The AI token market cap climbed 7.5% to $27.3B as regulatory momentum from Coinbase, Uniswap, and new US staking guidance lifted overall sentiment. Mindshare nudged up to 28%, with capital rotating into ICP, NEAR Intents.

Privacy-aligned intent systems moved into focus as NEAR Intents gained traction. Robotics also climbed up the narrative stack with Peaq’s robo-farm launch.

Now this is a chart that makes me sit up and pay close attention..

NEAR Intents just printed its strongest week ever.

Fees climbed toward $400K a day, volumes reached $200M, and total flow pushed past $4B after adding another billion in just eight days. It’ll likely be past $5B by the time you read this.

Still, the past week showed these Intents can absorb cross-chain demand at scale. They have been pulling a lot of attention for that reason alone, and the integrations keep coming: Ledger just added support last week.

A year ago, I remember everyone questioning whether any of the intent stuff would scale. The volume and integrations make that clear now: it did.

Markets don’t move on charts. They move on people…doing things.

Deep3 Labs just unveiled Hoku, their AI powered personalized token recomendation engine.

Hoku runs more than 50 proprietary machine learning models that dissect wallet clusters, liquidity migrations, and behavioral patterns across chains. The goal: forecast the next moves of the wallets that actually move markets.

You connect your wallet, and Hōkū finds your “trader twin” network: people who take similar risks, ape into the same themes, or exit at the same moments.

From there, it starts predicting what your cluster will do next. Instead of general market sentiment, you get micro-forecasts tuned to your own trading fingerprint. The system watches billions of wallet events across different chains, then surfaces signals before capital flows.

The cool part is seeing it all in 3D: a galaxy of wallets, glowing brighter as capital rotates between clusters. There’s also a cluster leaderboard that shows which trading groups and tokens are gaining traction fastest.

If its accuracy numbers hold (they claim 90%), this could be a cheat code for tracking on-chain narrative rotation before it hits CT. The clustering depth alone makes Hoku one of the more interesting DeAI tools I’ve tested this month.

It won’t replace your trading instinct, but it can surface shifts in your cluster before they show up elsewhere.

Are you building something awesome in crypto × AI? Or spotted a startup or product that more people should know about? Fill in this form and share it with us. We’ll feature the sharpest picks in the newsletter each week.

Google just pulled a quiet mind-bender: A new way of learning.

They call it Nested Learning, and it turns the deep learning stack inside out. Instead of one giant network learning in a straight line, picture a model made of smaller minds, each updating at its own rhythm, layered like Russian dolls.

It’s a direct shot at one of AI’s oldest flaws: forgetting what it once knew. Nested Learning trains models to remember like a brain does, balancing fast-changing short-term memory with slower, stable long-term knowledge.

Their prototype, Hope, takes this from concept to working system. It self-modifies by organizing knowledge into layers that age at different speeds. The surface absorbs new data; the deeper strata consolidate experience. The result is a living memory that updates without overwriting itself.

In tests, Hope beat standard transformers on reasoning and long-context tasks, recalling details most models lose halfway through. That’s big. To me, this feels like a foundational turn. It’s a re-architecture of learning itself.

👋 ICYMI: Last week’s deep dive was on TinyFish, a stealth startup that’s raised $47M to repair the broken corners of the internet we’ve all quietly accepted.

I only found it while tracing an agent-related rabbit hole, and it turned out to be one of the most compelling builds I’ve seen. Worth a look

Siam Kidd pitches TAO as GOOG and subnets as Alphabet’s stack. His takeaway: Bittensor isn’t one use case, it’s the world’s best at 17 and counting.

Grok’s agents delivered 36.6% APY, avoided losses, and out-executed every human-managed fund in the test.

Russia’s first humanoid robot fell on stage. I almost felt bad, until i read what it could do.

Recent AI launches are down 50–85 percent across the board, with only a few names holding up.

She held a mixed-reality ceremony after deciding the AI understood her better than her real partner.