833,000%. The total market capitalization of subnet tokens has exceeded $1.5 billion.

On the other hand, cracks at the governance level concentrated and erupted during the same period. On April 10, 2026, Covenant AI—one of the top development teams in the Bittensor ecosystem that successfully trained the 72-billion-parameter large language model Covenant-72B—suddenly announced its withdrawal from the network. Its founder, Sam Dare, publicly accused Bittensor co-founder Jacob Steeves of implementing “centralized control” over the network, and referred to the entire decentralized AI narrative as a “decentralized performance.”

After the announcement, TAO’s price fell noticeably in a short time. Covenant AI’s founder then sold off the TAO subnet tokens it held, triggering a chain of liquidations and pushing the price even lower.

It should be clarified that Covenant AI’s withdrawal behavior and the subsequent token sell-off mentioned in this paragraph are all publicly recorded, verifiable statements. The direction of price volatility can also be confirmed by market data; they are not analytical inferences.

The Gap Between Scale Expansion and Income Reality

Structural Differentiation of the Subnet Ecosystem

Bittensor’s expansion shows clear stage-like characteristics. From just 1 subnet in early 2023, to surpassing 100 by April 2025, and then stabilizing at 128 or above in early 2026. The protocol design caps the number of subnets at 256, leaving room for later scaling. After the first halving was completed in December 2025, the daily TAO issuance fell from 7,200 to 3,600, significantly tightening the supply side.

However, scale expansion does not mean a synchronized improvement in quality. Out of approximately $2.36 billion worth of TAO staked, over 30% has flowed into the subnet side, indicating that funds are migrating from the main network to the application layer. But the staking distribution shows clear structural differentiation: about 48% of TAO staking remains in the root network, while only 19% is staked within subnets. This means most TAO holders still tend to choose root network staking with relatively lower risk and more stable returns.

The Core Contradiction in the Income Model

Bittensor’s biggest structural problem lies in the sustainability of its economic model.

Bittensor’s subnet architecture covers 129 active subnets. Miners and validators receive TAO emission rewards for providing “intelligent” contributions across these subnets. The network’s external income that has been confirmed lies between $3 million and $15 million per year. Specifically, Chutes (SN64) generates about $5.5 million annually from paid API calls, while also receiving TAO emissions on the scale of tens of millions of dollars. Targon (SN4) earns over $10 million per year and has connections to large enterprise partners.

Multiple analysts point out that most subnets still heavily rely on TAO emissions to incentivize participants, and there is a clear gap between subsidies and external income. Taking Chutes as an example, its $5.5 million in external income is orders of magnitude lower than the tens of millions of dollars in TAO emissions it receives.

As the December 2025 halving continues to play out, if subnets’ external income cannot achieve significant growth within a reasonable time window, the gap between inflation incentives and real demand may further widen. But it is also worth noting that on-chain activities involving inference and agent deployment are increasing, which some observers see as an early signal of moving toward income independence.

Below is a comparison of key structural indicators in the Bittensor ecosystem:

| Indicator dimension | Data | Explanation | | --- | --- | --- | | Number of active subnets | 129 | Protocol limit of 256 | | Total subnet staking | Over $620 million | Up 833,000% in one year | | Total market capitalization of subnet tokens | Over $1.5 billion | Strong performance from subnets such as Templar, Quasar | | Chutes (SN64) annual income | About $5.5 million | Paid API calls for inference services | | Targon (SN4) annual income | Over $10 million | Includes enterprise partner income | | Total external annual income for the whole network | $3 million - $15 million | A clear gap versus market cap | | TAO circulating market cap | About $2.626 billion | As of May 25, 2026 | | Fully diluted valuation | About $5.8 billion | Total supply of 21 million | | Daily issuance | 3,600 TAO | After the December 2025 halving | | Root network staking share | About 48% | Preference for low-risk allocation |

Public Sentiment: Institutional Endorsement and Community Disagreement Coexist

Multiple Perspectives: Continued Institutional Capital Escalation

Since 2026, narratives about institutional participation in Bittensor have heated up significantly. The following is a list of core events (factual statements):

Grayscale filed an S-1 document with the New York Stock Exchange in December 2025 to apply for a TAO spot ETF. Grayscale also launched a compliant trust product, the Grayscale Bittensor Trust, to provide a channel for institutional investors to participate.

In addition, Nvidia CEO Jensen Huang has publicly recognized the training results of the Covenant-72B model within the Bittensor ecosystem, calling it a noteworthy technological achievement. Social Capital founder Chamath Palihapitiya also mentioned this achievement in a public podcast.

Bearish View: Governance Concentration and an Income “Desert”

The bearish argument also has internal logic. The key contradiction exposed by the Covenant AI incident is this: Bittensor’s Yuma consensus mechanism is theoretically a decentralized game system, but in actual operation, the staked capital behind the top validator nodes is highly concentrated among early investors, foundations, and addresses related to the founders. This means the founders are not only rule-makers, but—to a significant extent—also the biggest arbiters.

