Why OpenAI and Anthropic Are Bet Everything on Robots Now

The next battlefield is not your screen

For two years, the AI story was about language models getting smarter at text and images. OpenAI and Anthropic competed to make chatbots more helpful, more fluent, more creative. That competition has now hit a ceiling. Chatbots reduce friction inside software. But the world has more friction outside of it.

Both companies are pivoting hard into what the industry calls physical AI — systems that understand and act in the real world. OpenAI is hiring robot engineers and building hardware. Anthropic is building the protocol that will let those robots talk to machines. Neither approach is obvious. Both are calculated bets that the next trillion-dollar opportunity lives in factories, warehouses, and homes, not in your inbox.

OpenAI is betting on everything from gears to data

Sam Altman finally confirmed what the market already suspected. In a podcast appearance this month, he said OpenAI will build humanoid robots and other robotic forms. It is the first time he has acknowledged the effort publicly, and the company is moving fast enough to need no announcement.

OpenAI posted eleven robotics job listings in May. It now has twenty-five. The growth is not incremental. It is structural. The company is hiring for hardware roles — actuator designers, precision gear engineers, thermal management specialists — positions that have nothing to do with LLMs. Aditya Ramesh, who led DALL-E and worked on the Sora video model, is running the robotics team. The message is clear: OpenAI wants full-stack control over both the brain and the body.

The salary numbers tell the same story. OpenAI is offering base pay of $157,000 to $500,000 a year for robotics roles, with equity and performance bonuses on top. The highest-paying position is a machine learning engineer who builds distributed data systems — the plumbing that turns sensor readings and video feeds into training material. That is the real bottleneck. Robots cannot learn the way language models do. There is no equivalent of “the internet” for physical behavior. OpenAI is hiring aggressively to solve that problem.

Anthropic is playing a different game

Anthropic is not building a robot. It is building the language robots will use.

Last month the company released MHS, the Model Hardware Standard. It is an abstraction layer that lets AI agents discover, connect to, and operate physical equipment — robotic arms, microscopes, lasers — regardless of manufacturer. The promise is dramatic. Installing and integrating hardware in a lab or factory used to take weeks or months. MHS reduces that to hours or minutes.

The standard is model-agnostic. Claude is not required. Any AI system that implements the spec can drive compatible hardware. That is a deliberate choice. Anthropic is positioning MHS as infrastructure, not a product wrapper. If the standard catches on, Anthropic gains influence over an entire category without owning a single factory floor.

The early adopters are telling. Doosan Robotics, a Korean manufacturer, is testing MHS to automate quality inspection and coordinate tasks across multiple arms. AWS plans to support the standard through its Strands Robots tooling. Universal Robots, the Danish arm maker, is integrating it. Google DeepMind is pursuing a parallel path with Gemini Robotics 2, a model designed to run on heterogeneous hardware — Apollo 2 humanoids from Apptronik, Franca robotic arms, Spot quadrupeds from Boston Dynamics — without being tied to any one platform.

The market is real and it is huge

Grand View Research projects the physical AI market to grow from $110.8 billion this year to $960.4 billion by 2033. JP Morgan analyst John Rhie called it one of the largest investment opportunities of the next decade. The numbers are impressive, but the direction matters more. AI is moving from cognitive tasks to physical ones. That shift changes everything about who captures value.

Who wins, who loses

The winners are the companies that control either the hardware layer or the integration layer — or both. OpenAI is going for full-stack dominance. Anthropic is going for standard-setting power. Google is doing both through DeepMind. Startups that build the sensors, actuators, and simulation environments these systems depend on will also profit.

The losers are the incumbents who built their moat on proprietary interfaces and closed ecosystems. If MHS or a similar standard becomes the default way AI agents connect to physical equipment, hardware manufacturers that resist will find themselves locked out of the AI-driven upgrade cycle. Factory owners who invest in equipment that cannot speak the new language will face expensive retrofits or obsolescence.

Korean manufacturers are in a privileged position. Doosan Robotics is already on the MHS trial track. The country’s strength in precision hardware — gears, actuators, collaborative arms — aligns directly with what physical AI demands. But that advantage is not automatic. It depends on how quickly Korean firms can embed these standards into their product roadmaps.

What happens next

The talent war is only beginning. OpenAI’s twenty-five postings are a signal, not a summit. expect hundreds more in the coming quarters as the companies that move fastest lock in the engineers who can bridge silicon and steel. Salaries will keep rising for people who can design actuators and train vision-language models in the same breath.

Regulation will lag. Physical AI introduces safety questions that text-based AI never raised. A chatbot that gives bad advice causes reputational damage. A robot that miscalculates a grip force causes physical harm. Expect jurisdiction-specific safety standards within two years, and global fragmentation in how they are defined.

The most consequential shift is strategic. AI is no longer just about information. It is about action. The companies that figure out how to make intelligent systems operate reliably in unstructured physical environments will define the next era of computing. OpenAI and Anthropic are both racing toward that future. They just arrived by different roads.