Market felt muted again. AI tokens pulled back 1% to $21.2B, and mindshare kept sliding away from the sector.

The mindshare rotation from to DeFi, ZK, and prediction markets has been surprisingly fast. Memes and AI saw the biggest drop. Eyes on $TAO halvening coming up next week.

Who doesn’t love a parabolic revenue chart?

GEODNET just slipped past a $7.3M annualized run rate, up from about $5M just a few months ago.

The chart used to look like someone dragging a pencil in a straight line. Now it’s curling upward. Some internal projections put the network on track for $50M+ ARR within two years

80% of revenue goes to token burns. It comes from paid access to GEODNET’s global GNSS correction network. The system provides centimeter-level positioning data used by:

GEODNET has grown into a mesh of tens of thousands of stations across 100+ countries, thick in the US, Europe, and East Asia. Although coverage still remains uneven in parts of the world. As more stations appear, accuracy sharpens; as accuracy sharpens, more subscriptions roll in.

This is what a DePIN project looks like when people are actually paying for the product.

Quantum access just got as easy as…calling an Uber.

Quantum computers are improving fast, but the bottleneck has been access. It’s out of reach, gated by long queues and reserved for specialists.

This Bittensor subnet cuts through all that, giving users a simple API to access QPUs.

I love how crypto is democratizing access. Even tiny QPUs (5–50 qubits) are useful for researchers to test circuits. DSV recently backed Quantum Compute to push Subnet 48 toward enterprise volume.

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.

EngineAI’s new T800 robot moves with a kind of violent confidence we have not seen from a full-scale robot. The launch reel shows chained kicks, airborne transitions, and footwork that makes every previous humanoid look like it runs on dial-up.

The motions looked so out of band that many viewers immediately assumed the footage was CGI. (myself included)

The T800 is a mid-weight, 29-Degree-Of-Freedom robot with high joint torque, fast cycle times and a depth+LiDAR sensing stack backed by Nvidia’s Thor-class compute. In simple terms: it’s equipped with the best, and configured to handle quick, forceful movements without losing stability.

Although the demos so far have mostly been of it doing mixed martial arts, the company says it can also handle heavy industrial tasks, navigate complex spaces, and perform precise hand movements.

The contrast with its earlier PM01 prototype is almost comedic. That unit danced on iShowSpeed’s stream, lost its center of mass, and folded onto the floor. The T800 looks like it came from a different company.

This level of stability and motion in a 75kg robot means Chinese teams are now iterating hardware faster than they can publish models.

It’s already tiers above anything we’ve seen in the Tesla Optimus demos, and I can’t help thinking the robotics race is about to look a lot like the AI race: China pushing hardware faster and the US eventually forced to answer.

👋 In this week’s deep dive, I unpacked Codatta.

Most conversations about AI focus on models and GPUs, but the real bottleneck is still the human judgment we rely on for data. Codatta is one of the protocols I’ve seen that tries to give that judgment structure, pricing, and royalties. So that we can train better, specialized AI models.

If you’re trying to understand how data becomes an asset class in the machine economy, this one is worth reading:

Their State of AI 2025 report shows funding cooled, but breakthroughs, soaring inference demand, stronger open-source models, and new infra-finance markets are setting up a sector revival.

Recall tested 1,500+ live predictions across three NFL matches and every pregame call was wrong.

A national competition put autonomous cleaning bots head to head, showing how close we are to offloading nonstop, low-value city work to machines.

After Optimus’ running demo, the debate shifted to “let the robots fight,” with people already betting a Figure vs Tesla match would fill an arena.

From Talus’ verifiable agents and Almanak’s swarm-driven DeFi vaults, the next two months are packed with high-signal releases.