What did Jensen Huang say about OpenClaw?

At the Morgan Stanley TMT Conference on March 5, 2026, NVIDIA CEO Jensen Huang called OpenClaw 'probably the single most important release of software, probably ever.' He compared it to Linux, saying OpenClaw surpassed Linux's adoption in three weeks versus Linux's thirty years. He also said OpenClaw agents consume up to 1,000,000 times more tokens than a standard prompt, driving massive demand for GPU compute.

Did OpenClaw really surpass Linux?

In GitHub stars and downloads, yes. OpenClaw crossed 250,000 GitHub stars in roughly 60 days - a milestone that took React over a decade. But GitHub stars measure hype, not production usage. Linux runs the vast majority of the world's servers, phones (via Android), and supercomputers. OpenClaw is a four-month-old project that most users are still experimenting with. The comparison measures virality, not impact.

Why is Jensen Huang promoting OpenClaw?

NVIDIA sells the GPUs that power AI inference. According to Huang's own math, an agentic task consumes roughly 1,000x more tokens than a standard prompt, and continuously running OpenClaw agents consume up to 1,000,000x more. More tokens means more compute demand, which means more GPU sales. When Huang calls OpenClaw the most important software ever, he is also describing the most important demand driver for NVIDIA's core business.

Should founders use OpenClaw based on Jensen Huang's endorsement?

The endorsement validates the AI agent category, not OpenClaw specifically. Huang is saying that autonomous AI agents running continuously are the future of compute. That thesis can be true whether you use OpenClaw, Claude Cowork, a managed AI agent, or build your own. OpenClaw itself still has significant security concerns, unpredictable costs, and requires meaningful technical investment to run safely.

Jensen Huang's OpenClaw Hype Is Really About Selling GPUs

Q: What happened at ClawCon NYC?

ClawCon NYC was held on March 5, 2026, and drew over 1,300 attendees. It featured lobster-themed decor, live demos, and networking. However, when one presenter asked the crowd whose lives had been changed by OpenClaw, almost no hands went up, according to Business Insider's reporting. The event had more energy around the community and the idea of personal AI than around proven daily utility.

On March 5, Jensen Huang told a room of Morgan Stanley investors that OpenClaw is "probably the single most important release of software, probably ever." He compared it to Linux. He said it surpassed Linux's adoption in three weeks. He called the adoption curve "vertical" and said it "looks like the Y-axis."

That same evening, at ClawCon NYC - OpenClaw's own fan event, packed with 1,300 true believers in lobster headbands - a presenter asked the crowd whose lives had been changed by OpenClaw. Almost no hands went up.

Both of these things are true at the same time. And the gap between them is the most important thing for founders to understand about where AI agents actually are right now.

What Huang actually said

The setting matters. This wasn't a product launch or a tech conference. It was the Morgan Stanley Technology, Media & Telecom Conference - an investor event. Huang was answering questions about enterprise AI demand.

His key claims: OpenClaw surpassed Linux as the most downloaded open-source software in history, doing in three weeks what Linux took thirty years to achieve. The adoption curve is "straight up" even on a semi-logarithmic scale. And the compute implications are staggering - a standard prompt produces one response, an agentic task consumes roughly 1,000x more tokens, and continuously running OpenClaw agents consume up to 1,000,000x more.

That last number is the one that matters most to NVIDIA. More tokens means more inference compute. More inference compute means more GPU sales. When Huang describes OpenClaw as the most important software ever, he's simultaneously describing the most important demand driver for his company's core business.

This doesn't make him wrong. It makes him interested.

The Linux comparison doesn't mean what you think

Huang's comparison to Linux is technically accurate and deeply misleading at the same time.

OpenClaw crossed 250,000 GitHub stars faster than any project in history - surpassing React, which took over a decade to reach the same number. By the metric of "how fast did developers click a star button," OpenClaw is unprecedented.

But GitHub stars measure curiosity, not production usage. Linux runs the majority of the world's servers. It powers Android, which runs on billions of phones. It runs most of the world's supercomputers. It is foundational infrastructure that the entire internet depends on.

OpenClaw is a four-month-old project that most users are still experimenting with. Its own maintainer has warned that "if you can't understand how to run a command line, this is far too dangerous of a project for you to use safely." A zero-click exploit was disclosed on March 2 that lets attackers hijack an instance just by getting the user to visit a webpage.

Comparing OpenClaw's GitHub velocity to Linux's three-decade adoption arc is like comparing a viral TikTok to a highway system. One is fast. The other is infrastructure.

The ClawCon reality check

The same day Huang made his declaration, ClawCon NYC told a different story.

The event drew 1,300+ attendees to a West Village venue. There were lobster headbands, lobster tails, silver balloons, and a DJ. The host, Michael Galpert, kicked things off by telling the crowd: "All your friends and family probably think you're crazy, and the whole point is for you to be in a room with other crazy people so it's normal."

