- Authors
- Name
- 오늘의 바이브
Vertical on a Semi-Log Scale
On March 3, 2026, OpenClaw passed 250,829 GitHub stars. It overtook React to become the most-starred software project in GitHub history. 48,274 forks. 1,075 contributors. Time from launch to record: roughly 60 days.
NVIDIA CEO Jensen Huang looked at the adoption curve and said this: "If you look at the line even in semi-log, this thing is straight up. It's vertical. It looks like the Y-axis." A semi-log graph exists to flatten exponential growth into a straight line. If the line is still vertical on a semi-log scale, "exponential" is an understatement.
React took 8 years to reach this point. Linux took 12. Kubernetes took 10. OpenClaw did it in two months. Nothing in open source history has moved this fast.
Why Huang Called It "The Most Important Software Ever"
Jensen Huang's exact words: "OpenClaw is probably the single most important release of software, you know, probably ever." That is not hyperbole from a tech executive. It is arithmetic from a GPU company CEO.
NVIDIA sells GPUs. GPU demand scales with AI compute. AI agents like OpenClaw consume dramatically more compute than traditional generative AI.
Huang provided the numbers himself. A standard generative AI prompt produces a single response. An agentic task consumes approximately 1,000 times more tokens. A background agent running continuously can consume up to 1 million times more tokens. "The amount of compute every company needs is skyrocketing," he said.
The CEO of a GPU company calling the software that multiplies GPU demand by 1,000x "the most important ever" is not a technical assessment. It is a business statement. OpenClaw's success means more agents. More agents mean more tokens. More tokens mean more GPUs sold.
From Queries to Actions
To understand why OpenClaw grew this fast, look at the shift it represents. Traditional AI is query-based. You ask "what is" or "when is" and get a text response. You read the answer in a chat window. The AI stays inside that window.
OpenClaw and the AI agent paradigm are action-based. You say "create," "do," "build," or "write," and the agent actually operates the computer. It creates files, sends emails, executes shell commands, calls APIs.
The difference matters. Query-based AI assists human productivity. It finds information faster, drafts text, suggests code snippets. But the human still has to execute. Action-based AI executes. You say "do it," and the agent runs until it is done.
As Huang described it, agents can "autonomously research information, read manuals, apply tools, and execute complex tasks." This is not an evolution of chatbots. It is a fundamental change in how software gets used.
Where OpenClaw Came From
OpenClaw's history is short but chaotic. Austrian developer Peter Steinberger launched it in November 2025 under the name Clawdbot. It is an open-source platform that gives large language models -- ChatGPT, Claude, and others -- direct access to a computer.
It supports WhatsApp, Telegram, Signal, Discord, Slack, and iMessage. It can access local file systems, shell commands, email, calendars, and web browsers. A plugin system called "skills" makes it infinitely extensible.
On January 27, 2026, a trademark dispute with Anthropic forced a rename to Moltbot. Three days later, citing pronunciation difficulties, it became OpenClaw. The chaotic rebranding went viral.
Then the numbers exploded. 34,168 stars in 48 hours. At peak, 710 stars per hour. React needed 8 years, Linux needed 12, Kubernetes needed 10 to hit 100,000 stars. OpenClaw passed that mark in days.
Growth Comparison: React, Linux, Kubernetes vs OpenClaw
When you line up the numbers, the scale of this growth becomes visceral.
| Project | Time to 250k GitHub Stars | Launch Year | Core Purpose |
|---|---|---|---|
| React | ~8 years | 2013 | Frontend UI library |
| Linux | ~12 years | 2011 (on GitHub) | OS kernel |
| Kubernetes | ~10 years | 2014 | Container orchestration |
| OpenClaw | ~60 days | Nov 2025 | AI agent platform |
React is the standard for web development. Meta built it. Millions of developers use it in production. OpenClaw surpassed its star count in two months.
The eras are different, of course. GitHub had far fewer users in 2013. The star button meant something different then. But as Huang pointed out, OpenClaw's growth curve remains vertical even after adjustment. When a line is vertical on a semi-log scale, the era difference cannot explain it away.
1,000x the Tokens, 1,000x the Opportunity for NVIDIA
Look at Huang's token consumption data again. Agentic tasks consume approximately 1,000x the tokens of a standard generative AI prompt. Continuous background agents can reach up to 1 million times more.
The math for NVIDIA is straightforward. Processing tokens requires GPUs. If token consumption rises by 1,000x, GPU demand rises proportionally. Every major AI model today -- GPT, Claude, Gemini, DeepSeek -- runs on NVIDIA hardware.
NVIDIA is already using OpenClaw internally. In Huang's words, multiple OpenClaw agents are "all continuously running, doing things for us." The company uses them for internal development and software creation.
This is not a casual endorsement. NVIDIA's business model and OpenClaw's growth are structurally coupled. OpenClaw succeeds, agent usage grows, token consumption explodes, GPUs sell. When Huang called OpenClaw "the most important software ever," he was not making a technical judgment. He was stating a business conviction.
The Shadow Side of Fast Growth
OpenClaw's growth is not all upside. As of March 3, security researcher Maor Dayan scanned the internet with a custom tool called ClawdHunter v3.0 and found over 42,665 OpenClaw instances exposed on the public internet. Among verified instances, 93.4% had authentication bypass vulnerabilities.
SecurityScorecard independently confirmed 40,214 exposed instances, of which 12,812 were vulnerable to remote code execution (RCE). Another 341 malicious skills had infiltrated the marketplace.
Growth outran security. Users deployed OpenClaw faster than security teams could monitor it. 250,000 stars prove technical enthusiasm. 40,000 exposed instances prove the infrastructure is not ready.
This is not just an OpenClaw problem. It is structural to the AI agent category. Agents have permissions to access email, documents, messages, and API keys. Connect one to Slack, and it reads your messages. Connect it to Google Workspace, and it accesses your documents. With an OAuth token, lateral movement becomes trivial. An agent exposed to the internet without authentication is not an open door. It is an open vault.
It Did Not Beat React. It Changed What Software Means.
Let's be honest: OpenClaw passing React in GitHub stars is not inherently meaningful. Stars are not a quality metric. They are a blend of curiosity, hype, and FOMO. React underpins millions of production applications. OpenClaw is not there yet.
But Huang is not looking at the star count. He is looking at the slope of the adoption curve. What that slope indicates is that the way people use AI is fundamentally shifting. From asking to acting. From assistant to autonomous agent. From single responses to continuous operation.
Translate this shift into tokens and you get 1,000x to 1,000,000x. That means redrawing the demand curve for the entire computing industry. Whether OpenClaw is the catalyst for this transition or merely a symptom of it remains unclear. But one line on a graph is vertical even on a semi-log scale, and that line maps directly to GPU demand. That much is certain.
Sources