Bittensor’s $TAO has rallied 90% so far this month, and the tokens in its ecosystem are running up even harder.
The network’s subnet token category reached a combined market cap of $1.47 billion on Monday, with $118 million in 24-hour trading volume, according to CoinGecko data.
The surge follows $TAO‘s own run from $180 to above $332 in March, but the subnet tokens are where the real action is. Templar, the token for Subnet 3, gained 444% in 30 days. OMEGA Labs rose 440%. Level 114 added 280%. BitQuant gained 230%. Even the larger subnet tokens posted significant returns, with Chutes up 54% and Targon gaining 166%.
Bittensor is a decentralized network that creates marketplaces for artificial intelligence. Instead of one company building and controlling AI models, Bittensor incentivizes 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 focused on a different AI task, from training language models to running compute 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 into it.
Several catalysts contributed to these moves of the 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 commodity internet hardware.
The model was trained on 1.1 trillion tokens and achieved a 67.1 MMLU score, confirmed in a March 2026 arXiv paper. That puts it in competitive range with Meta’s Llama 2 70B, a model built by one of the most well-resourced AI labs in the world. (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 compute power and compete to produce useful training gradients for large language models, while validators evaluate the quality of their contributions and distribute $TAO rewards accordingly.
Think of it as a way to train AI models the same way bitcoin mines 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 All-In Podcast on March 20, framing decentralized AI training as complementary to proprietary models. Coming from the CEO whose blog post earlier this month briefly helped reverse a tech stock selloff, the endorsement carried weight beyond the usual crypto echo chamber.
How subnet tokens work
The subnet token mechanics explain why the gains are so outsized relative 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 staked into that subnet’s reserves. When $TAO appreciates, every subnet’s reserve becomes more valuable, inflating token prices and attracting more stakers. The relationship is reflexive and amplifies moves in both directions.
With $TAO at roughly $3 billion in market cap and individual subnet tokens ranging from $1 million to $137 million, the subnet tokens function 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 on converting the Grayscale $TAO Trust into a spot ETF could provide institutional access by 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 sustains depends on whether Bittensor keeps producing competitive AI models or whether Covenant-72B was a one-off that got lucky timing with a Huang endorsement.
