Why Car Companies Will Win the Humanoid Robot Race — Not Startups

The robotics industry is repeating a pattern that most founders refuse to see: the winners will be the companies that already know how to manufacture things at scale, not the companies that built the smartest software.

Jan Liphardt, CEO of OpenMind (which makes the OM1 operating system for robots), has spent enough time around both the software and hardware sides of robotics to call the outcome clearly: car companies will emerge as the real suppliers of humanoid robots. Not Tesla as a software shop. Not some venture-backed robotics startup in San Francisco. The car industry — people like Tesla, BYD, Hyundai, and Xiaomi — because they have something AI companies will never own: manufacturing DNA.

“My suspicion is that car companies will emerge as the real suppliers of hundreds of thousands or millions of robots,” Liphardt explained in his interview. And the reasoning isn’t sentimental or backwards-looking. It’s structural.

Why Software Companies Can’t Manufacture

Here’s the problem every robotics startup faces: scaling from prototypes to millions of units requires not just engineering skill, but institutional knowledge about supply chains, factory operations, quality control, logistics networks, and customer support at scale.

Tesla is the exception that proves the rule. Elon Musk built Tesla as a manufacturing company that happens to be software-heavy, not a software company that dabbled in manufacturing. That distinction matters. Tesla owns the stamping plants, the battery lines, the assembly processes. Even OpenAI-backed robotics startups don’t have that.

Liphardt put the gap in sharp terms: “For many robotics companies, they don’t necessarily have large-scale manufacturing in their DNA.”

What does manufacturing in your DNA mean? It means when demand doubles unexpectedly, you don’t call a contract manufacturer and hope. It means your cost structure is built for volume — your supply chain relationships, your purchasing power, your labor organization. It means you’ve already debugged the problems that kill startups: yield rates, supply disruption, warranty logistics, field service.

A startup can hire great engineers. A startup can raise $500M. But you can’t buy a supply chain that hums. You have to grow one, and it takes years.

The Historical Precedent

Every major technology platform eventually gets seized by the players who could manufacture at scale.

PCs were designed by Apple and Commodore, but were ultimately dominated by companies with distribution relationships and supply chain power — Dell (direct sales + manufacturing partnerships), HP (printers gave them retail relationships), and later Lenovo (owning IBM’s hardware division).

Smartphones were pioneered by startups, but iPhone’s dominance came partly because Apple already had manufacturing relationships through iPod and was willing to invest in Chinese supply chains. Android’s domination came through Samsung and eventually Chinese manufacturers with deep hardware expertise.

The pattern is consistent: the company with the smartest technology often isn’t the company that captures the market. The company that can manufacture and distribute at the lowest cost, fastest, and most reliably wins.

Humanoids are early. OM1 is software. Other startups (Figure, Boston Dynamics, etc.) build their own hardware. But the capital requirements and timeline to reach millions of units favor the incumbents.

What Car Companies Already Have

Take Tesla for a moment. Tesla runs factories. Tesla has relationships with battery suppliers, semiconductor manufacturers, and logistics networks across continents. When demand spiked, Tesla could ramp production faster than any startup because the infrastructure already existed.

Now extend that to BYD and Hyundai. BYD already manufactures millions of electric vehicles and has more EV production capacity than most countries. Adding humanoid manufacturing to an existing factory floor isn’t starting from zero — it’s a line extension. Hyundai acquired Boston Dynamics partly to get the IP, but the real value is being able to manufacture at Hyundai scale.

Chinese companies like Xiaomi (which invested in robotics startups) have the same advantage. They don’t just have manufacturing expertise — they have the cost structure to make robots under $1,000, not $20,000. And that becomes a market-making advantage over time.

A robotics startup would need to raise another $1B, build factories, negotiate supply agreements, hire operations teams, and operate at negative margins for years to compete on manufacturing cost. Most won’t make it that far.

The Software Layer Becomes Extractable

This outcome is actually healthy for the AI and software side of robotics. When manufacturing becomes commoditized (car companies making reference humanoids), the software layer becomes the moat for specialized capabilities.

This is exactly what Liphardt is building with OpenMind. OM1 isn’t tied to hardware. It can run on any humanoid. As car companies ship millions of reference humanoids (just like Android became the software that ran on any manufacturer’s hardware), the software layer will be where differentiation happens.

Companies will compete on the operating system, the middleware, the development tools, and the cloud platform for robot coordination — not on the robot hardware itself. This happened in phones (Android vs. Apple’s iOS), it happened in PCs (Windows vs. Mac), it’ll happen in robotics.

