The most important thing today is the Nvidia GTC conference—it's basically an AI version of a brief history of humanity.

The most important thing today is the Nvidia GTC conference—basically an AI-version of the history of mankind.

Even before Huang Renxun takes the stage, the amount of leaked information in advance is already enough to write a book.

Wenwen has put together three big takeaways. Come on, fat friends—follow me.

1）AI compute costs get cut by 90%

The previous-gen Blackwell is already very strong, right? Next up, Nvidia is about to announce mass production of the next-gen chip, Vera Rubin.

What’s so powerful about Vera Rubin? To put it plainly: it’s cheap.

Run the same AI model— the number of chips drops to one quarter, and the inference compute cost falls by 90%. By 90%, friends. AWS, Microsoft, and Google—the three biggest cloud providers—are the first batch to get on board.

2）Groq, bought for $20 billion last year, turns in homework today

Earlier, at an earnings call, Huang Renxun said Groq would be connected to Nvidia’s ecosystem as an expansion architecture—like how back then, buying Mellanox helped complete networking capabilities.

Groq’s LPU and Nvidia’s GPUs sit in the same data center: GPUs understand the problem, and the LPU rapidly spits out the answers.

With these two kinds of chips sharing responsibilities, Agent use-case latency gets knocked down directly.

AI Agents do work for people—one task can involve tweaking the model dozens of rounds, and each round burns inference compute, while users are waiting there. If it’s even a bit slower, the experience collapses.

Inference has two steps: first understand your question, then output the answer character by character.

GPUs are good at the first step, but for the second step—the speed and stability of output—Groq’s LPU is stronger.

Is $20 billion really expensive?

Think about it: in the future, every company will run hundreds of Agents, and each Agent will tweak the model thousands of times per day.

3）Nvidia’s OpenClaw launches, called NemoClaw

It’s an open-source platform: when enterprises install it, they can deploy AI employees to run workflows for real people—handle data, and manage projects. It’s said to already be in talks with Salesforce and Adobe.

The interesting part is that NemoClaw doesn’t require you to use Nvidia chips. Think about that logic carefully. Selling chips only makes money on the hardware layer; setting the rules is how you earn across the whole chain. Huang Renxun clearly算得门儿清—he’s done the math perfectly.

4）Huang Renxun says he wants to showcase “chips the world has never seen before”

Most likely, the next-next-gen architecture, Feynman, will make its first appearance—mass production in 2028, using TSMC’s most advanced 1.6nm process.

Also, there’s one less-discussed tidbit I find pretty interesting.

Nvidia is making laptop computer processors—two models, focused on gaming. The GPU sellers are coming to grab the CPU business too.

Wenwen, I feel like Huang Renxun is going to become a great figure of an era in the future.