Author: Max.S
Capital markets' faith in 'Decentralized AI' (DeAI) is facing an unprecedented stress test.
Recently, Bittensor ($TAO), the absolute leading project in the decentralized AI sector, experienced a highly destructive internal earthquake. Covenant AI, one of the top development teams within the Bittensor ecosystem, which had just successfully trained a 72B large language model, suddenly announced via social media its complete withdrawal from the Bittensor network. In its exit statement, Covenant AI directly targeted Bittensor founder Jacob Steeves, vehemently criticizing his "absolute and dictatorial" control over the network, accusing him of arbitrarily cutting off token rewards for subnets, and bluntly stating that the so-called decentralized AI is nothing more than an elaborately staged "charade."
Affected by this black swan event, the $TAO token price faced panic sell-off in the secondary market, plummeting 15% to 25% in a single day, with its market capitalization evaporating by hundreds of millions of dollars instantly. While the crypto community was "enjoying the drama" of the public fallout between a top team and the founder, it also began to seriously examine a deeper industry proposition: In the AI field, which heavily relies on computational power capital and complex engineering, is the token-economics-driven "decentralization" a utopia that reshapes production relations, or a华丽的 facade masking centralized power?
To understand the destructive power of this event, one must first recognize Covenant AI's weight within the Bittensor ecosystem.
In Bittensor's multi-subnet architecture, most subnets are still in the low-level stages of API calls, model fine-tuning, or simple task routing. Teams truly capable of training models from scratch or handling large-scale parameter models are extremely rare. Covenant AI was a "hardcore" representative in this ecosystem. Just before announcing their exit, the team had delivered a milestone achievement to the community: successfully training a 72-billion parameter (72B) open-source large model in a decentralized network environment.
Given current computing power costs, training a 72B model means mobilizing a massive GPU cluster (typically equivalent to thousands of H100s running for weeks) and incurring extremely high hardware and electricity costs. The core logic behind Covenant AI's willingness to bear such huge upfront sunk costs lay in Bittensor's "Emissions" mechanism—as long as the models and computing power they provided scored high in subnet evaluations, they would continuously receive $TAO token emissions as substantial rewards. This is precisely the most attractive flywheel effect in the DeAI narrative.
However, the flywheel stopped abruptly at its peak. According to Covenant AI's disclosure, after they spent huge sums to complete the 72B model training and deployment, founder Jacob Steeves and his stakeholders, by controlling validator nodes, cut off the token rewards flowing to Covenant AI's subnet without any warning or transparent governance process.
For miners and developers, cutting off Emissions is tantamount to "pulling the plug." The ROI on massive computing power expenditure instantly dropped to zero. This highly unpredictable systemic risk directly triggered Covenant AI's angry departure.
The word "Charade" used by Covenant AI in its exit statement precisely hit Bittensor's most vulnerable nerve: network control.
Bittensor's underlying design relies on the Yuma consensus, whose core is that "validators" evaluate the contributions of "miners" and decide how the system's newly issued $TAO tokens are distributed. Theoretically, this is a decentralized game theory system based on staking amounts and algorithms. But Covenant AI's accusations reveal a harsh reality: computing power is decentralized, but power and capital are highly concentrated.
In the current Bittensor root network, the leading validator nodes that dominate the flow of token distribution have their staking chips highly concentrated among early investors, the foundation, and addresses associated with founder Jacob Steeves. This means the founder is not only the rule-maker but also the biggest referee.
Covenant AI pointed out that when a subnet's output did not align with Jacob's personal wishes, or potentially threatened the interests of other "favored" subnets, Jacob could easily use his substantial staking weight to alter the distribution outcome of the Yuma consensus. This "one-person rule" type of intervention renders the decentralization at the smart contract level meaningless. Developers spending millions of dollars on computing power ultimately found their fate dependent on the subjective will or behind-the-scenes manipulation of a single founder.
Objectively, Jacob and his supporters might defend their actions by citing reasons like "maintaining overall network quality" or "preventing specific subnets from exploiting rule loopholes to farm tokens." However, in the absence of a transparent DAO governance mechanism and on-chain hearing and appeal channels, this kind of centralized intervention, done in the name of "carrying out justice," severely undermines the network's core value as a "credibly neutral infrastructure."
$TAO's single-day plunge of 15-25% is not merely a panic-induced stampede by retail investors; it is also institutional funds repricing the "governance risk discount" for Bittensor.
The reason Bittensor could support its huge market capitalization and enjoy extremely high valuation premiums was that the market viewed it as the only realistic candidate for a "decentralized OpenAI." The foundation of this grand narrative is that the system must possess极强的 predictability: as long as you contribute computing power and quality models, the protocol will automatically guarantee your收益 through code.
The Covenant AI event shattered this expectation. Top financial professionals and institutional investors most despise "unpredictable single points of failure," and here, that point of failure is Jacob Steeves's power.
If even the absolute top-tier team capable of training a 72B model could instantly receive nothing due to the founder's intervention, then for other computing power providers and AI research institutions holding tokens and观望, deploying heavy assets on Bittensor is undoubtedly playing a game of Russian roulette where the table can be flipped at any moment. When the high-quality supply side (miners and developers) refuses to enter out of fear of centralized tyranny, the application scenarios and intrinsic value of the $TAO token become a river without a source. The frantic capital flight is a提前 vote on this fundamental weakness.
Covenant AI's departure is not just a PR crisis for Bittensor alone; it is also the inevitable growing pain for the entire decentralized AI sector as it moves into deeper waters. It残酷ly reveals the "Impossible Trinity" of the DeAI field: Model Quality & Scale, Decentralized Credible Neutrality, and Anti-Malice Incentive Alignment.
Centralization of Scale vs. Decentralization of Mechanism: Cutting-edge AI (like 72B+ large models) training is a typical heavy-capital, centralized engineering effort requiring highly coordinated GPU clusters. This has a natural physical chasm with the permissionless,分散式 nodes advocated by Web3.
Anti-Sybil (Preventing Fake Volume) vs. Credible Neutrality: To prevent low-quality nodes from farming tokens through mutual traffic brushing (Sybil attacks), the network must introduce subjective "quality assessment." But today, with AI evaluation standards not yet fully objective and mathematical, this assessment power, once handed to a few validators, can easily evolve into centralized power rent-seeking.
Bittensor attempted to use token economics to build a bridge connecting the two, but the Covenant event proves that the bridge's load-bearing pillars (governance mechanisms) remain incredibly fragile.
Covenant AI's exit popped the romanticized bubble of Bittensor's "absolute decentralization." For $TAO, this might be a painful moment of disenchantment, but for the entire DeAI industry, it is a necessary wake-up call.