The startup that matters five years from now might not be a robot manufacturer at all. It might be the company that built the middleware layer, or the SaaS platform that lets you manage fleets, or the development framework that makes it trivial to create new robot behaviors.

But the startup that manufactures the robot body at scale? That company probably doesn’t exist yet, and if it does, it’ll get acquired by a car company or go out of business.

The Uncomfortable Implication

This is hard for founders to swallow because it means the exciting part — the robot hardware — won’t be where the venture returns are. The returns will be in the software layer, the services layer, or in becoming a supplier to the car companies.

It also means that if you’re a robotics startup with a prototype humanoid, your exit strategy probably involves one of two paths: (1) get acquired by a car company as a talent acquisition, or (2) become the software/OS layer that runs on car-company hardware.

Liphardt’s hedge here is important: he’s not trying to compete on manufacturing. OpenMind is being the software layer. The bet is that when millions of humanoids are running OM1, OpenMind becomes the platform on top of which new capabilities get built.

That’s a learned lesson from how software platforms win. Android didn’t win because Google manufactured phones. It won because it was the abstraction layer that manufacturers could build on. OpenMind is trying to be that for robotics.

FAQ

Isn’t OpenMind vulnerable if a car company makes their own OS?

Yes. But the timing matters. Car companies are years behind on software AI expertise. By the time they could build a competitive OS, OpenMind will have years of developer community, library ecosystem, and deployment history. That switching cost is real. Plus, many car companies might just license OpenMind rather than build competing software.

Has this pattern happened before in robotics?

Not quite. Industrial robotics (ABB, KUKA) were built by incumbents in heavy machinery, not startups. Drones saw competition from DJI (China) and 3D Robotics (US startup), but DJI won by controlling manufacturing costs and scale. The pattern holds: if you need volume and reliability, manufacturing incumbents win.

Could a startup avoid manufacturing by outsourcing to Foxconn or similar?

In theory, maybe. In practice, Foxconn is already working with every robotics company and several car manufacturers. Contract manufacturing takes time, quality can be inconsistent, and you’re competing with their other customers’ priorities. Owning the factory gives you priority and the ability to iterate on design based on what happens at scale.

Why is Tesla considered an exception?

Tesla started as a luxury car company with direct sales, so profit margins were high enough to invest in manufacturing from the start. Most robotics startups started with the software and thought they’d “figure out manufacturing later.” By then, it’s too late. If OpenMind (or any software company) suddenly decided to manufacture humanoids at the scale BYD needs, they’d be starting a decade behind.

Doesn’t this mean innovation slows down?

Not necessarily. Car companies are not monolithic — they’re competitive with each other. And the software layer (OS, middleware, developer tools) is where innovation will accelerate. What might slow is the creation of one-off robot designs for niche markets. Most humanoids will be reference designs manufactured by car companies. Customization will happen in software.

If car companies dominate, why does OpenMind exist?

Because the software layer still needs to be built, and car companies are terrible at it. Your Tesla isn’t running great software because Tesla owns car manufacturing — it’s running great software because Tesla hired software talent and focused on the stack. OpenMind is making the bet that the robot software layer is complex enough and specialized enough that car companies will license it rather than build it in-house.

What should a robotics startup do if they can’t compete on manufacturing?

Build software, not hardware. Build a tool, library, or platform that runs on reference hardware. If you’re building custom hardware, find a manufacturing partner early (even if it hurts your margins) and focus on becoming a software-first company. The hardware is table stakes, not the moat.

Will we see millions of humanoids in homes by 2030?

Only if car companies are building them. And only if the software (like OM1) makes them useful enough to justify the cost. The hardware problem is solved — car companies can make humanoids. The software problem is unsolved. Whoever makes the software indispensable wins. Whoever tries to compete on hardware after 2028 loses.

Is BYD actually going to make humanoids?

BYD has stated interest in humanoids and robotics. They have the manufacturing capacity, the capital, and the ecosystem. Whether BYD themselves builds the physical product or partners with others, the point is the same: whoever manufactures at scale will be a company with existing automotive/hardware expertise.

Can OpenMind stay independent, or will a car company acquire them?

That depends on whether OpenMind becomes a de facto standard for robot software. If they do, staying independent and licensing is more profitable than being acquired. If they become one of several competing OS layers, acquisition becomes more likely. The path forward is the same either way: become essential software, not secondary hardware.