NVIDIA announced this week entry into the OpenClaw craze with their introduction of NemoClaw, marketed as “secure, always on AI assistants.”
It acts as a wrapper around OpenClaw by putting it inside a Docker container and granting it no permissions by default on anything outside of the container. This gives OpenClaw an isolated sandbox while maintaining similar behavior you would get by putting it on its own server, allowing you to have multiple claws running on a computer without interacting with each other.
It opens up some new possibilities for deploying OpenClaw for companies- you can host them on the same server, which saves costs and you can more effectively control agent permissions- you control them at the the OpenShell level instead of the OpenClaw level.
What started as a series of viral memes are now being taken very seriously by the top AI companies. Meta acquired Moltbook recently and Anthropic released a new feature allowing you to communicate with Claude Code directly from your phone and be triggered by external events- some of the core features of OpenClaw.
Of course these always-on use cases bring about the obvious problem of token costs. If you’re not careful and are using the most expensive models you could rack up $100 of cost in tokens in one day easily, which means using OpenAI / Anthropic’s subscription (which actually breaks Anthropic’s ToS) is the only viable option to use the best models.
With the increased inference demand there have been growing questions on how long we will have access to models at current prices, OpenAI is expected to lose $14 billion this year. They are planning on going public in Q4 so will require the support of the public market to keep subsidizing their tokens for us, along with the other labs which we can assume have similar margins.
These factors explain the NVIDIA $1T investment in Vera and Blackwell chips. CEO Jensen Huang has said they cannot currently keep up with demand, revealing the bottleneck for this industry. I recently added GPU pricing to the site as I plan on adding more sophisticated analysis to make the picture more clear on how viable the current costs we get are- on subscriptions and APIs.