Today, in my view, can be said to be a big day for the AI community.

The NVIDIA GTC 2026 conference opened in San Jose, USA, and Jensen Huang will deliver a keynote speech at 2 AM Beijing time. The market is waiting for the new 'Feynman' chip, which is rumored to use TSMC's 1.6-nanometer process, replacing metal wires with optical communication between chips, which can reduce energy consumption by more than 70%.

But there is one piece of news I've been watching for a long time: NVIDIA is also going to 'raise lobsters'. It's rumored that Jensen Huang will release an enterprise-level AI agent open-source platform similar to OpenClaw—NemoClaw, allowing companies to deploy AI agents in bulk. The GTC official website has already leaked that there will be a 'GTC Park' on-site, where attendees can personally create their exclusive AI assistants powered by OpenClaw, along with an image of a Q-version Jensen Huang sewing a lobster plush toy.

The comments are all guessing how amazing NemoClaw is, but I am focusing on this message, thinking about another question: Nvidia sells chips and builds platforms, then... who pays the salary for the AI running on the platform?$ROBO 🤔🤔

Does anyone still remember Jensen Huang proposed the 'five-layer AI cake'? Let me take everyone back through it: energy, chips, infrastructure, models, applications. Each layer is spending money — the energy layer buys electricity, the chip layer streams chips, the infrastructure layer builds factories, the model layer trains, and the application layer calls. But how does the money flow? Who settles the accounts for the robots at the bottom layer? Have you thought about this?

@Fabric Foundation What is supplemented is this sixth layer: the 'settlement layer' of the machine economy.

The OM1 operating system gives each robot an on-chain identity, equivalent to an employee number. The x402 protocol has revived the HTTP 402 status code — this thing was left for 'payment required' use 30 years ago, and now it has finally been awakened by USDC. Robots can handshake with charging piles, deduct fees, and settle accounts, all in milliseconds. The VPU chip reduces the cost of proving 'real work done' to two orders of magnitude cheaper than Nvidia's H100, and just a few cents can self-verify.

The key point is the PoRW mechanism — Proof of Work for robots. Up to 29.7% of the ecological tokens are only awarded to robots that actually do the work. The HGV algorithm only recognizes 'tasks paid for by real users'; even if you create ten thousand virtual accounts to brush through, you won’t earn a penny. Each robot also has a unique on-chain identity, and its behavior is traceable and verifiable.#英伟达GTC大会

Currently, the daily task call volume exceeds 25,000 times, with 12,400 active nodes and a task completion rate of 98.7%. The robots have started to 'calculate' for themselves.

Jensen Huang wants to transform Nvidia from a 'chip seller' into a platform company that 'defines the rules of AI infrastructure.' But once the infrastructure is built, someone has to keep the accounts. Fabric might be that accountant.

Nvidia launched NemoClaw, allowing AI to do more work.@Fabric Foundation What it is doing is: letting AI finish the work and be able to receive a salary, which can be said to be a closed loop in a certain sense!

Imagine the day you use NemoClaw, and AI helps you get all the work done; it might send you a message at the end of the month: 'Boss, I paid the electricity bill this month, I also sorted out the skill package upgrade, is it okay to directly transfer you the net profit of 10000?' Doesn’t that sound quite interesting?#robo