The Web3 + AI Daily #27
First large-scale LLM training run on a fully permissionless and decentralized network, BTC miners pivoting to AI, a Machine Economy report, and much more.
Hello, everyone, and welcome to all newly-joined subscribers!
This is The Web3 + AI Daily, your definitive guide to the world of Decentralized AI (DeAI/dAI)! As 0xSammy reported yesterday, the overall DeAI market cap continues to grow, jumping by 3.8% to $33.6B this week. This publication provides a front-row view to the emerging machine economy, so make sure to subscribe.
Thank you for being here! Let’s dive in.
What’s Hot in Web3 + AI?
Bitcoin Miner IREN Continues Shift to AI
IREN, a Nasdaq-listed bitcoin miner, is aggressively pivoting into AI cloud services by securing new multi-year contracts for NVIDIA Blackwell GPU deployments. The move signals a major strategic shift from being a pure miner to a hybrid of mining + AI infrastructure provider.
The company has expanded its planned GPU fleet to ~23,000 units and already locked in contracts for 11,000 of them, representing about $225 million in annualized revenue by the end of 2025. IREN’s longer-term target is to achieve over $500 million in annual run-rate revenue from its AI cloud operations by Q1 2026. To support that, it’s building out new data centers (Horizon 1 & 2), scaling its power and land portfolio, and converting its ASIC mining infrastructure to GPU capacity.
While the AI cloud business is growing, IREN continues to operate ~50 EH/s of bitcoin mining capacity.
Web3 + AI R&D and Innovation
Covenant72B: First Large-Scale Globally Distributed 72B Model
Covenant AI has announced a major milestone with Covenant72B, its first large-scale 72-billion-parameter LLM training run on a fully permissionless, decentralized network.
Built on the Bittensor blockchain, Covenant AI coordinates global participants through three interconnected platforms: Covenant AI for pre-training, Basilica for decentralized compute, and Grail for post-training reinforcement learning.
The Covenant72B run launched on September 12th and aims to process 1.2 trillion tokens. It leverages novel technologies like SparseLoCo, a communication-efficient optimizer, and Gauntlet, a blockchain-based reward mechanism, enabling participants to exchange compressed pseudo-gradients securely and transparently.
The first checkpoint is already available on Hugging Face, showing that Covenant72B outperforms previous decentralized LLM runs and is approaching parity with centralized baselines in benchmarks like ARC-C, ARC-E, HellaSwag, and MMLU.
The network currently involves over 20 participants, each contributing at least 8xB200 GPUs or equivalent, and maintains low communication overhead (around 6%). Challenges such as validator instability and peer desynchronization have been identified and mitigated. Moving forward, Covenant AI plans to release more checkpoints, optimize system efficiency, and expand participation while developing its Basilica compute rental platform and Grail post-training system.
Venice’s V2 Brings Video Generation, Tokenomics Upgrade
The permissionless and privacy-focused ChatGPT counterpart Venice.ai presented a Q4 roadmap, offering a sneak peek into Venice V2. The forthcoming upgrade will deliver video generation and a tokenomics reform.
Video generation (both text-to-video and image-to-video), supported by a new credit system and access to state-of-the-art models (e.g. Sora 2, Veo3, Kling Turbo), will soon be rolled out to beta testers. On the tokenomics side, Venice plans to introduce a buy-and-burn mechanism to reduce the supply of its VVV token and to cut emissions from 10 million to 8 million VVV per year. The goal is to gradually shift VVV toward being a deflationary asset with native yield, more tightly aligned with Venice’s overall growth.
Web3 + AI Research & Stats
Peaq’s Machine Economy Report Q3 2025
peaq published its latest Machine Economy report, covering the significant strides, pioneering initiatives, and technological enhancements the company undertook in Q3 2025.
Key Developments
Tokenized Robo-Farm: Peaq unveiled the world’s first tokenized vertical robo-farm at Korea Blockchain Week 2025. Developed by Kanaya AI Technology Limited, tokenized by dualmint, and deployed on Peaq, this Hong Kong-based farm offers token holders a share of its operating profits, with an estimated annual percentage yield (APY) of approximately 20%.
