According to Amsterdam-based startup data provider Dealroom, robotics companies had already raised $55.8 billion by early June 2026—nearly double the previous annual record. June alone read like a never-ending pipeline of deals: more than $1.75 billion in just twelve disclosed funding rounds, driven by embodied AI and defense robotics. Generalist AI: $400 million for models that teach robots real-world tasks. Allen Control Systems: $200 million for drone defense.

But the real showstopper came from a mid-sized company in Baden-Württemberg. NEURA Robotics, based in Metzingen, announced a Series C round of up to $1.4 billion on June 10. The list of investors reads like a “Who’s Who” of the industry: Tether, Qualcomm, Amazon, and Nvidia, along with Bosch, Schaeffler, and the European Investment Bank. According to a Financial Times report, the valuation stood at around seven billion dollars—though NEURA itself did not confirm the figure. Physical AI made in Germany, financed by half of Silicon Valley. We haven’t seen anything like this in a long time.

The question no one is asking out loud: Who’s going to buy all this?

This is where things get uncomfortable. A mid-June assessment sums it up bluntly: In the humanoid robot market, supply already exceeds demand. The factories are up and running, the capital is there, the robots are rolling off the assembly line—but the paid orders in the hoped-for numbers are failing to materialize. Meanwhile, the real work is being done by their unspectacular colleagues: autonomous robots are arriving at Toyota’s plants, and overall robot orders remain stable. Stable—not euphoric.

A robot that earns its keep

Ironically, it’s a humanoid that proves things can be done differently—and right where it counts: in accounting. At BMW’s plant in Spartanburg, South Carolina, what the entire industry dreams of is a reality: a real business operation with a price tag. According to consistent market observations, Figure AI has an initial fleet of 40 Figure-03 units in commercial use there, billed at around $25 per robot-operating hour—figures that, for the first time, make humanoid automation a quantifiable expense rather than a research budget item.

Physical AI Can Deliver Measurable Added Value Under Real-World Conditions

The key lies in the lead-up. Before billing began, the technology had to prove itself over several months. And here, BMW itself provides the figures: Within ten months, the predecessor model, Figure 02, supported the production of more than 30,000 BMW X3s, working ten-hour shifts daily from Monday through Friday. In total, the robot handled over 90,000 components and took approximately 1.2 million steps over roughly 1,250 operating hours—all while retrieving and positioning sheet metal parts with millimeter precision for the welding process. These aren’t stage numbers; this is shift operation. And BMW draws a clear conclusion from this: The pilot confirmed that Physical AI delivers measurable added value under real-world conditions. The findings will now be used to further develop and scale the applications.

Technically, the whole system is powered by an in-house “brain.” Figure parted ways with OpenAI in early 2025 and now runs Helix—a vision-language-action model developed entirely in-house—on every unit. In other words: The robot sees, understands a command in natural language, and acts—without anyone having to write code for every single action. What makes this example so valuable isn’t the hardware, but the fact that someone can provide reliable figures: hours, unit counts, hourly rates. Not a glossy video, but an invoice line item—and thus the exact opposite of what most competitors deliver.

This year, pilot projects will transition to mass production

This is exactly what will separate the wheat from the chaff in 2026. According to the Association for Advancing Automation, North American manufacturers ordered 36,766 robots worth $2.25 billion last year—this year is now seen as the moment when pilot projects give way to mass production. Solid craftsmanship, not a gold rush! Especially since the global growth forecast for manufacturing has been lowered from 2.9 to 2.6 percent due to geopolitical turbulence. Investors are betting on a future that today’s order intake has not yet delivered.

Those who look away now will pay the price later

Ignoring this would still be a mistake, because the wave of investment is driving down component prices, broadening supply chains, and accelerating software development. Those who look away now will pay the price later—buying into the hype is the more expensive option. The uncomfortable issues are: throughput, availability, cost per operating hour, and documented real-world deployments with paying customers. The BMW example sets the standard—and at the same time exposes just how few meet it. Today, value creation lies in edge AI, cobots, vision-guided systems, and digital twins—not in every headline that comes striding in with a big bang. The real risk of this boom, then, is not getting in too early, but investing in the wrong format.