🚀 Assister AI (ASRR) is launching on Binance Alpha, with trading kicking off on May 30. If you’re holding Alpha Points, you’re in luck. Eligible users can claim their airdrop as soon as the event page drops.

💰  GAIB AI just opened Cap 3 of its fundraising push for its synthetic dollar, pulling in another $35M and raising the cap to $50M. Deposits come with a 5x multiplier, and support is live across Ethereum (ETH), Arbitrum, and Base. Yes, it’s moving fast.

🌐 Donut Browser raised $7M to build the first Agentic Crypto Browser on Solana. It’s like Chrome got jacked on DeFi steroids, letting you trade, swap, and act on-chain straight from your browser window.

On the Bittensor front, subnet demand is rising. Since the dTAO upgrade went live 100 days ago, more than $300M in TAO (roughly 7.5% of the circulating supply) has flowed into subnets.

This surge isn’t just capital rotation. Subnets are quickly becoming the main engine for TAO yield. As staking dynamics evolve and Root continues to thin out, expect even more pressure on the demand side.

⚙️ IO Net Daily Earnings Double

IO net is quietly assembling one of the largest distributed GPU networks in crypto, tapping unused compute from gaming rigs, crypto miners, and idle data centers. The goal? Power AI workloads for a fraction of AWS’s cost—up to 70% cheaper.

And it seems to be working:

• 📈 $5.7M in Q1 2025 revenue, up 82.6% QoQ

• 💸 $60K+ in daily earnings, more than double from last week

• 🌍 83,000+ verified GPUs across 130+ countries

It runs on Solana, which means every job and payout hits the chain. Contributors earn IO tokens when their machines are tapped; developers burn tokens to rent compute.

GPUs are grouped by type, speed, and geography to reduce latency for high-load AI jobs like model inference and training.

Earlier this week, they shipped Retrieval Engine, a system that adds source traceability to LLM outputs. Think: built-in receipts for AI. A big deal for enterprises building AI apps where hallucinations aren’t… funny.

There’s a massive appetite for cheaper, programmable compute. IO net’s bet is that decentralized infra can meet that demand. The question now is whether it can scale sustainably and efficiently.

🌐 AIOZ Network launches AIOZ AI

AIOZ Network launched AIOZ AI V1 this week, a Web3 marketplace for building, sharing, and running AI models on AIOZ’s DePIN network.

Here’s the flow: developers upload models and datasets to AIOZ. Each one is verified, versioned, and tied to its creator. Once approved, the model is pushed out to devices in the AIOZ decentralized network, ready for training or inference.

When a task comes in, AIOZ AI sends it to a device that already has the model. That device provides compute, storage, and bandwidth, earning AIOZ tokens for every successful job. Model creators also earn rewards whenever their models are used.

Everything is tracked on-chain. Uploads, executions, and payouts are transparent. Contributors control access, set their own prices, and monitor usage in real time.

The platform includes collaboration tools like profile pages, discussion threads, and version control to support open, community-driven AI development.

Project Updates

• Venice just rolled out a $27M Incentive Fund to back builders using its private, uncensored AI API. It’s targeting dev tools, agents, dApps, and integrations building scalable, permissionless AI infra.

• Pluralis just teased a new paper on scaling decentralized training by linking up everyday consumer devices over the internet to train massive models collaboratively.

• Loud, the decentralized attention protocol built with KaitoAI, kicked off its Initial Attention Offering. The LOUD token is live, with weekly SOL rewards tied to engagement and user mindshare.

• Aethir launched the EigenLayer ATH Vault. Users can stake ATH, mint eATH, and earn rewards from real compute usage across Eigen’s AVS network.

• Dria rolled out Decentralized Batch Inference, cutting inference costs by 5x and onboarding 20K daily contributors. Its Batch API now supports synthetic data generation, offline evals, and price-sensitive jobs.

• Hyperbolic just shipped the first MCP server that lets Claude autonomously rent an H100, spin up workloads, run system commands, and shut it all down, with no human in the loop. They also hinted at a full enterprise cloud offering coming soon.

• Render Network unveiled OctaneBench 2025. Benchmark your GPU in seconds and plug your score into their calculator for exact cost estimates.

• Gensyn’s RL Swarm just leveled up. Mac users now get Metal GPU support, training stats integrate with Weights & Biases, dashboards are personalized per node, and local logging simplifies debugging.

Bittensor News

• TaoFi just launched Uniswap V3 on Bittensor EVM, with bridged USDC from Ethereum and Base using Hyperlane. Users can now bridge, swap, and provide liquidity directly at taofi.com.

• Templar completed the first open-access LLM training run for a 1B parameter model. No whitelist, no gatekeeping. Powered by Gauntlet, their incentive layer, contributors earned rewards based on how much they improved the model.

• NATIX launched StreetVision Subnet 72 on Bittensor, merging its camera-powered DePIN with decentralized AI for real-time mapping and autonomous driving applications.

• TaoBase: The U.S.-based company focuses on scaling the Bittensor ecosystem by staking TAO, backing subnets, and driving research and incubation efforts.

• FLock OFF (SN96) is now live. It compresses high-quality data for small language models and enables on-device optimization.

AI Agents

• OpenGradient just open-sourced BitQuant, the first crypto-native AI quant. After 50K beta users and 330K on the waitlist, the full agent stack is now live under the MIT license.

• Solace, a voice-native AI companion for real-time emotional support, is now live on Virtuals Protocol after taking top prize at the Virtuals hackathon. It features voice check-ins, Spotify sync, calendar awareness, and wearable integration.

• Warden’s beta is out. Just message one AI to trade, bridge, stake, and handle your on-chain tasks like a personal crypto concierge.

• Masa launched Indexed Search on its Data API, giving AI agents lightning-fast access to historical X data. It supports hybrid, semantic, and real-time search, ideal for tracking trends and surfacing context.

Web2 AI

• DeepSeek has just released R1-0528, featuring sharper reasoning, fewer hallucinations, native JSON outputs, and built-in function calling. The front-end got a glow-up too. API users? No changes needed. It’s now live on Chutes.

• Mistral launched its Agents API, giving devs everything they need to build autonomous AI agents that can code, search, generate images, and retain long-term memory.

• Perplexity rolled out Labs, a new workspace designed for complex workflows. It bundles coding, design tools, and headless browsing into one interface for building interactive apps, dashboards, and reports.

• Hume AI unveiled EVI 3—a big step toward true voice intelligence. It’s a speech-language model that understands and generates human-like voices with nuance, personality, and speed.

We’re marking one year of research with a new series that unpacks everything we’ve uncovered by digging deep into the AI and crypto frontier.

🔑 Current Unlock Progress: 4 out of 10 essays live

This week’s focus: Bittensor is the most compelling place to watch Crypto x AI unfold in the near term.

Here’s everything we’ve learned about Bittensor over the past year  👇

🔥 Our Weekly Top 5

Will AI replace software?

Chutes is scaling faster than it’s centralized counterparts ChatGPT & Gemini

Bye SEO, say hello to GEO

Research from DCG shows growing public preference for decentralized AI

Dario (Anthropic CEO) says AI could wipe out half of junior white collar jobs soon