Reporting from San Jose, California at GTC 2026…

There's a moment happening right now in San Jose.

Jensen Huang—arguably the most important man in tech, the guy whose chip obsession accidentally minted trillions in wealth—is on stage at the SAP Center in front of 20,000 people from 100+ countries.

And, as we predicted last week, he's talking about a lobster: OpenClaw.

Astute Altucher Confidential readers will recognize OpenClaw as the AI agent platform that, according to Huang himself, is "probably the single most important release of software, you know, probably ever.

“This is as big of a deal as HTML,” he said on stage. “This is as big of a deal as Linux.”

Moreover, he said: “Every single company needs to have an OpenClaw strategy… This is going to become a multi-trillion dollar opportunity.”

That's a big claim. Let me explain why he's not wrong.

And the BIGGEST opportunity hiding in the weeds. (The same one James and I will talk about during our Tech Turning Point event tomorrow, with a FREE pick in tow.)

The Guy Who Built It in an Hour

Peter Steinberger spent 13 years building PDF tools in Austria.

Then, in November 2025, he sat down and built a prototype in roughly one hour.

He wasn't trying to change the world.

He was just annoyed that something didn't exist—so he “vibe coded” it into existence.

The origin story is surprisingly… boring.

Steinberger wanted a local AI agent he could talk to through WhatsApp. Not a chatbot. Not a web app. An autonomous agent that could research, draft, code, and automate tasks—all through the messaging app he already used every day. He looked for something like this.

It didn’t exist. So he built it.

That one-hour experiment became OpenClaw—the fastest-growing repository in GitHub history, with over 175,000 stars.

If you look at the adoption curve, Linux took roughly 30 years to reach similar milestones.

OpenClaw did it in three weeks.

Huang's description of the chart: "If you look at the line even in semi-log, this thing is straight up. It's vertical. It looks like the Y-axis. I've never seen anything like it."

Okay, But What Actually Is It?

Here's the part most tech journalists skip—the plain-English version.

Every AI assistant you've ever used works the same way: you ask it something, it answers, conversation over. It's reactive. It waits for you.

OpenClaw doesn't wait.

It's an autonomous agent that can execute tasks via large language models, using messaging platforms as its main user interface. You connect it to WhatsApp or Telegram. You give it access to your computer, your files, your email, your calendar. And then it just... runs.

It feels less like using software and more like texting a very capable friend who happens to have root access to your computer.

The key word is autonomous. You don't have to ask it for everything. You give it a goal—"manage my inbox," "research this company," "book me a flight"—and it figures out the steps on its own.

Steinberger even connected his AI to the door-lock system.

Theoretically, the AI could lock him out of his home.

But it's precisely this risky design that makes OpenClaw a real AI agent, rather than just another chatbot.

This is the shift Huang was pointing at on stage today.

"The last prompt," he said, "was 'what is', 'when is', 'who is.' This now prompt goes 'create', 'do', 'build', 'write.'"

AI went from answering questions to taking actions. That's the whole ballgame.

Why This Breaks the Math

Here's the investment angle—and it's enormous.

A normal AI conversation uses a modest amount of computing power. A quick answer, a quick response.

A typical generative AI prompt produces a single response, while agentic tasks can consume around 1,000 times more tokens.

Continuous OpenClaw agents running in the background may consume up to one million times more tokens.

One million times.

"We have a whole bunch of OpenClaw in the company," Huang said. "They're all continuously running, doing things for us, writing, developing tools, developing software. The amount of compute in our company that we need has just skyrocketed. The amount of compute every company needs is skyrocketing."

This is why NVIDIA is celebrating a software project it didn't build. Every OpenClaw agent running in the background is another reason to buy more chips. More agents = more compute = more Blackwell GPUs = more revenue.

The hardware thesis just got a second wind.

BUT… There’s a BIG Bottleneck

Here's the open secret Huang didn’t linger on.

Agents are getting smarter. Faster. Cheaper. The hardware is catching up. The software stack is locking in.

But there's one problem that doesn't get solved by a faster GPU or a cleaner policy engine.

Memory.

Right now, every time an AI agent starts a new task, it forgets everything it's ever done.

It's like hiring the world's most capable employee—and then wiping their brain every morning before they come into the office. No context. No history. No accumulated knowledge about you, your company, your preferences, your patterns.

That's not a super-employee that’s going to revolutionize your workflow.

That's a goldfish.

For agents to truly work—to help manage your inbox, run your experiments, coordinate your team, operate your business—they need to remember.

Persistently. Across sessions. Across tasks. Across time.

Right now, the entire agentic AI industry is built on a foundation that doesn't (yet) have this solved.

Whoever cracks it first doesn't just win a product category. They become the connective tissue of every agentic deployment on earth.

The claw is the law. But memory is the throne.

That's the conversation James and I are having tomorrow.

Beyond that, we've got an incredible lineup covering all things tech.

If you're serious about understanding where this is all going—and where the real money gets made—you need to be in the room.

→Grab your spot at Tech Turning Point right here.