Original article | Odaily Odaily( @OdailyChina )
Author｜Azuma ( @azuma_eth )
Do you remember the story of Nvidia CEO Jensen Huang praising Bittensor (TAO)?
On March 20th, during an appearance on Chamath Palihapitiya's All-In podcast, Huang was asked whether he was optimistic about decentralized AI systems/computing networks. Palihapitiya cited Bittensor as an example (with a hint of personal bias), stating that a team on a subnet of Bittensor had successfully trained a Llama model with 4 billion parameters (actually 72 billion parameters), and that the entire process was completed collaboratively by distributed computing power. Huang's assessment upon hearing this was "a remarkable technological achievement."
Benefiting from this positive development, TAO surged against the market trend last month, once breaking through $370, and Bittensor was regarded by the cryptocurrency industry as "the hope of the whole village".
Then, just half a month later, the situation took a sharp turn for the worse due to a sudden statement. As of the morning of April 10, TAO had fallen below $290, experiencing a sharp decline for three consecutive days, and Bittensor was caught in what may be the biggest media storm since its inception.
The team that Lao Huang liked was actually a subnet team called Covenant AI.
Before explaining the ins and outs of the event, we need to understand the subnetting architecture of Bittensor.
Bittensor is a decentralized machine learning network with token incentives at its core. Bittensor uses a subnet mechanism to allow different teams to build various AI task markets, and miners and validators jointly participate in the computation and evaluation to distribute TAO rewards.
The "subnet team" mentioned by Palihapitiya earlier is actually called Covenant AI (formerly Templar), and the model that Huang praised is called Covenant-72B. It is a model with 72 billion parameters, which is collaboratively trained by more than 70 independent contributors on general-purpose hardware in a permissionless manner. It is also the largest decentralized large model pre-training project in history.
In short, Bittensor can essentially be understood as the underlying infrastructure for projects like Covenant AI, responsible for providing incentives, governance, and network rules, rather than directly developing specific AI models or applications; while subnets like Covenant AI are more like "application layer builders" that provide specific AI tasks and model capabilities on the underlying network.
Covenant AI's Breaking Announcement
On the morning of April 10, Covenant AI founder Sam Dare suddenly issued a statement (considering the continuous decline of TAO, the actual conflict may have been brewing for a longer time), stating that Covenant AI has decided to withdraw from the Bittensor network because Bittensor and its founder Jacob Steeves (online name Const) violated the concept of decentralization.
In its statement, Covenant AI pointed out that the team’s core belief is that “the training of cutting-edge AI models should not be controlled by any single entity.” However, when a single actor can suspend subnet emissions, overturn the subnet owner’s management rights over their own community space, publicly abandon projects without a process, and use token sales as a coercive mechanism to force others to comply, this is not decentralization, but centralized control disguised as decentralization.
Covenant AI further alleges that every participant in the Bittensor ecosystem—miners, validators, and investors—should understand that this power does exist and has been exercised by Const. Const exercises this power not for network health, but to regain control of a team that has become "too independent" and difficult to manage—a subnet owner capable of building its own community, making independent decisions, and operating without permission, because this threatens his power throughout the ecosystem. Specifically, while Bittensor employs a so-called "three-person leadership" structure, where three people manage multisignature for network upgrades and claim it as distributed governance, this is not the case. Const effectively retains absolute power and resists any genuine transfer of power—power in the Bittensor ecosystem has never left one person's hands.
Covenant AI also mentioned that in the past few weeks, Const has taken a series of actions against the team’s operations that conflict with the principles claimed by Bittensor, including suspending emissions from the Covenant AI subnet, removing the team’s administrative rights over its own community channels, unilaterally abandoning the subnet infrastructure, and exerting economic pressure by conducting a large-scale public token sell-off during the operational conflict.
Therefore, Covenant AI has decided to withdraw from the Bittensor network. The team concluded by stating that decentralized, permissionless AI training is not a feature unique to Bittensor, but rather a technological capability that the Covenant AI team hopes to continue developing. Covenant AI's research, team, models, and vision will continue to move forward, and there are already very exciting projects underway that will soon be announced to the public.
The conflict became public, and Bittensor was caught in a media storm.
Due to the success of Covenant-72B (SubNet-3), coupled with the fact that the Covenant AI team also runs two other key subnets, Basilica (SubNet-39, positioned as an AI model evaluation/inference-related subnet) and Grail (Sub-81, positioned as a more complex task-driven AI subnet), the team holds a pivotal position in the Bittensor ecosystem. Perhaps it is precisely because of Covenant AI's increased influence in terms of community, resources, and voice that the "power struggle" conflict with Const has been triggered.
With the conflict between the two sides becoming public, the Bittensor ecosystem quickly fell into a media storm.
From a product perspective, with Covenant AI's departure, the community has begun to question the future development and value of the Bittensor network. As one of the teams with the most technical narrative and practical achievements in the current Bittensor ecosystem, Covenant AI's exit means that this capability chain has been directly removed. Bittensor's technological progress in AI model training and its ecosystem activity will face uncertainty, and the market's judgment on its long-term value has therefore become more cautious.
Reputational damage is impacting Bittensor's decentralized narrative, which is facing its biggest challenge since its inception. Covenant AI's accusations directly target Bittensor's core narrative—the "decentralized AI network." For Bittensor, which relies on its decentralized narrative to attract developers and computing power participants, the impact of this governance controversy extends far beyond short-term price fluctuations; it is more likely to shake the confidence of ecosystem participants.
From a brand perspective, Covenant AI used this controversy to gain a significant advantage over Bittensor in the community's mindshare. Before this statement, the general impression of Nvidia's endorsement was that it praised Bittensor, with few realizing that Covenant AI was the real protagonist, and even fewer knowing of the team's existence. As the incident unfolded, Covenant AI's brand awareness grew, while Bittensor became the one perceived as "bleeding" in the community's eyes.
As of this writing, Bittensor's official social media accounts have not commented, while Const offered a vague response on his personal account: "This will drive the emergence of Bittensor's first truly 'headless' (presumably referring to subnets not relying on a single team), truly commercial subnets... Thanks to Covenant AI for making Bittensor more decentralized."
Below Const's response, a large number of Bittensor community users (especially TAO holders) are urging Const to provide a more detailed response to the allegations made by Covenant AI, but Const has not yet replied.
Odaily will continue to follow this story, so please stay tuned.
