1. A 90% Monthly Rally With an Unlikely Catalyst
TAO — the native token of the Bittensor decentralised AI network — entered March 2026 trading near $180 per coin and ended the month above $330, having posted one of the more remarkable price recoveries in the AI token sector. At a market capitalisation of approximately $3 billion, TAO ranked among the top 30 cryptocurrencies by value. The rally had multiple contributing forces, but the single most identifiable inflection point came on March 20, when Nvidia CEO Jensen Huang appeared on the All-In Podcast alongside investor Chamath Palihapitiya and commented on Bittensor's recent technical achievement. Huang compared Bittensor to a modern version of Folding@home — the pioneering distributed computing project that mobilised millions of volunteers to create one of the world's most powerful supercomputers for protein folding research. The analogy, from the most prominent figure in AI hardware, landed with force in a market acutely sensitive to AI narrative validation.
2. The Technical Achievement That Prompted the Endorsement
Huang's comments were a direct response to Palihapitiya's description of Covenant-72B, a large language model trained permissionlessly on Bittensor's Subnet 3 — known as Templar — by a distributed group of more than 70 contributors. The model has 72 billion parameters and was trained on 1.1 trillion tokens entirely through decentralised computation, using consumer-grade GPUs connected over residential internet connections rather than the centralised data centre infrastructure that typically underlies models of this size. The results were confirmed in a March 2026 arXiv paper: Covenant-72B achieved a 67.1 score on the MMLU benchmark, the Massive Multitask Language Understanding test that evaluates AI models across 57 academic subjects. That score places it in competitive range with Meta's Llama 2 70B — a model built by one of the most resource-rich AI laboratories in the world. The achievement demonstrated that Bittensor's incentive structure, which rewards contributors of useful computational work with TAO tokens, could produce frontier-capable models without centralised coordination.
3. How Subnet Tokens Work and Why They're Moving Harder Than TAO
Bittensor's architecture divides the network into subnets — specialised mini-networks each focused on a specific AI task, ranging from language model training and decentralised compute provision to sports prediction, cybersecurity, and data services. Each subnet has its own native token, priced through an automated market maker that is backed by staked TAO. This mechanism creates a structural relationship between subnet token valuations and the underlying TAO price: when TAO appreciates, the backing value for subnet tokens rises, creating a leverage effect that amplifies both gains and losses relative to TAO. With TAO's market cap at approximately $3 billion and individual subnet tokens ranging from approximately $1 million to over $130 million, the subnet token category offers substantially higher volatility than TAO itself — functioning, in market terms, as leveraged bets on the Bittensor network's overall trajectory.
4. The Subnet Token Performance Leaders
The combined market capitalisation of Bittensor's subnet tokens reached approximately $1.47 billion during the reporting period, with 24-hour trading volume exceeding $118 million. τemplar, the token of Subnet 3 — the network behind the Covenant-72B model — led the category with a market capitalisation of approximately $137 million after a 444% monthly gain, making it the most prominent beneficiary of the Huang endorsement cycle. Other notable 30-day performers include OMEGA Labs on Subnet 24 at approximately 440%, Level 114 on Subnet 114 at 280%, BitQuant on Subnet 15 at 230%, Nova on Subnet 68 at 218%, and Grail on Subnet 81 at 211%. Even the more established, larger-cap subnet tokens posted significant monthly returns: Chutes on Subnet 64, with a market cap near $132 million, gained 54%; Targon on Subnet 4 rose 166%; Ridges AI on Subnet 62 gained 85%; and Hippius on Subnet 75 added 115%.
5. The Folding@home Comparison and Its Significance
Jensen Huang's characterisation of Bittensor as a modern Folding@home carries specific implications that help explain why the endorsement had such pronounced market impact. Folding@home mobilised millions of individual participants to contribute idle computing power to scientific research — protein folding simulations that required supercomputer-scale resources — through voluntary participation incentivised by recognition rather than financial reward. At its peak, it was among the world's most powerful distributed computing projects, demonstrating that decentralised coordination could match or exceed the output of centralised infrastructure for sufficiently well-structured problems. Huang's framing positions Bittensor's approach to AI model training as a potential analogue — suggesting that the incentive design of TAO rewards for useful computational contributions could mobilise a comparable scale of distributed GPU capacity toward AI model development. For a market looking for narrative validation of decentralised AI, that comparison from the CEO of the world's dominant AI chip manufacturer is close to the most credible possible endorsement.
