Decentralized artificial intelligence (AI) protocol Bittensor’s native $TAO token has rallied roughly 90% over the past month from around $180 at the start of March to above $332 as of March 24, and the momentum is spilling over into its subnet token ecosystem.
According to CoinGecko data, the Bittensor Subnets category is up 30% over the past 24 hours to a combined market capitalization of $1.47 billion, with trading volume topping $118 million. Of the subnet tokens tracked on the platform, a significant number have posted triple-digit percentage gains over the past month.
Top Subnets
What Are Subnets?
Subnets are specialized mini-networks within the Bittensor ecosystem, each dedicated to a specific AI task, ranging from language model training and decentralized compute to sports prediction and cybersecurity. Miners within each subnet compete to produce high-quality AI outputs, while validators evaluate performance and allocate $TAO rewards accordingly.
Since the launch of dynamic $TAO (dTAO) in February 2025, each subnet operates as its own automated market maker with a natively assigned token. The valuation of each subnet token is determined by the amount of $TAO staked into that subnet’s reserves, creating a direct economic link between $TAO‘s price and subnet token performance. The subnet ecosystem first crossed $500 million in April 2025, and by July 2025, the cumulative subnet market cap had neared $1 billion as $TAO treasury companies like xTAO and Synaptogenix emerged.
The relationship is reflexive: as $TAO appreciates, each subnet’s $TAO reserve becomes more valuable, inflating subnet token prices, which, in turn, attract more stakers and attention to the ecosystem.
Subnet Standouts
The $TAO rally accelerated on March 20 after Nvidia CEO Jensen Huang and investor Chamath Palihapitiya endorsed Bittensor’s decentralized AI training model on the All-In Podcast.
That enthusiasm has cascaded into subnet tokens. τemplar (SN3) — the subnet behind the Covenant-72B model that triggered the Nvidia-fueled rally — is now the top subnet token with a $137 million market cap after rallying 444% over the past month.
Other notable 30-day performers include OMEGA Labs (SN24) at 440%, Level 114 (SN114) at 280%, BitQuant (SN15) at 230%, Nova (SN68) at 218%, and Grail (SN81) at 211%.
Even the more established subnet tokens with larger market caps have posted strong monthly returns. Chutes (SN64), which has a $132 million market cap, is up 54% over 30 days. Targon (SN4) has gained 166%, Ridges AI (SN62) is up 85%, and Hippius (SN75) has risen 115%.
The primary catalyst for the broader rally was the reveal of Covenant-72B, a large language model trained permissionlessly across Bittensor’s Subnet 3 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, putting it in a competitive range with Meta’s Llama 2 70B.
The Nvidia CEO endorsed both proprietary and open-source AI models as complementary during the podcast appearance, framing foundational AI technology as benefiting from decentralized innovation.
Beta Play on Decentralized AI
The outsized returns in subnet tokens relative to $TAO itself reflect a dynamic familiar to crypto markets: smaller-cap ecosystem tokens tend to act as leveraged bets on their parent protocol. With $TAO‘s market cap sitting at roughly $3 billion and individual subnet tokens ranging from $1 million to $110 million, the tokens offer significantly higher volatility in both directions.
Looking ahead, the Bittensor network plans to expand its active subnet capacity from 128 to 256 later this year, which could bring a fresh wave of new subnet token launches and further broaden the category.
Before the first $TAO halving in December 2025, subnets had already reached a cumulative $1.28 billion market cap, with Yuma, a subsidiary of Digital Currency Group, contributing to 14 different subnets. A potential regulatory decision on converting the Grayscale $TAO Trust into a spot ETF could also provide additional institutional access by late 2026.
For now, the surge in subnet tokens signals that market participants are betting the decentralized AI narrative has staying power, and that Bittensor’s expanding network of specialized AI subnets will be at its center.