Another analysis points out that after subnet operators receive TAO token rewards, they have no mandatory obligation to feed back the value of the generated AI models, data, or services to the ecosystem. They may choose to privatize the results and detach from Bittensor, causing TAO holders to be unable to effectively capture value.

Below is a side-by-side comparison of mainstream pro and bearish views:

| Pro views | Bearish views | | --- | --- | | Grayscale submitted an ETF application, with high institutional attention | There is a huge gap between TAO market cap and actual network income | | The halving mechanism creates a supply contraction effect | Subnet income depends on inflation subsidies, with a low share of external paid revenue | | Subnet ecosystem expands quickly, leading to an independent token economy | Governance power is highly concentrated, and founders have too much control over validator nodes | | Recognized publicly by the Nvidia CEO | Ecosystem development exit risk: core contributors can take their results and leave the network |

Four Key Issues

Has the level of decentralization met market expectations?

There is a significant gap. The reality revealed by the Covenant AI incident is this: compute power can be distributed, but network control power and capital are highly concentrated. This fact creates structural tension with Bittensor’s “decentralized AI” narrative. According to public information, Jacob Steeves proposed restarting a community-voting governance mechanism on Discord, which indirectly confirms that the current governance architecture needs improvement, but validating the execution and rollout of specific improvement plans still requires time.

Is the token economic model sustainable?

There is objective pressure on the income side. The verified on-chain fact is that the Chutes subnet’s $5.5 million in external income is far lower than the tens of millions of dollars in TAO emissions it receives. This ratio may still be supported by market narratives in the current stage, but whether it can be sustained in the medium to long term depends on the growth speed of external paid demand; at present, there is insufficient data to make a definite judgment.

What drives institutional entry?

Institutional inflows mainly reflect the demand for allocating capital across the AI sector as a whole. The Grayscale ETF application makes TAO one of the more liquid institutional investment options within the decentralized AI narrative. But it needs to be distinguished: institutional allocation behavior is the result of “being bullish on the AI track and TAO being a major selectable option,” and it does not equate to a comprehensive endorsement of Bittensor’s current economic model.

Does the competitive positioning against centralized AI hold?

Differentiation exists, but still needs verification. Bittensor’s subnet Chutes claims that its AI inference service pricing is lower than that of traditional cloud providers. If this cost-performance advantage can be sustained and expanded, Bittensor could have genuine differentiated competitive strength in the AI infrastructure market. However, current income data shows that this advantage has not yet translated into a sufficiently large base of paying users.

Industry Impact: A Structural Reference Point for Decentralized AI

Expanding the lens from Bittensor as a single project to a more macro view of the industry, several trends are worth watching.

The integration of decentralized AI with blockchain is accelerating. The demand for AI agents to rely on paid infrastructure is becoming a new growth pole for on-chain economies. In the decentralized compute space, various decentralized GPU computing protocols are growing rapidly, forming complementary solutions to traditional centralized cloud services. The global AI market size was about $350-$400 billion in 2025, and is expected to grow to $1.5-$2 trillion by 2030.

Bittensor’s position in this landscape is special. It is not a market that rents out GPU hardware compute capacity; rather, it incentivizes intelligent AI output itself—forming a decentralized machine intelligence market through its subnet architecture. This differentiated positioning makes it complementary to other decentralized compute protocols rather than directly competing with them.

Below is a differentiated comparison of Bittensor and representative decentralized AI projects:

| Project | Core positioning | Differentiation features | | --- | --- | --- | | Bittensor (TAO) | Decentralized AI model incentive coordination layer | Subnet architecture incentivizes intelligent output, not hardware leasing | | Render (RENDER) | Decentralized GPU rendering | Focuses on rendering scenarios and expands toward general AI computing | | Akash (AKT) | Decentralized cloud computing marketplace | Reverse-auction pricing; general compute resource market |

Conclusion

Bittensor’s current situation is, in essence, a classic proposition about value anchoring: when a protocol gains a market valuation far beyond its actual current income by leveraging a grand industry narrative, a carefully designed token economic model, and strong institutional endorsements, what exactly is the market doing—discovering forward-looking value, or front-loading valuation driven by narrative?

The 833,000% growth in subnet staking from $74,400 to $620 million is indeed striking. But the gaps are equally hard to ignore: the difference between Chutes’ $5.5 million in annual revenue and the tens of millions of dollars in TAO emissions; the difference between the network’s total external income of $3 million to $15 million and its market cap of about $2.6 billion; and so on. Grayscale’s ETF application reflects a long-term bet from an institutional perspective, but Covenant AI’s withdrawal statement reminds the market that governance risks under a decentralized narrative are not something that can be dismissed as negligible tail events.

The next breakout point for decentralized AI may not be simply stacking more subnets, higher staking amounts, or larger-scale institutional allocations. Instead, it hinges on whether it can achieve structural breakthroughs across the two fundamental dimensions of income generation and governance credibility. For market participants, continuously tracking quarterly changes in subnet external income, the progress of inflation absorption after the halving, and the on-the-ground implementation of governance proposals will be the core observation framework for evaluating how this proposition evolves.