Jensen Huang Drops a Bombshell Again! Humanoid Robot + PC Chip, AI Hardware Has Completely Changed

Hello, I'm Wanfeng, a programmer who writes about AI.

June 1, Taipei.

Jensen Huang stood on the GTC Taipei stage and launched two big moves in a row, blowing the AI hardware world apart.

After watching the livestream, I had only one feeling:

The war in AI hardware has shifted from "compute power" to "the edge."

1. First Move: NVIDIA x Unitree, the Humanoid Robot H2+ Arrives

What Is It?

Jensen Huang announced: NVIDIA has partnered with Unitree Robotics to launch the next-generation humanoid robot reference design H2+, also known as the Isaac GR00T system.

Specific specs:

Total height: about 1.8 meters
Total weight: about 68 kg
Robot body: 31 degrees of freedom
Each mechanical hand: 25 degrees of freedom

What does that mean?

An adult human has about 250 degrees of freedom in total.

This robot's upper body (torso + two hands) is already close to a human's.

What does that imply?

It implies that the actions it can perform are already similar to those of a human.

Why Build This?

Honestly, the humanoid robot race has been heating up for years.

Boston Dynamics, Tesla Optimus, Xiaomi CyberOne… everyone is building them.

But what Jensen Huang is doing this time is not "building his own robot."

He is building a "reference design."

What does that mean?

It means NVIDIA bundles the brain (AI chip) + body (robot reference design) so that any company wanting to build a humanoid robot can use it.

Just like the Android system back then — Google didn't make phones, but every phone maker used Android.

What Jensen Huang wants to build is "the Android of the robot era."

What Does It Mean for Us?

Honestly, humanoid robots are still a bit far from ordinary consumers.

But they are already very close to industrial applications.

I checked, and Unitree's humanoid robots have already started testing in several scenarios:

Car factory assembly lines: tightening screws, moving parts
Warehousing and logistics: carrying boxes, sorting goods
Hazardous environment operations: nuclear power plant maintenance, fire and rescue

This isn't science fiction.

This is happening right now.

If you work in manufacturing or logistics—

Your human-machine collaboration partner may turn into a humanoid robot within 3 years.

2. Second Move: RTX Spark, a PC Chip Built for AI Agents

What Is It?

Jensen Huang unveiled RTX Spark (N1X) in his keynote—a PC chip designed specifically for AI Agents.

Key specs:

Architecture: Based on the Arm architecture (not x86)
System: Runs Microsoft Windows (yes, Windows on Arm)
Positioning: Local compute for personal AI agents

What does that mean?

Simply put: Your next computer may not be "Intel Inside," but "NVIDIA Inside."

And this chip is not for binge-watching or gaming.

It is for running local tasks for AI Agents.

Why Build This?

There's a key point here that many people miss:

Jensen Huang said: "The AI era has truly arrived."

He's not bragging.

The context in which he said this is: AI Agents have started to be deployed at scale.

What is an AI Agent?

It's an AI that can autonomously complete multi-step tasks.

For example:

You tell the AI "book me a flight to Shanghai tomorrow," and it checks flights, compares prices, places the order, and adds it to the calendar by itself.
You tell the AI "write me a competitive analysis report," and it searches for data, organizes it, writes the article, and adds charts by itself.

These tasks cannot be handled by simple Q&A.

They require the AI to run continuously on your local machine, calling various tools, and completing a full task.

That requires compute—and it's running on your own computer.

Why Does NVIDIA Want to Make PC Chips?

Actually, the answer is pretty obvious:

Because the battlefield of AI Agents has shifted from "the cloud" to "the edge."

Over the past 2 years, all AI companies have been racing on "cloud-based large models":

ChatGPT runs on Microsoft's cloud
Claude runs on Amazon's cloud
ERNIE Bot runs on Baidu's cloud

But the cloud has several issues:

Privacy: Your data has to be uploaded to someone else's server
Latency: Every request requires a network round trip, which is slow
Cost: Cloud inference is expensive, billed by token

So the next stage of AI Agents must be "edge-first, cloud-secondary."

In other words: simple tasks are handled by local AI; only complex tasks ask the cloud for help.

And what does "edge AI" need?

It needs compute—a local chip capable of running AI Agents.

That's why NVIDIA is making the RTX Spark.

3. What Does This Mean for Us?

Honestly, when Jensen Huang dropped these two moves, my first reaction was:

"I might need to replace my next computer."

But that's not the most important thing.

The most important thing is: the AI hardware battlefield has completely changed.

1. From "General-purpose Compute" to "AI-dedicated Compute"

For the past 30 years, the logic of PC chips has been "general-purpose compute":

Intel: my CPU has a higher clock speed
AMD: my multi-core performance is better

But in the AI era, you need more than "general-purpose compute."

You need "compute that efficiently runs AI models."

That's where NVIDIA has the advantage—it started out making GPUs, which are inherently designed for parallel computation.

Now, it has packed GPU capabilities into PC chips.

2. From "Human-Computer Interaction" to "Agent Interaction"

In the past, the way you used a computer was:

Open an app
Click around the interface
Complete a task

But in the AI Agent era, the way you use a computer will become:

You say a sentence to AI
AI opens apps, operates them, and completes the task by itself
You just look at the result

What does that mean?

It means the form of "apps" will change.

Before: Apps were independent pieces of software that you opened one by one.

After: Apps may just be "capabilities" that AI Agents call on demand.

This is the core logic behind OpenAI's "app-less phone" demo—phones no longer install regular apps; interfaces are generated in real time by on-device local models.

Jensen Huang's RTX Spark is preparing for that future.

3. From "Tool" to "Companion"

The release of the H2O+ humanoid robot actually points to a longer-term future:

AI is no longer just an app on your phone.

AI will become a "physical presence"—a humanoid robot that can do things for you in the real world.

This isn't science fiction.

Unitree's humanoid robots are already working in factories.

Jensen Huang's Isaac GR00T system is providing a "brain" for all humanoid robots.

Within 5 years, you may really see humanoid robots working at home or in the office.

4. What Should We Do?

Honestly, my feelings about these two pieces of news are mixed.

On one hand, I think it's exciting—

The pace of AI hardware evolution is much faster than I imagined.

On the other hand, I think it's urgent—

If your job involves "sitting in front of a computer and using software to complete tasks"—

AI Agents + humanoid robots will completely change the way you work within the next 5 years.

Concrete Action Steps:

Step 1: Start using AI Agents

Download Cursor and experience AI-assisted programming
Use Claude and experience AI doing research for you
Use Windsurf and let the AI Agent complete tasks autonomously

You don't need to fully understand the principles, you just need to start using them and feel "what AI Agents can do."

Step 2: Learn AI programming

The essence of an AI Agent is "an AI that can call tools"
If you can write your own tools and connect to APIs, you can let AI Agents do more complex things
My Bilibili course has free previews (link at the end)

Step 3: Follow AI hardware progress