Bittensor TAO Climbs 15% After Nvidia CEO Backs Decentralized AI
Bittensor’s TAO token jumped more than 15% between March 19 and 20, 2026, after Nvidia CEO Jensen Huang and venture capitalist Chamath Palihapitiya publicly acknowledged the project’s decentralized AI training milestone on the All-In Podcast. At press time, TAO was trading around $304, with a 24-hour gain of 17.27% according to CoinMarketCap. That kind of validation from the head of the world’s dominant AI chip company does not happen in a vacuum, and the market priced it in fast.
The catalyst was a discussion on the All-In Podcast in which Palihapitiya spotlighted Bittensor’s Covenant-72B, describing how a large language model had been trained using spare compute contributed by independent operators across the open internet. He framed it as “an extraordinary technical achievement.” Huang agreed, comparing the model to a modern version of Folding@home, where distributed contributions collectively accomplish something no single actor could efficiently do alone. His exact words: “These two approaches aren’t mutually exclusive; they’re complementary. That’s absolutely certain.” That was enough. TAO appreciated roughly 5% within hours of the episode airing, and volume nearly doubled overnight.
What Covenant-72B Actually Is — and Why the Detail Matters
There is a number discrepancy worth addressing directly. During the podcast, Palihapitiya referenced a 4-billion-parameter LLaMA model trained in a distributed fashion. The Opentensor Foundation subsequently clarified that the model in question is Covenant-72B, a 72-billion-parameter language model pre-trained across more than 70 global contributors using standard internet-connected hardware. The methodology is documented in a March 2026 arXiv publication, and the model posted a 67.1 MMLU benchmark score. The Blockonomi report is more credible here precisely because it corrects the parameter count and cites the foundation’s own statement. The podcast got the spirit right but the specifics wrong, which is a reminder that narrative and fact are not the same thing, even when the narrative moves markets.
Covenant-72B represents the largest decentralized LLM pre-training run ever recorded. That is a genuine milestone, not a marketing claim. Training a model of this scale typically requires centralized data centers, massive capital expenditure, and access to hardware that most of the world cannot touch. Bittensor’s Subnet 3, run by Templar AI, proved the architecture can produce competitive results from disaggregated, commodity hardware. That is the real story underneath the price action.
Derivatives markets moved decisively. Futures open interest tracked by CoinGlass climbed from $131.94 million on March 4 to $361.15 million by Monday, a shift that reflects genuine positioning rather than spot-only enthusiasm. Daily trading volume peaked at $677.06 million on Sunday, the highest reading since November 7, according to Santiment data. By Friday, volume had settled at $521.92 million, still elevated. These are not the numbers of a dead-cat bounce. They suggest capital with a thesis behind it.
Subnet tokens inside the Bittensor ecosystem also moved. Templar was up 42% in the 24-hour period to $25.11, and Targon gained 12.28% to $14.14, both per CoinMarketCap. The mechanism is straightforward: as TAO gains visibility, participants rotate into subnet tokens to capture exposure to specific AI verticals running on the network. Social momentum accelerates that rotation, sometimes faster than fundamentals justify. Worth watching whether Templar’s 42% holds or retraces once the initial narrative heat dissipates.
Bitcoin BTC$77,253.00▲1.44%
Ethereum ETH$2,107.85▲1.87%
Tether USDT$0.9991▲0.03%
BNB BNB$661.36▲1.72%
XRP XRP$1.35▲1.36%
USDC USDC$0.9998▲0.01%
Solana SOL$85.32▲1.47%
TRON TRX$0.3714▲1.93%
Figure Heloc FIGR_HELOC$1.03▲0.00%
Dogecoin DOGE$0.1023▲1.42%Institutional Access Widens as Grayscale and Safello Move
Two institutional developments are running parallel to the price action. Grayscale has filed to convert its Bittensor Trust into a spot ETF, a structure that would meaningfully broaden access for traditional finance allocators. The trust currently trades at a premium, which analysts read as a sign that demand is outpacing the supply of regulated exposure vehicles. Whether the SEC approves the conversion is a separate question, but the filing itself signals that Grayscale sees institutional appetite as real and durable enough to pursue. The regulatory environment for crypto assets has been shifting in ways that make these filings more viable than they were eighteen months ago.
Separately, Nordic crypto exchange Safello cross-listed the Bittensor Staked TAO ETP on Nasdaq Stockholm as of March 19, 2026. The product, trading under the ticker STAO, was previously available only on the SIX Swiss Exchange. Expanding to a second regulated exchange opens the product to a broader pool of Nordic retail and institutional investors who require on-exchange access. These two moves together, Grayscale’s ETF application and Safello’s cross-listing, reflect a pattern of infrastructure building that typically precedes sustained institutional inflows rather than a one-week trade.
Technically, TAO is trading above its 50-day, 100-day, and 200-day exponential moving averages. The RSI sits at 77, firmly in overbought territory, which is a signal that short-term consolidation or a pullback is statistically more likely than continued vertical movement. Immediate resistance is mapped near $305.30, with the next meaningful target at $341.10 if that level clears. Analyst Ardi has identified a right-angled broadening descending wedge forming since October, with breakout confirmation requiring a daily close above $302. If that confirms, the projected range extends to $360 to $370. Over the past month, TAO has gained 55.66%, but it remains 61.13% below its all-time high of $767.68 set in April 2024. That gap is either an opportunity or a ceiling, depending on whether the decentralized AI thesis converts from narrative into sustained network utility.
The honest read on this moment: markets are responding to something real. A 72-billion-parameter model trained across 70 contributors on open internet infrastructure, then validated by the CEO of Nvidia on a podcast with tens of millions of listeners, is a legitimate proof-of-concept for a distributed AI future. But the price reaction always outruns the technology, and that gap between what is real and what is priced in is where most retail participants get hurt. The Covenant-72B milestone earned this attention. Whether TAO can hold $300 when the podcast cycle moves on to the next story is the question nobody on social media is asking right now, which is exactly why it is the right one to ask.
