OpenClaw's ChatGPT moment sparks concern that AI models are becoming commodities - CNBC
OpenClaw Is Real Infrastructure. Here Is What the Numbers Say.
Three months ago, Peter Steinberger was an Austrian developer with a lobster-themed side project. Today, OpenClaw has crossed 250,000 GitHub stars, an independent foundation, Nvidia building enterprise security tooling around it, and Chinese regulators restricting it at state enterprises. Whether that trajectory represents a genuine inflection point for agentic AI or a media narrative amplified by Jensen Huang's keynote is the question worth asking. The numbers, read carefully, lean toward inflection.
OpenClaw launched in November 2025 as Clawdbot, a personal AI assistant that runs locally and uses messaging platforms — WhatsApp, Telegram, Signal, Discord, Slack, and roughly 16 others — as its user interface. Steinberger renamed it Moltbot on January 27, 2026, after Anthropic raised trademark concerns, then OpenClaw three days later on January 30, 2026, when Moltbot "never quite rolled off the tongue." [CNET, DEV community] As of March 2, 2026, OpenClaw had accumulated 247,000 GitHub stars and roughly 47,700 forks; by March 3 it had crossed 250,829 stars and 48,274 forks, per the project's own blog. [OpenClaw blog] That makes it, per Nvidia's CEO, "the fastest-growing open source project in history." That framing is self-serving — Nvidia has obvious reasons to amplify a platform that runs its GPUs — but the underlying numbers are not fabricated.
The more telling metric is the fork-to-star ratio. As of early March, OpenClaw had accumulated roughly 47,700 forks alongside those 247,000 stars, a ratio of approximately 19%. For typical open source projects, that number sits closer to 5-10%. The gap suggests developers are not passively bookmarking OpenClaw — they are cloning it, modifying it, and shipping derivatives. NanoClaw, a security-hardened fork built by Israeli developer Gavriel Cohen, emerged precisely because the base version's permission model was too loose for enterprise use. Cohen's wife used it to track baby stroller prices on WhatsApp; Cohen and his brother spun it into a startup called NanoCo and partnered with Docker. That downstream ecosystem activity — not the headline stars — is what infrastructure actually looks like when it moves.
The security picture is where the narrative frays. Cisco's AI security research team tested a third-party OpenClaw skill and found it performed data exfiltration and prompt injection without user awareness. OpenClaw's own Discord maintainers have warned that the project is "far too dangerous" for users who cannot read command-line output. In March, Chinese authorities restricted state enterprises and government agencies from running OpenClaw on office computers. These are not minor footnotes — they are the actual constraints on adoption, and they explain why Nvidia built NemoClaw.
NemoClaw, announced at GTC on March 16, installs in a single command and layers policy-based security, network guardrails, and an isolated sandbox beneath OpenClaw. It uses Nvidia's Agent Toolkit to optimize the framework and can run open models locally on RTX PCs, DGX Station, or DGX Spark. Huang's framing — "Mac and Windows are the operating systems for the personal computer. OpenClaw is the operating system for personal AI" — is the kind of line that sounds profound until you notice it was delivered at a conference where Nvidia sells the hardware that runs it. But the NemoClaw announcement itself is the telling part. Nvidia is not building a competing platform; it is building a security wrapper around an existing one. That is a bet that OpenClaw is the substrate, not the product.
Steinberger joined OpenAI on February 14, 2026, but the project transferred to an independent 501(c)(3) foundation before he did, per the OpenClaw blog. [OpenClaw blog, Fortune] The governance structure matters: OpenAI does not own OpenClaw. This separates it from the typical pattern where a solo developer builds something interesting and a large lab acquires it. It also means the foundation — not Steinberger, not OpenAI — controls the roadmap going forward. That is an unusual arrangement and one with real consequences for how the project evolves.
The analyst reception is divided in a useful way. Jerry Chen at Greylock, an Anthropic investor, frames OpenClaw as a tangible demonstration of what intelligent agents can do but does not see it displacing foundation models. "The interesting question now is whether OpenClaw becomes the de facto standard — the Linux of the market, as Jensen puts it — or just the first of many open and closed-source agentic operating systems," he told CNBC. David Bader at NJIT offered a cleaner framing: "The models become the engine; the agent framework becomes the car." That division of labor — commoditized models below, differentiated orchestration above — is what the open-source community has been building toward, and OpenClaw is the clearest instance of it working at scale.
For builders and investors, the relevant read is narrower than the headlines suggest. OpenClaw is not threatening to displace GPT or Claude. It is demonstrating that the value layer in AI is shifting down the stack, from model providers to orchestration frameworks. The GitHub stars are real. The fork activity is real. The security problems are also real, and they are the binding constraint on enterprise adoption — which is exactly why Nvidia built NemoClaw and why NanoClaw exists. The framework that solves the security and permission model problem cleanly will own the next phase of this market. That may be OpenClaw itself, a hardened fork like NanoClaw, or something that has not shipped yet. What is clear is that the race has started, and it is being run on open infrastructure.
Sources: