Physical AI Has 3,110 Open Roles. Here's What the $200K Median Actually Signals.
A September 2026 snapshot of Physical AI hiring counts 3,110 open roles across 67 companies, with a reported $200K median salary. The useful signal is not a single headline number—it is the spread of engineering work across robotics, autonomy, simulation, perception, controls, and deployment.
# Physical AI Has 3,110 Open Roles. Here's What the $200K Median Actually Signals.
Published: September 2, 2026
The AI hiring market is not one market. It is a collection of technical labor markets that happen to share a label.
The latest example is Physical AI. A September 1 snapshot from Physical AI Jobs counts 3,110 open roles across 67 companies and 228 cities, with a reported $200,000 median salary where compensation data is available. The tracked market includes robotics, autonomy, simulation, perception, controls, SLAM, motion planning, and other engineering work that connects models to machines and physical environments. (Physical AI Jobs — State of Physical AI Jobs 2026)
That number is useful, but it needs to be read correctly. It is a point-in-time count of public listings from one specialist job source, not a census of the global AI workforce and not a guarantee that every role pays $200,000. The more durable signal is the shape of the demand: employers are hiring for people who can make intelligent systems operate reliably outside a notebook or a benchmark.
Physical AI is a deployment market
The report describes an engineering-led market. Its role families include robotics software, autonomy, perception, controls, simulation, SLAM, and motion planning. The work is practical: debugging robots, scaling autonomy stacks, building simulation and data tooling, validating behavior, and integrating software with hardware constraints.
That makes Physical AI different from the way many people imagine frontier AI hiring. A research title may focus on a new model or learning method. A Physical AI role may instead ask whether a robot can localize in a changing environment, whether a vehicle can make a safe decision under sensor noise, or whether a simulation pipeline produces useful data for the next training run.
The distinction matters for candidates because the portfolio signal changes. A polished chatbot demo says little about whether someone can operate a system with sensors, latency, failure modes, and physical consequences. A smaller project with a reproducible evaluation harness, a simulation environment, telemetry, and an explanation of failure handling is much more legible to a robotics or autonomy hiring team.
The Vibe Coding Ebook covers the agentic development tools that can accelerate parts of this work. The important word is *parts*: an AI coding agent can help generate a test scaffold or inspect a codebase, but the engineer still owns the system boundary, the evaluation design, and the decision about what evidence is sufficient.
The $200K median is a market marker, not a promise
Physical AI Jobs reports a $200,000 median salary for its tracked market, but also notes that salary coverage is incomplete because many employers do not publish pay. Candidates should therefore treat the figure as a market marker and inspect the underlying role, location, level, and compensation definition before comparing offers.
The broader AI jobs market shows why scope matters. Prof H Lab's September 1 tracker counted 3,733 open roles across 20 AI, robotics, and autonomy boards, including 2,955 U.S. roles and 2,210 postings with a printed pay band. Its company breakdown listed 769 OpenAI roles, 571 Anthropic roles, and 350 Waymo roles on that day's reading. Those numbers come from employer career feeds and use a different coverage set from the Physical AI report, so they should not be added together. (Prof H Lab AI Jobs Tracker)
The comparison is still revealing. Specialized trackers can surface compensation and role detail that disappear inside a broad “AI engineer” count. For a candidate, the right question is not “What is the average AI salary?” It is “Which technical market am I actually entering, and how does that market price the work I can do?”
Remote work exists, but hardware changes the map
The Physical AI snapshot reports 940 onsite roles, 267 hybrid roles, and 77 remote roles. Another 1,826 listings have unspecified work mode. In the report's classification, remote roles represent only 2% of the tracked inventory.
That is not a statement that remote Physical AI is impossible. It is a reminder that hardware creates a different operating constraint from ordinary web software. Someone may be able to write perception code from anywhere, but testing the result may require a lab, a vehicle, a robot, a sensor rig, or access to a physical test range. A role that looks like “machine learning engineering” in the title may still be deeply coupled to equipment and field operations.
Candidates should ask where the work is actually performed. Is the team training models, building simulation, operating a fleet, or integrating software at a customer site? How much of the role is remote coding, and how much is lab or field validation? What happens when the simulation passes but the hardware fails? These questions are more informative than a generic remote label.
The software-engineering slice is smaller and more specific
Looking for AI-native engineers?
Post your role for free on LLMHire and reach thousands of verified engineers actively exploring opportunities.
CronJobs' September 1 snapshot offers a useful adjacent view: it observed 117 eligible U.S. AI/ML software-engineering jobs, or 7.1% of a 1,655-job inventory. AI/ML was the fourth-largest normalized role category in that dataset, behind backend, DevOps/SRE, and security. CronJobs explicitly warns that this is a descriptive inventory and cannot establish whether AI is creating or eliminating software-engineering jobs. (CronJobs AI/ML software-engineering snapshot)
That caution is the right one for Physical AI as well. Listing counts show where employers are advertising work; they do not show offer acceptance, hiring velocity, attrition, or the number of roles that will remain open after a hiring plan changes. They are a map, not a causal model.
Still, the category mix gives job seekers a practical search strategy. Instead of applying only to “AI engineer,” search across the adjacent systems that make AI deployable:
- Simulation and data tooling: build the environments and datasets used to train and test behavior.
- Perception and localization: turn sensor data into a stable representation of the world.
- Controls and motion planning: convert decisions into actions under constraints.
- Robotics software and infrastructure: connect model outputs to production systems, observability, and fleet operations.
- Evaluation and safety: define what “works” means and measure behavior before deployment.
This is the same career principle LLMHire has covered in its AI evaluation engineering and forward-deployed AI engineering coverage: the scarce skill is increasingly the layer between a capable model and a dependable system. AgenticNode explores the orchestration side of that layer, where tools, workflows, permissions, and feedback loops have to work together.
What to put in a Physical AI portfolio
The best portfolio project does not need to imitate a billion-dollar robotics company. It needs to make the engineering judgment visible.
Show the input data and the assumptions. Define the environment and the success criteria. Include a baseline, not just the final demo. Record failure cases and explain whether you changed the model, the data, the controller, or the evaluation. If an AI coding tool helped write the system, document how you reviewed generated code and how you tested safety-critical boundaries.
For software-heavy candidates, a simulation-first project can be a credible bridge. Build a small navigation or manipulation environment, expose the state and actions clearly, add telemetry, and make the evaluation repeatable. For candidates coming from data or ML, add the production layer: versioned datasets, reproducible runs, monitoring, and a clear rollback or fail-safe story.
The point is not to claim that a toy robot proves production readiness. It is to demonstrate the habits that production teams need: measurement, debugging, systems thinking, and honest limits.
The hiring signal
Physical AI's 3,110 listed roles do not prove that every AI engineer should move into robotics. They do show that AI demand is spreading into specialized engineering markets where deployment constraints matter. The reported $200,000 median is a useful indicator of scarcity, while the role and work-mode breakdown explains why the market is not interchangeable with general-purpose software hiring.
For candidates, follow the work rather than the label. Learn the model layer, but also learn the environment around it: simulation, sensors, systems integration, evaluation, observability, and safety. Those are the capabilities that remain valuable when a model changes—and they are the capabilities hiring teams can verify after the demo ends.
Related: AI Evaluation Engineer: The Role That Keeps AI Products From Failing in Production · Forward-Deployed AI Engineer: Why Amazon, OpenAI, and Anthropic Are Racing to Staff the Role
For the developer workflow behind AI-enabled systems, see the Vibe Coding Ebook and AgenticNode.