According to rootData’s Hot Index, here’s what’s buzzing:

🧭 Wayfinder just launched its token, PROMPT, yesterday. We’re letting the dust settle. Full breakdown coming next week.

🧠 Mind Network’s FHE token just hit the markets. They’re bringing Fully Homomorphic Encryption (FHE) to Web3, meaning your data and AI computations can stay private even while being used.

🐦 Meanwhile, Flock dropped some head-turning numbers from their AI training platform: Delegating yields are pushing 600% among other stats (more on this later!).

Over in Bittensor land, a few of the newer subnets are moonwalking:

• Agentao (SN62) — up 251%

• Storb (SN26) — up 211.4%

• Gaia (SN57) — up 193.8%

Some are saying there is no bear market in Bittensor subnets. Why? TAO stakers are rotating out of root as yields cool off—and diving into subnets instead.

Is this the next rotation before the rest of the market catches on?

AI Training on FLock Grows

Since launching in December 2024, we noticed that FLock’s network has quietly grown:

• 600+ nodes

• 1,000+ validators

• 5,000+ models hosted on AI Arena with 450K validations logged

• 63K community members.

So…FLock is building a decentralized AI training platform that puts privacy first.

Instead of sending your data to a central server, FLock uses federated learning to keep it on your device (secure and encrypted) while still contributing to model training.

The protocol runs on Base and uses the FLOCK token for payments, staking, and community governance.

At the center of the ecosystem is AI Arena—a gamified battleground where devs compete to train the best models.

Basically, it’s Kaggle if it went full crypto and moved on-chain.

Contests run on both public and local datasets, and high-performers earn FLOCK token rewards. Weekly challenges—like translating Web3 data using Text2SQL—help keep things fresh. Over 10,000 users have participated since launch.

In January, the team introduced the FL Alliance—an initiative to make model training accessible to anyone with a laptop or tablet.

• Users contribute compute, vote on model quality, and stay in full control of their data.

• The goal: democratize access to advanced AI without compromising privacy.

Of course, the space is heating up. FLock’s not alone. Bittensor’s Gradients subnet is already pulling in the crypto-native AI crowd.

This race is just getting started.

🌴 NOVA (Bittensor Subnet 68)

Drug discovery is notoriously slow, expensive, and dominated by Big Pharma + huge amounts of red tape.

NOVA (Bittensor Subnet 68) offers a new approach.

It reframes molecular R&D as a decentralized, AI-driven search problem where anyone can compete, not just folks with billion-dollar labs

Here’s how it works:

• The SAVI 2020 database is a large vault of 1.75 billion synthesizable molecules. Miners plug in AI models or clever heuristics to scan it for potential drug candidates.

• Submitted molecules are scored by PSICHIC, an in-house evaluator.

• Every hour, the top-ranked molecule wins token rewards

Who’d guess that we could turn drug discovery into a real-time, on-chain game?

Current stats:

• 3.82% TAO emissions daily, split between miners and validators (τ 112.86 each), with τ 49.55 for the owner.

• NOVA’s clocking at a $387M FDV and is one of the top 5 subnets by mindshare this week.

• About 3.5% of the total supply is staked.

The long-term bet is bold: Open-source pharma driven by decentralized AI. All outputs are monetized: new chemical IP, target chemical libraries, and wet lab datasets.

If NOVA works, it won’t just move markets. It could literally save lives.

Project Updates

• Plastic Labs announces its pre-seed investment, launching Honcho to tackle privacy-preserving personalization and identity for AI applications, backed by Variant Fund and Betaworks.

• Nesa teases “DAIs, a new type of decentralized AI product - not agents, not dapps.” Also invites applicants for the first DAI cohort. 🤔

• Phala Network launched its TEE-MCP Server, combining secure enclaves (TEEs) with the open MCP standard to give AI agents memory, context, and tool access.

• Allora Network upgraded to v0.10.0 on Testnet, bringing on-chain storage for network inferences, better validation, and reduced volatility in topic rewards. The update sets the stage for a Dev Mainnet launch soon.

