Projects We Back

Bittensor: Decentralized AI on the Blockchain

We are long-term believers in Bittensor, a decentralized protocol where machine learning models compete and collaborate on-chain. Our focus lies in the subnet layer, where independent teams build specialized domains of intelligence.

🔹 Subnets We Support:

Subnet(s)	Operator	Focus	Our Involvement
3	Templar	Decentralized pre-training of LLMs using SparseLoCo compression over commodity internet—trained Covenant72B (72B parameters) matching centralized baselines	Team Member
39	Basilica	Decentralized compute infrastructure	Team Member
81	Grail	Decentralized RL post-training for LLMs with cryptographic verification—PULSE enables 100x weight sync compression, matching centralized training speeds	Team Member
2	DSperse	Decentralized zero-knowledge proof infrastructure for verifiable AI inference	Alpha Holder
17	404-GEN	Decentralized text-to-3D generation for gaming, VFX, and virtual worlds	Alpha Holder
44	Score Vision	Decentralized computer vision network for AI video analysis—training autonomous systems to understand visual content at scale, with football/soccer as the primary training domain	Alpha Holder
56	Gradients	Decentralized fine-tuning infrastructure with enterprise partnerships for domain-specific NLP model training	Alpha Holder
62	Ridges AI	Incentivized Agentic Training	Alpha Holder
64	Chutes	Serverless model inference processing ~160B tokens daily across 8,000+ GPU nodes—Bittensor's largest inference subnet	Alpha Holder
68	NOVA Labs	AI for pharmaceutical drug discovery	Alpha Holder
75	Hippius	Decentralized cloud storage powered by a custom Substrate blockchain, IPFS, and S3-compatible storage—censorship-resistant, community-run infrastructure for AI-ready data	Alpha Holder
120	Affine	Decentralized RL evaluation platform where AI models compete on tasks like code generation and reasoning—sybil-proof, open-source, uses Chutes for model hosting	Alpha Holder
22	Desearch	Decentralized AI-powered search engine infrastructure	Alpha Holder
23	Trishool	Decentralized AI alignment and evaluation subnet—benchmarking and validating model quality across training pipelines	Alpha Holder
26	Kinitro	Incentivized competitions for embodied AI and robotics—advancing robotic policy and planning models through decentralized challenges	Alpha Holder
100	Platform	Collaborative AI research through parallel challenge tasks using sub-subnet technology and Intel TDX confidential computing	Alpha Holder

We are most excited about:

Covenant AI—developing three complementary subnets: Templar (SN3) for decentralized pre-training of large language models, Basilica (SN39) for decentralized compute infrastructure, and Grail (SN81) for RL post-training with cryptographic verification. Together, these form the core pipeline for building a Bittensor-native frontier model—from pre-training through compute to post-training. As a member of the Covenant AI team and token holder in all three subnets, I'm directly involved in advancing this vision.

Chutes (SN64): serverless model inference processing over 160 billion tokens daily across 8,000+ GPU nodes. Bittensor's first $100M+ subnet, underscoring the scale and adoption of decentralized AI infrastructure.

Gradients (SN56): on-demand decentralized fine-tuning infrastructure with enterprise partnerships. Anyone can fine-tune text and image models without managing compute, at costs well below centralized cloud providers.

Hippius (SN75): decentralized cloud storage powered by a custom Substrate blockchain, IPFS, and S3-compatible storage. We're actively using Hippius for our own infrastructure needs, and that's always a good sign. Censorship-resistant, community-run storage is essential infrastructure for the decentralized AI stack.

404-GEN (SN17): decentralized text-to-3D generation for gaming, VFX, and virtual worlds. The quality of the 3D assets is strong, and the subnet is underpriced relative to the scope of the problem. As game engines and virtual environments demand procedural 3D generation, a decentralized pipeline matters.

NOVA Labs (SN68): bringing decentralized science to drug discovery. NOVA uses AI to accelerate pharmaceutical research by incentivizing miners to discover novel molecular compounds. This represents the intersection of two movements we care deeply about: decentralized AI infrastructure and DeSci. Traditional drug discovery is notoriously slow and expensive, dominated by a handful of pharmaceutical giants. NOVA opens this frontier to permissionless participation, where anyone with compute can contribute to finding the next breakthrough treatment.

