Physical AI Crosses the Factory Floor: Figure at BMW, Gemini in the Loop, and a $6 Billion Quarter for Robotics

The conversation about humanoid robots in factories has shifted from "when" to "how fast." In the span of a single quarter, Figure AI completed an 11-month production deployment at BMW, Google DeepMind embedded its Gemini foundation models into 20,000 industrial robots through a partnership with Agile Robots, China opened its first 10,000-unit humanoid production facility, and global robotics startups raised over $6.1 billion across 134 funding rounds. The physical AI sector is no longer pre-revenue hype. It is entering the production phase.

Figure AI at BMW: From Demo to Data

The most consequential data point in robotics this year may be a single number: 30,000. That is how many BMW X3 vehicles Figure AI's humanoid robots contributed to producing at the Spartanburg, South Carolina plant over 11 months. The Figure 02 robot ran 10-hour shifts Monday through Friday, handling over 90,000 sheet metal parts with five-millimeter precision in two-second cycle times. It logged approximately 1,250 hours of runtime and covered an estimated 1.2 million steps.

Those results were consequential enough for BMW to take the next step. In late February 2026, the automaker announced it will deploy humanoid robots at its Leipzig plant in Germany, marking the first time Physical AI of this kind has entered European automotive production. The Leipzig project will test humanoid robots in high-voltage battery assembly and component manufacturing, expanding the use case from simple sheet-metal loading into more complex production environments.

For researchers tracking the economics of physical AI adoption, the BMW deployment offers real operational data. The key metric to watch is not just throughput but intervention rate: how often a human must pause or reset the robot. Figure AI has stated its goal is zero interventions per shift. The closer that number gets to zero, the stronger the business case becomes for replacing ergonomically demanding tasks with autonomous systems.

Google DeepMind's Factory Strategy

While Figure AI is proving that humanoid robots can perform reliably on a production line, Google DeepMind is making a different bet: that the foundational intelligence layer matters more than the hardware form factor. In March 2026, DeepMind announced a strategic research partnership with Munich-based Agile Robots, which has already deployed over 20,000 robotic solutions worldwide. The partnership embeds DeepMind's Gemini Robotics foundation models into Agile's industrial hardware, creating a feedback loop where operational data from real factories continuously improves the underlying AI.

This follows a similar partnership between DeepMind and Boston Dynamics earlier in the year. The pattern is clear: DeepMind is positioning Gemini as the operating system layer for industrial robotics, partnering with hardware companies rather than building its own robots. The sectors targeted include electronics manufacturing, automotive, data centers, and logistics.

The strategic question for the industry is whether the value will accrue primarily to the hardware makers, the software layer, or the companies that control the deployment data. DeepMind's approach suggests it believes the intelligence layer will ultimately command the largest share of the value chain, a thesis that echoes how Android captured mobile economics without manufacturing phones.

China's 10,000-Unit Production Line

On March 29, a high-capacity humanoid robot production facility in Guangdong, China, officially began operations. Built by Leju Robotics, it is reportedly the first facility in the country capable of manufacturing up to 10,000 humanoid robots per year, a milestone that marks China's transition from prototype experimentation to industrial-scale production.

The competition behind Leju is fierce. Agibot recently celebrated production of its 1,000th humanoid robot. Unitree Robotics is pursuing a $580 million funding round to support a facility targeting 75,000 units per year. UBTECH Robotics aims to produce 5,000 units annually while driving per-unit costs below $20,000. Xpeng Robotics has begun factory trials of its "Iron" humanoid in Guangzhou.

The production numbers underscore a critical divergence between the U.S. and Chinese approaches to physical AI. American companies like Figure AI and Physical Intelligence are optimizing for intelligence and general-purpose capability, commanding multi-billion-dollar valuations with small teams. Chinese manufacturers are optimizing for unit economics and manufacturing scale, racing to drive per-unit costs down to levels where mass deployment becomes commercially viable. Both strategies carry risk. The U.S. approach risks building expensive intelligence without a scalable body. The Chinese approach risks building scalable bodies without sufficiently capable intelligence.

The Funding Numbers

March 2026 recorded 134 robotics funding rounds totaling $6.13 billion globally, significantly up from January and February, according to Robots and Startups. The standout rounds illustrate the breadth of capital flowing into the sector:

• Physical Intelligence: In talks to raise approximately $1 billion at an $11 billion valuation, roughly doubling its $5.6 billion valuation from just four months earlier. Founders Fund is expected to participate alongside Lightspeed Venture Partners and returning backers Thrive Capital and Lux Capital. The company has raised over $1 billion total with approximately 80 employees and no commercial product.
• Mind Robotics: The Rivian spinout raised $500 million in a Series A co-led by Accel and Andreessen Horowitz, following a $115 million seed round. Founded by Rivian CEO RJ Scaringe, it is building AI-enabled robotic systems for complex manufacturing tasks.
• Shield AI: Raised $2 billion in a Series G for autonomous defense systems.
• Rhoda AI: Emerged from stealth with a $450 million Series A.
• Arda: Founded by former OpenAI VP of Research Bob McGrew, raised $70 million at a $700 million valuation to automate manufacturing using AI and video models to train robots.

For context, 2026 is on pace for over $20 billion in total robotics funding. The capital is not concentrated in a single sub-sector. It spans humanoid robots, industrial automation, defense autonomy, agricultural robotics, and the increasingly important data infrastructure layer that enables robot learning.

Tesla's Optimus Timeline

Tesla has confirmed that Optimus 3 production begins this summer at the Fremont factory, with a production line designed for one million units per year. Elon Musk has stated that high-volume production should follow by summer 2027, with Optimus 4 already in development. Gigafactory Texas is being built to eventually handle 10 million units per year. Inside Tesla's own facilities, Optimus robots are already handling 4680 battery cells, folding materials, and performing basic logistics tasks. More than 200 Optimus Gen 3 units are planned for deployment at Gigafactory Texas in 2026.

Tesla's cost target of $20,000 per unit at scale, down from current estimates of $30,000 to $80,000, would fundamentally reshape the economics of humanoid deployment if achieved. At that price point, a robot that operates 16 hours per day costs less than minimum wage in most developed markets within its first year.

What This Means for the Research Landscape

The convergence of factory deployments, foundation model partnerships, production scaling, and record funding suggests that physical AI is entering a phase transition. The BMW data proves the technology works in production. The DeepMind partnerships prove that the intelligence layer can be decoupled from hardware. The Chinese manufacturing push proves that unit economics are improving rapidly. And the funding numbers prove that institutional capital has conviction in the sector's trajectory.

The educational takeaway for anyone tracking alternative technology markets: physical AI is following the classic deep-tech adoption curve. The prototyping phase is over. The pilot phase produced real data. The production phase is beginning. The companies, partnerships, and deployment data emerging in 2026 will define which players capture value as this market scales toward the trillions. This is the landscape AdValorem's AI Robotics research vertical tracks in depth, and the developments of Q1 2026 have made it one of the most data-rich periods in the sector's history.