Bittensor’s TAO is up 90% so far this month, and tokens in its ecosystem are running up even harder.
The network’s subnet token category reached a total market capitalization of $1.47 billion on Monday with $118 million in 24-hour trading volume, according to CoinGecko data.
The rise follows TAO’s own run from $180 to over $332 in March, but the subnet tokens are where the real action is. Templar, the symbol of Subnet 3, rose 444% in 30 days. OMEGA Labs rose 440%. Level 114 added 280%. BitQuant rose 230%. Even the larger subnet tokens saw significant returns, with Chutes up 54% and Targon up 166%.
Bittensor is a decentralized network that creates marketplaces for artificial intelligence. Instead of one company building and controlling AI models, Bittensor encourages a global network of participants to contribute computing power, data and machine learning models in exchange for TAO, the network’s native token.
The network is divided into specialized sub-networks called subnets, each of which focuses on a different AI task, from training language models to running computing infrastructure to cybersecurity analysis. There are currently 128 active subnets, each with its own token whose value is tied directly to the amount of TAO staked on it.
Several catalysts contributed to these movements of Bittensor’s ecosystem tokens.
Subnet 3 produced Covenant-72B, a large language model trained permissionlessly across Bittensor’s decentralized network by over 70 contributors using common Internet hardware.
The model was trained on 1.1 trillion tokens and achieved a 67.1 MMLU score, confirmed in an arXiv paper from March 2026. That puts it in competitive range with Meta’s Llama 2 70B, a model built by one of the most resourceful AI labs in the world. (The MMLU, or Massive Multitask Language Understanding, is a standardized test for AI models that scores them across 57 academic subjects.)
Subnet 3, called Templar, is Bittensor’s decentralized AI training network. Miners contribute GPU computing power and compete to produce useful training gradients for large language models, while validators evaluate the quality of their contributions and allocate TAO rewards accordingly.
Think of it as a way to train AI models similar to bitcoin mining blocks, with distributed participants around the world contributing hardware and getting paid for useful work.
Elsewhere, Nvidia CEO Jensen Huang and investor Chamath Palihapitiya endorsed Bittensor’s approach on the March 20 All-In Podcast, framing decentralized AI training as a complement to proprietary models. Coming from the CEO whose blog post earlier this month briefly helped reverse a tech stock selloff, the endorsement weighed beyond the usual crypto echo chamber.
How subnet tokens work
The subnet token mechanics explain why the gains are so large compared to TAO itself.
Since Bittensor launched dynamic TAO in February 2025, each subnet operates its own automated market maker with a native token whose valuation is determined by the TAO deposited into that subnet’s reserves. As TAO appreciates, all of the subnet’s reserves become more valuable, increasing token prices and attracting more players. The relationship is reflexive and reinforces movements in both directions.
With TAO at around $3 billion in market capitalization and individual subnet tokens ranging from $1 million to $137 million, the subnet tokens act as leveraged bets on the parent protocol.
The network plans to expand from 128 to 256 active subnets later this year, which would bring a new wave of token launches.
A potential regulatory decision to convert the Grayscale TAO Trust into a spot ETF could allow institutional access in late 2026. And Digital Currency Group subsidiary Yuma is already contributing to 14 different subnets, suggesting the smart money is treating this as infrastructure rather than speculation.
Whether the subnet rally continues depends on whether Bittensor continues to produce competitive AI models or whether Covenant-72B was a one-off that got lucky timing with a Huang endorsement.