Polkadot

Polkadot continues to stand out as one of the most principled and forward-thinking ecosystems in Web3. Built on Wasm-based runtime environments and a modular, forkless upgrade system, its architecture is designed for truly sovereign blockchains to interoperate seamlessly.

We're especially excited about Polkadot's next evolution: the Join-Accumulate Machine (JAM), a next-generation scalable compute fabric that brings Ethereum-like smart contract functionality and Rollup-style scalability directly into the Polkadot ecosystem. JAM not only preserves Polkadot's core principles of security and decentralization but extends its potential to support a universal, multi-runtime network.

Polkadot Parachain Projects We Support:

• 

  PEAQ
  Powering the Machine Economy and enabling DePIN infrastructure for autonomous machines and dApps.
• Frequency: A Polkadot parachain enabling decentralized social media, backed by Project Liberty and the DSNP protocol.
  • We stand behind initiatives like the People's Bid to reclaim TikTok and decentralize the social graph.
• Ex Machina DAOs (DEUS): We were Genesis OG holders of the DEUS token. After thoroughly reviewing the legal structure and terms, we've decided to exit our position at TGE. Read our full analysis: Why I'm Exiting DEUS at TGE.We remain excited about the robotics industry and will be seeking alternative ways to gain exposure to humanoid robotics and physical AI companies.

🤖 Robotics: On Our Radar

• OpenMind: Building the universal operating system for intelligent machines. OpenMind offers OM1, an open-source, hardware-agnostic OS for robots (humanoids, quadrupeds, drones), and FABRIC, a decentralized AI control layer enabling robots to verify identity, share context, and coordinate securely across environments. Founded by Stanford Professor Jan Liphardt and backed by Pantera Capital, Coinbase Ventures, and Lightspeed Faction ($20M Series A). We're not yet investors or token holders, but we're watching this one closely as a potential path to robotics exposure without the governance complexities we encountered with DEUS.

🔒 Privacy Infrastructure

Aztec Network

Aztec is a privacy-first Layer 2 on Ethereum: a fully programmable private world computer. After eight years of development, the network launched fully decentralized from day one with permissionless validators, decentralized proving, and community governance.

What makes Aztec different: every transaction is a zero-knowledge proof by default. Smart contracts can have private and public components that compose atomically. Users generate proofs locally, so sensitive data never leaves their device. Through "private intents," users can access any DeFi protocol on any L2 or Ethereum mainnet while maintaining full privacy.

The project also funded ZK Passport, a breakthrough in private identity verification using NFC chips in government passports. With AI deepfakes about to break traditional KYC, cryptographic identity may be the only verification that survives.

We also hold alpha tokens in DSperse (Bittensor SN2), which operates the largest decentralized zkML proving cluster in the world. ZK proofs for AI inference are the natural complement to private transactions: verifiable computation without exposing the underlying data.

BIO Protocols & Decentralized Science

We actively support DeSci (Decentralized Science), a movement changing how research is funded, verified, and shared. These DAOs and protocols fund public-good biotech, regenerative medicine, and mycology science.

BioDAOs We Support:

Project	Focus Area
VitaDAO	Longevity research, Matrix Bio, and the Longevity Prize
PsyDAO	Psychedelic science and mental health governance
MycoDAO	Fungal science, sustainability, and the LAB-IN-A-BOX program
ReflexDAO	Health data systems and research funding models

These DAOs represent a return to science as a commons, where peer-review and access are decentralized.

🤖 AI-Native Economies & Autonomous Agents

As AI-native protocols emerge, we're laying groundwork to develop, fund, and coordinate autonomous agents that are free, composable, and governed by users rather than centralized APIs.

Our Alignment:

• AGNO: Currently researching this AI agent platform for its potential in autonomous task execution and decision-making.
• Basilica OpenClaw / 8004.org: We're building a Bittensor-native agent that runs entirely on decentralized infrastructure: Basilica for compute, Chutes for inference, Hippius for storage, Desearch for search. Built on the OpenClaw codebase and Tau Ninja (an agent framework from Bittensor founder const). Covenant AI is developing subnet plugins to make this stack easier for others to adopt. Build your own.
• Tibbir: We liquidated our speculative token position to consolidate into Bittensor, but remain interested in the AI agent coordination space. Read our original analysis.

We see these platforms as tools for coordinating machine labor and agency in the open web.

We disclose these positions because transparency matters. If we write about a project, you should know whether we hold tokens in it.