Business Insider's reporter was in the audience. His observation was pointed: attendees weren't that enamored by OpenClaw itself. When a presenter asked whose lives had been changed by it, almost no hands went up. People seemed more excited by the branding and the community than by proven daily utility.

The demos reinforced this. Some were genuinely interesting - a researcher used OpenClaw to manage his colony of mice. Another used it for paper trading. But several had technical difficulties and devolved into screenshots. Many felt like product pitches for OpenClaw wrapper services. By the end of the evening, seats were emptying and people had stopped listening, choosing to network instead.

An Amazon engineer told an Uber engineer he was "doubtful of OpenClaw" but came because he was "open-minded and loved AI." That sentence captures the state of the entire ecosystem: curious, hopeful, not yet convinced.

Meanwhile, a Hacker News thread asked the same week why people claim OpenClaw has "completely changed their life" when every use case could be replicated with Zapier, Automator, or a basic script.

Why Huang is right about the category (but wrong about the timing)

Here's the part founders should actually pay attention to: Huang's thesis isn't really about OpenClaw. It's about the shift from AI as a query tool to AI as an action tool.

"The last prompt was 'what is,' 'when is,' 'who is,'" he said. "This now prompt goes 'create,' 'do,' 'build,' 'write.' The last prompt was queries. This prompt is actions. Tasks. Do something for me."

That framing is correct. AI agents that can take actions on your behalf - managing calendars, updating CRMs, handling meeting follow-ups, writing documents - represent a genuine shift in how work gets done. The reason Claude Cowork caused a $30 billion software stock selloff wasn't because the demos were flashy. It was because enterprise buyers recognized that agents doing actual work changes the economics of entire software categories.

But Huang's framing collapses the distance between "this direction is correct" and "this specific tool is ready." OpenClaw has 250,000 GitHub stars and real security problems. It has a passionate community and unpredictable costs. It has a Shenzhen government subsidy program and a maintainer who warns beginners away. It is a proof of concept for the agentic future, not the finished product.

Follow the compute, not the hype

Huang's clearest tell is the token math. He said NVIDIA is running "a whole bunch of OpenClaw in the company" and that they're "all continuously running, doing things for us, writing, developing tools, developing software." The amount of compute NVIDIA needs has "skyrocketed."

This is honest - and it's exactly why founders should be skeptical of taking his OpenClaw endorsement at face value. NVIDIA's business model depends on compute demand growing faster than anyone expects. An AI agent that runs 24/7 and consumes a million times more tokens than a chat prompt is, from NVIDIA's perspective, the perfect product. It doesn't matter whether that agent is OpenClaw, Claude Cowork, a custom enterprise deployment, or something that doesn't exist yet.

The endorsement is for the compute pattern, not the specific software.

For founders at Series A-B companies evaluating what to do with all of this, the takeaway isn't "go deploy OpenClaw because Jensen Huang said so." It's:

The agentic direction is real. AI that takes actions rather than answering questions is where the industry is heading. If you're not thinking about how agents fit into your workflows, start.

The specific tools are still immature. OpenClaw has massive adoption and massive security gaps. Claude Cowork is powerful but locked to Anthropic's ecosystem. Every option involves meaningful tradeoffs today.

Self-hosting an AI agent is a commitment, not a weekend project. Whether you choose OpenClaw, NanoClaw, or something else, the setup is just the beginning. Running an agent that has access to your company's data and systems requires ongoing maintenance, security monitoring, and cost management.

Managed alternatives exist for a reason. Not every team needs to self-host. If you want AI handling your admin work - CRM updates, meeting follow-ups, action item tracking - without becoming an infrastructure project, managed AI platforms that own their own security and reliability are the more practical path for most startups.

The real signal underneath the noise

Jensen Huang is one of the most consequential technology executives alive. When he calls something the most important software release ever, it moves markets. Literally - his comments at Morgan Stanley contributed to a broader narrative about AI compute demand that affects how billions of dollars get allocated.

But his job is to sell the future of compute. OpenClaw's job is to be a useful product. Right now, those two things are in tension. The future Huang describes - millions of agents running continuously, consuming orders of magnitude more compute - may well arrive. But at OpenClaw's own celebration, in a room full of its most enthusiastic users, almost nobody could say it had changed their life yet.

The category is real. The specific moment is early. And the person telling you to go all in is the one selling the infrastructure.

This is part of a series on AI agents in 2026. See also: Is OpenClaw Safe?, How Much Does OpenClaw Actually Cost?, NanoClaw vs OpenClaw, Claude Cowork vs OpenClaw, and Best OpenClaw Alternatives That Don't Require Coding.