6. The Supply Dynamic: December's Halving Amplified the Rally
The Huang endorsement arrived against a tokenomic backdrop that had been tightening since December 2025, when Bittensor completed its first-ever token halving event. The halving cut daily TAO emissions from 7,200 to 3,600 tokens — a supply reduction structurally analogous to Bitcoin's halving mechanism, which TAO's tokenomics deliberately mirror with a hard cap of 21 million coins. At the time of the rally, approximately three-quarters of the total circulating supply was already staked within the network's subnet ecosystem, reducing the volume of TAO available for sale on exchanges. The combination of reduced daily issuance, high staking participation, and surging demand driven by the Huang endorsement and Covenant-72B's performance created the supply-demand imbalance that drove the month's extraordinary price movement. Q1 2026 also saw Bittensor generate approximately $43 million in AI customer revenue — a figure that positions the network as a commercially viable infrastructure project rather than a speculative token.
7. Institutional Interest Building Around TAO
Beyond retail and speculative demand, institutional access infrastructure around TAO has been developing in parallel. Grayscale filed an S-1 with the SEC to list shares of its Bittensor Trust on NYSE Arca — the first potential U.S. exchange-traded product for TAO. A spot ETF regulatory decision is possible by late 2026, and if approved, it would open the asset to institutional allocation channels that currently cannot access TAO directly. In Europe, Deutsche Digital Assets has already listed a staked TAO exchange-traded product, providing regulated exposure to institutional investors in that market. Digital Currency Group subsidiary Yuma is contributing to 14 separate Bittensor subnets — a pattern more consistent with infrastructure investment than speculative token acquisition, suggesting some institutional participants are treating the network as a durable AI infrastructure layer rather than a near-term trade.
8. The Network's Planned Expansion
Looking ahead, Bittensor's development roadmap includes expanding the network's active subnet capacity from its current ceiling of 128 subnets to 256. That expansion would create the conditions for a new wave of subnet launches and corresponding token issuances, broadening the range of AI tasks being addressed through the network's incentive structure. Total subnet staking has grown by an extraordinary 833,000% year-over-year to exceed $620 million — a figure that reflects the scale of capital that has been committed to the network's computational economy in a relatively short period. The breadth of AI tasks currently covered by active subnets — spanning language model training, decentralised inference, data provision, confidential computation, cybersecurity, and sports prediction — indicates that the network has moved beyond a single use case into a diversified AI infrastructure ecosystem.
9. The Risks That Come With Leverage Dynamics
The same structure that has allowed subnet tokens to generate 200–440% monthly gains also makes them significantly more fragile than TAO itself. Their valuations are mechanically dependent on TAO's price through the automated market maker backing mechanism, meaning a sustained TAO downturn would compress subnet token values disproportionately. The broader market question for Bittensor's continued momentum is whether Covenant-72B represents the beginning of a reliable production-grade AI model development pipeline, or whether it was an exceptional result that happened to coincide with a major public endorsement. If the network can produce additional competitive AI models at consistent intervals — demonstrating that the incentive structure scales beyond a single headline achievement — the narrative foundation for sustained institutional engagement becomes considerably more durable. If it cannot, the current subnet valuations embed a level of expectation that the underlying technology may not yet be able to fulfil reliably.
10. The Broader Significance of Decentralised AI Validation
The market response to Huang's Bittensor comments carries implications that extend beyond TAO's price chart. Nvidia's CEO is, more than any other individual, the person most credibly positioned to assess which approaches to AI development have structural merit — and his public framing of Bittensor's permissionless training model as comparable to Folding@home represents a form of technical legitimacy that no marketing campaign could purchase. For an asset class that has spent years seeking institutional recognition of its AI-native projects as substantively different from speculative tokens, the Huang endorsement of Bittensor's specific technical achievement may prove to be a defining moment in how decentralised AI infrastructure is perceived by the mainstream technology industry. Whether that perception translates into durable value creation depends on whether Bittensor's subnet ecosystem continues to deliver on the technical promise that the Covenant-72B paper established.