• Arena releases its iOS app—a crypto + AI platform that combines group chat, AI discovery, and built-in trading across Solana, Base, Arbitrum, Ethereum, and BNB Chain.

• Huddle01 just launched Testnet Act II: The Nexus—a major step toward decentralized, real-time connectivity. Users can now test apps like Huddle01 Meet and Farhouse, earn Huddle Points (HPs) through quests and referrals.

• Ohara just went live with a Farcaster integration—every app built on Ohara now instantly works as a Farcaster Mini App. You can create apps using natural language, and buy a stake in them.

And not to forget some Bittensor updates:

• BitMind is now live on the App Store and Play Store, bringing its AI-or-Not deepfake detection service from Telegram to native mobile.

• Rayon Labs has launched Music Gen on Chutes, enabling users to create full songs with custom lyrics in seconds using the Diffrhythm model. The feature will also be available on SQD.io soon.

AI Agents

• Theoriq AI launched Phase 1 of its Build in Public campaign with the OLP Swarm—an agentic system where Signal Agents process real-time DeFi data to optimize liquidity.

• ​SwarmNode released the LinkedIn Recruiter Agent Template—an AI agent that takes a job title and experience level, then ranks LinkedIn candidates for you. It fetches, analyzes, and shortlists profiles automatically.

• Vainguard just launched the first version of its Investment Committee Simulation Room—an interactive AI experience where two Virtuals agents, Vain and Guard, debate and evaluate crypto projects in real-time.

• Fetch.ai introduced AgentRank, a new way to surface high-quality agents in Agentverse by measuring how central and trusted they are in the network. Inspired by PageRank, it analyzes recent agent-to-agent interactions, making it easier to find the right one for your task.

Web2 AI

• Google is on fire this week.

  • 🔥 Deep Research powered by Gemini 2.5 Pro-Exp is out—and early feedback suggests it outperforms OpenAI’s version. Oh, and it’s free.

  • 🛠️ Firebase Studio launched as an all-in-one platform for building full-stack AI apps. It brings together Project IDX, Genkit, and Gemini into a seamless dev experience

  • 🤝 Agent2Agent (A2A) also dropped. It’s an open protocol for secure, cross-platform collaboration between AI agents. Tailored for enterprise, A2A supports long-running tasks, multimodal interactions, and robust authentication.

• OpenAI rolled out enhanced memory for ChatGPT, allowing it to remember past conversations for more personalized, contextual responses. This marks a major shift in human-AI interaction—turning ChatGPT into a long-term companion that adapts to individual needs, preferences, and life experiences.

• Fauna just raised $30M from Kleiner Perkins, Quiet Capital, and Lux to build safe, lightweight robots designed for human environments. Their platform combines hardware, software, and AI to accelerate embodied AGI.

• LiveKit just announced Agents 1.0 alongside a $45M Series B led by Altimeter. The update brings multi-agent orchestration, a new low-latency turn detection model, and a robust telephony stack—powering voice AI for 3B+ calls a year.

• WillBot just launched as the first AI performance marketing agent that actually watches your ads. It syncs with platforms like Meta and TikTok, analyzes ad performance. All right from Slack.

We just broke down our key takeaways from Jensen Huang’s keynote at NVIDIA GTC 2025—and why we’re still bullish.

Also, we just launched something new for absolute beginners 👇

• A 2-week email crash course to get you up to speed on AI + Crypto.

• Bite-sized explainers on core concepts like prompt engineering, RAG, and more.

Check it out and share it with a friend who’s curious but doesn’t know where to start!

🔥 Our Weekly Top 5

A leaked internal memo from the CEO of Shopify on AI

The AI has been lying to us all this while..

Chutes (Bittensor Subnet 64) is now doing 30b tokens per day on OpenRouter

LLM inference prices have fallen 9x to 900x/year, the steepest price drop with GPT-4o models

Are we entering a new era of coding?