Bittensor, a blockchain-based decentralized AI network, has moved closer to the mainstream after receiving high-profile public attention from investor Chamath Palihapitiya and Nvidia CEO Jensen Huang. The discussion centered on Covenant-72B, a large language model developed through Bittensor’s ecosystem, and has reignited debate over whether distributed model training can become a serious alternative—or complement—to traditional AI development built around centralized data centers.
A Crypto-Native AI Project Breaks Into the Broader Tech Conversation
The project gained visibility during an episode of the All-In podcast, where Chamath Palihapitiya pointed to Bittensor’s work as a concrete example of decentralized AI moving beyond theory. He described the effort in simple terms: a large language model trained without relying on a single centralized infrastructure provider, instead using a network of independent contributors who supplied excess computing power.
Palihapitiya said the achievement was technically remarkable, comparing it to familiar distributed computing efforts that coordinated idle hardware from participants around the world. That framing matters because it makes Bittensor’s approach easier for a broader audience to understand. Rather than seeing decentralized AI as an abstract crypto narrative, the podcast discussion presented it as a real engineering model with measurable output.
Bittensor operates as a peer-to-peer network powered by blockchain incentives, aiming to create a marketplace where AI models, machine learning outputs, and compute resources can be exchanged and rewarded. That places it at the intersection of crypto, open-source AI, and distributed infrastructure—three areas that have increasingly overlapped as demand for model training and inference continues to grow.
Covenant-72B Pushes the Boundaries of Distributed Training
At the center of the discussion is Covenant-72B, a model developed through Bittensor’s Subnet 3, known as Templar. According to the source material, the model was trained across a decentralized network involving more than 70 contributors connected through standard internet links, without a central authority coordinating the entire process in the way a conventional hyperscale data center would.
From a technical standpoint, the numbers are notable. Covenant-72B is reported to contain 72 billion parameters and to have been trained on approximately 1.1 trillion tokens. The training process relied on techniques such as compressed communication protocols and distributed data parallelism, making it possible to coordinate model development outside the walls of traditional centralized AI infrastructure.
That is what makes the story more than just a novelty. Distributed AI training has often been viewed as conceptually interesting but operationally weak compared with the tightly integrated clusters operated by major AI labs and cloud providers. The Bittensor effort challenges that assumption by suggesting that, under the right architecture, decentralized participants can contribute meaningfully to large-scale model creation.
Jensen Huang’s View: Open and Closed AI Will Coexist
Jensen Huang did not frame decentralized AI as a direct replacement for proprietary systems. Instead, he offered a more balanced view of the market, saying the future is not “A or B,” but “A and B.” In other words, open and decentralized approaches can grow alongside closed, polished commercial systems rather than eliminating them.
That statement is significant because it comes from the head of Nvidia, the company most closely associated with the hardware backbone of the modern AI boom. Huang’s remarks suggest that decentralized model development is not being dismissed outright by one of the most influential executives in the sector. Rather, it is being acknowledged as one part of a broader AI stack that will likely contain multiple business models, technical frameworks, and deployment patterns.
Huang also emphasized that models should be viewed as a technology, not as the final product itself. For most mainstream users, highly polished systems such as ChatGPT, Claude, and Gemini are likely to remain the preferred interface. These products offer reliability, usability, and broad-purpose functionality that ordinary users generally want without having to build or customize a model themselves.
At the same time, Huang pointed to areas where customization and control are essential. In certain industries, domain expertise must be embedded into a model in ways that organizations can directly manage. According to him, that kind of requirement can only be met through open models. This is precisely where decentralized or open ecosystems such as Bittensor may find durable relevance.
Why the Milestone Matters Beyond Crypto
The importance of Covenant-72B is not simply that it exists, but that its reported benchmark performance places it in the same conversation as established centralized models. The source notes that performance testing suggests the model is competitive with already recognized centralized systems. If that view holds up under continued scrutiny, it strengthens the case that decentralized training can move from experimentation into practical deployment.
That possibility has broad implications. First, it challenges the assumption that meaningful AI progress must always flow through massive, tightly controlled data-center environments. Second, it opens the door for more flexible compute coordination, where geographically dispersed contributors can help train sophisticated models. Third, it may provide a path for organizations that value sovereignty, cost efficiency, or infrastructure independence.
The appeal is especially strong for sectors that do not want to depend entirely on black-box commercial AI providers. Enterprises in regulated industries, research institutions, or specialized technical fields may prefer models they can inspect, adapt, or align more closely with internal requirements. A decentralized architecture will not automatically solve these needs, but it can support a different governance and incentive structure than centralized cloud-based AI alone.
Market Reaction and Strategic Implications
The market responded quickly. According to the report, Bittensor’s native token, TAO, rose 24% after clips of the discussion between Palihapitiya and Huang circulated on social media. While price reactions do not determine the long-term value of a technology, they do show how strongly investors are watching the intersection of crypto and AI for signs of external validation.
That validation appears to be evolving from a niche narrative into a more serious strategic discussion. Huang’s comments suggest that the bigger story is not pure disruption, but coexistence. Proprietary AI platforms are still likely to dominate general consumer use, while decentralized and open models may gain ground in specialized, cost-sensitive, or sovereignty-driven applications.
For startups, Huang outlined what appears to be a pragmatic roadmap: begin with open-source foundations and then layer in proprietary advantages over time. That model reflects what many venture-backed AI companies are already doing—using open systems to accelerate development and adoption before differentiating through productization, domain expertise, distribution, or private enhancements.
A Hybrid AI Future Comes Into Focus
The Bittensor episode highlights an increasingly important shift in AI infrastructure thinking. The future may not belong exclusively to giant centralized labs, nor solely to fully decentralized networks. Instead, the industry may move toward a hybrid structure in which different architectures serve different needs.
In that framework, centralized systems will continue to excel where integration, scale, and user experience matter most. Open and decentralized systems, meanwhile, may become especially relevant where transparency, customization, governance flexibility, and alternative compute coordination are needed. The significance of Huang’s endorsement lies less in declaring a winner than in recognizing that both tracks are likely to matter.
For Bittensor, the moment is consequential. A project once discussed mostly in crypto-native circles is now being referenced in a wider conversation about how AI should be built. Whether decentralized training ultimately captures a large share of the market remains uncertain, but the attention from Palihapitiya and Huang suggests it has crossed an important threshold: it is no longer easy to dismiss as merely an experiment.