Robotics SDK Launch: The introduction of Peaq’s Robotics SDK enables developers to integrate self-sovereign identities, payment capabilities, and data verification into robots, facilitating their participation in the decentralized Machine Economy.
Blockchain Performance: Peaq demonstrated its blockchain’s scalability by achieving over 49,000 transactions per second (TPS) and a block finalization time of approximately 500 milliseconds, showcasing its capacity to handle high throughput without compromising decentralization.
Machine Economy Free Zone: In collaboration with Pulsar Group, Peaq established the Machine Economy Free Zone in the UAE, a dedicated sandbox environment aimed at fostering innovation and deployment of machine-centric applications in the region.
These developments underscore Peaq’s commitment to building a decentralized infrastructure that supports the integration of machines into the global economy, paving the way for a more autonomous and efficient future.
Web3 + AI Investment News
Crunch Lab Raises $5M to Advance dAI
Crunch Lab, the organization behind CrunchDAO, has secured an additional $5M in strategic funding, bringing its total to around $10M, to advance its decentralized AI prediction network. The round included participation from Galaxy Ventures and Road Capital, with backing from VanEck and Multicoin Capital
Crunch Lab’s platform enables data scientists to compete anonymously through encrypted modeling competitions that preserve data privacy while rewarding the most accurate predictive models.
The network has already shown strong real-world impact, contributing to cancer gene and therapy research at institutions linked to Massachusetts Institute of Technology and Harvard University. It also improved computer vision models for cancer detection, and achieved double-digit accuracy gains for the Abu Dhabi Investment Authority’s research lab. Additionally, it has supported economic modeling research by Nobel laureate guido imbens.
Positioning itself as a decentralized intelligence layer for global enterprises, Crunch Lab plans to expand into industries beyond biomedical and finance. The project was also selected for Solana Labs’ Incubator 2025 cohort, reflecting its growing alignment with the Solana ecosystem and the broader decentralized AI movement.
Web3 + AI Readings & Conversations
Polosukhin: Intents + Agentic Payments Form the Real Future of Payments
In his latest blog post, NEAR Protocol‘s Illia Polosukhin argues that most new blockchain payment networks are missing the point: on-chain transactions alone can’t rebuild the global payments system. Payments involve many steps beyond fund transfer (authorization, escrow, dispute resolution, etc.), which blockchains don’t natively handle.
Polosukhin proposes Intents and Agentic Payments as the real future of payments:
Intents = user-defined goals (“buy X,” “book Y”) that abstract the entire payment flow and can work across crypto, fiat, and real-world assets.
This creates a universal market where users pay in any currency or stablecoin and merchants receive in their preferred one, with minimal friction or slippage.
Agents (AI assistants) automate commerce: they find products, handle payments, and optimize fees for both users and merchants.
This system enables lower costs, instant settlement, yield on funds, and reduced chargebacks.
Ultimately, Polosukhin envisions NEAR Intents + AI agents as the foundation of a new, interoperable payment ecosystem that unifies crypto and traditional finance, replacing today’s fragmented, high-friction infrastructure.
Argentum AI: Optimizing AI Via Decentralized Compute
In a conversation with DL News, Dr. Clark Alexander, co-founder and head of AI at Argentum AI, presented the company’s vision of matching user jobs to compute providers in a cost- and energy-efficient way. The platform employs auction mechanics and an AI agent to optimize resource allocation, while ensuring task correctness through verification protocols and arbitration in case of disputes.
Alexander also emphasized the inefficiencies of common AI tools, pointing to Python’s energy usage as orders of magnitude worse than more efficient languages like Rust, and plans to use pricing mechanisms to incentivize more efficient compute practices. He envisions Argentum supporting a broad array of real-world scientific, industrial, and AI tasks (from molecular simulations to financial modeling) within the next year, secured by post-quantum cryptographic methods.
Moreover, Alexander discussed the physical and computational limits of large language models (LLMs), arguing that energy demand is becoming a critical bottleneck as model size and data scale grow. He advocates for smaller, specialized “SLMs” (small language models) tailored to specific domains, and sees value in chaining them through a higher-level classifier that routes user queries to the right expert model.
Thank you for reading! I hope you found this article insightful.
If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.
And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.
The Web3 + AI Book Club is live on Fable. Follow the link below and come join our literary community.
I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.
Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.
Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.