Anthropic released the Model Hardware Standard (MHS) on Thursday as a research preview, marking the company's first move into so-called physical AI. The framework connects large language models like Claude directly to physical equipment, from manufacturing robotics to laboratory microscopes.
The numbers come first. Anthropic says integrating AI into equipment through MHS takes 'hours or minutes,' compared with the 'weeks, if not months' typically required for a custom, specialist-built connection. That is a productivity gain measured directly in engineering time and capital saved.
MHS is built on the Model Context Protocol (MCP), an open standard Anthropic introduced in 2024 to link AI systems with data sources. Alek Kemeny, a member of Anthropic's technical staff, described MCP to Fortune as 'kind of like the USB for AI to software connection.' Crucially, MHS is model-agnostic: it works with any LLM, not just Claude, including systems built by rival OpenAI or open-source alternatives.
The standard also lets multiple devices communicate with one another through shared commands such as 'read,' letting robotic arms, assembly lines and lab instruments act on the same instructions regardless of manufacturer.
Jonah Cool, Anthropic's head of partnerships and deployment of science, framed the pitch around a familiar market complaint: scientific equipment 'suffers from proprietary solutions that are very brittle and often don't meet the need of scientists.' 'We want to avoid vendor lock-in for scientists,' Cool said.
Not every existing machine can plug into MHS out of the box, since many lack a programming interface. Anthropic says it is working with device manufacturers to build new products pre-loaded with MHS and to retrofit existing equipment. Kemeny said the goal is a future where 'scientists can buy these devices and out of the box it works.'
Anthropic developed the standard with the HHMI Janelia Research Campus, a biomedical research center in Virginia, and gave early access to a handful of partners spanning biotech, robotics and quantum computing: Genentech, Carnegie Mellon University, QuEra, Universal Robots, Amazon Web Services, Doosan Robotics, Danaher and Hugging Face.
The move lands amid a broader corporate scramble into physical AI. Hugging Face separately unveiled a robotic duck the same day, though it does not run on MHS. Nvidia, which is set to purchase Hugging Face for $13 billion in a deal that has not yet closed, has long argued that 'every industrial company will become a robotics company,' per CEO Jensen Huang's March prediction.
Capital rewards clear rules, and Anthropic's bet is that an open, model-agnostic standard beats the walled gardens that have long defined industrial software. If manufacturers and labs no longer need custom, months-long integrations to deploy AI on the factory floor, the savings accrue directly to producers rather than to specialist integrators charging premium fees for proprietary systems.
The pending Nvidia-Hugging Face transaction, still unclosed at $13 billion, underscores how much private capital is racing toward the same physical AI thesis from a different angle -- chips and hardware rather than open protocols. Whichever approach wins, the market, not Washington, will decide it. Anthropic's move is a reminder that in a still-lightly-regulated frontier, competition among private firms -- not a government standard-setter -- is what is driving the pace of innovation.
