Robots Building Robots: Figure, Apptronik, and the $9.8 Billion Race to Scale Physical AI

The humanoid robotics industry crossed a threshold this month that separates research curiosity from industrial reality. At NVIDIA's GTC 2026 keynote on March 18, CEO Jensen Huang declared that "physical AI has arrived" and that "every industrial company will become a robotics company." That statement, once hyperbolic, now sits against a backdrop of billion-dollar funding rounds, factory-floor deployments measured in tens of thousands of production hours, and a cumulative funding pipeline that has surpassed $9.8 billion globally.

For anyone tracking the convergence of artificial intelligence and physical systems, March 2026 represents a turning point worth dissecting.

From Prototypes to Production Lines

The clearest signal that humanoid robotics has moved beyond demonstration comes from Figure AI. The company disclosed that its Figure 02 robot completed an 11-month deployment at BMW's Spartanburg plant, accumulating more than 1,250 hours of runtime across daily 10-hour shifts. During that period, the robots loaded over 90,000 parts and contributed to the production of more than 30,000 BMW X3 vehicles. Placement accuracy exceeded 99% per shift, with cycle times meeting the 84-second industrial requirement.

Those are not proof-of-concept numbers. They are production metrics that compete directly with human labor benchmarks. Figure has since retired the 02 platform, rolling its learnings into Figure 03, a third-generation humanoid that CEO Brett Adcock says will begin deploying on Figure's own production lines this year. The goal is striking: robots building robots, with fully recursive manufacturing targeted within 24 months.

Figure's Baku manufacturing facility currently has capacity for 50,000 robots per year across four production lines, with plans to scale to hundreds of thousands and eventually millions of units. The company's leasing model prices access at roughly $300 per month, or about $0.40 per hour, a fraction of minimum wage that could fundamentally reshape labor economics in logistics, manufacturing, and warehousing.

Apptronik's $935 Million War Chest

Figure is not alone. In February, Austin-based Apptronik closed a $520 million Series A extension backed by Google, Mercedes-Benz, B Capital, and the Qatar Investment Authority, bringing its total raise to over $935 million at a $5 billion valuation. The capital is earmarked for scaling production of its Apollo humanoid robot, expanding global deployments, and building a robot training and data collection center in Austin.

Apollo robots are already deployed in pilot zones at Mercedes-Benz, GXO Logistics, and Jabil facilities, performing component transport, sorting, and kitting tasks. Current deployments use external sensors and light curtains for safety, but Apptronik CEO Jeff Cardenas has indicated that next-generation models will support "collaborative safety," allowing the robots to work alongside humans without physical barriers.

The partnership with Google DeepMind is particularly notable. Apptronik is integrating Gemini Robotics AI models into Apollo, layering advanced reasoning capabilities onto a hardware platform that already has commercial traction. Morgan Stanley projects orders of $1 billion beginning in 2027, with per-unit costs around $80,000, comparable to a luxury vehicle.

NVIDIA's Physical AI Stack

What connects these deployments is NVIDIA's increasingly central role as the infrastructure layer for robotics AI. At GTC 2026, Huang outlined a three-computer architecture for robotics: a training computer, a synthetic data generation and simulation computer, and an onboard robotics computer. NVIDIA's open-source Isaac Lab for robot training and evaluation, Newton for GPU-accelerated physics simulation, Cosmos world models for neural simulation, and Groot foundation models for robot reasoning collectively form what the company calls the "physical AI stack."

The significance is hard to overstate. Classical robotics relied on painstakingly hand-coded behaviors for narrow tasks. NVIDIA's approach treats compute as data: developers pre-train world foundation models on internet-scale video and human demonstrations, then use simulation to generate massive synthetic training datasets. This closes what Huang calls the "physical AI data gap," the reality that no amount of real-world data can prepare a robot for every possible scenario.

Over 110 robotics companies are now working with NVIDIA's platform, including ABB, Universal Robots, KUKA, Caterpillar, and Hyundai. The ecosystem spans factory automation, logistics, healthcare, and autonomous vehicles. When Huang says every industrial company will become a robotics company, he means it literally: the platform is designed to make robotics a horizontal capability rather than a vertical specialty.

The Funding Landscape: China and the West

The capital flowing into humanoid robotics reflects a global conviction that this market has crossed from speculative to investable. Cumulative industry funding exceeded $9.8 billion by the end of 2025, according to ResearchAndMarkets, and the pace is accelerating. In the first nine months of 2025 alone, China's robotics sector reported 610 financing agreements totaling 50 billion yuan (approximately $7 billion), a 2.5x increase over the prior year. The embodied intelligence segment accounted for 243 deals in Q3 2025.

In the United States, Tesla has earmarked $20 billion in capital expenditures for 2026, a figure widely understood to include significant Optimus humanoid development. Figure AI was valued at $39 billion in its most recent round. Yann LeCun's AMI Labs, focused on "world models" for physical AI, raised $1.03 billion in March at a $3.5 billion valuation.

The International Federation of Robotics forecasts that the global industrial robot market has reached $16.7 billion, with humanoids emerging as the fastest-growing segment. Commercial adoption is expected to unfold in three waves: automotive manufacturing and logistics from 2025 to 2030 at $80,000 to $250,000 per unit; healthcare and service applications from 2028 to 2032; and consumer home deployment from 2029 onward.

What This Means for the Broader Market

Morgan Stanley Research estimates that nearly $3 trillion of AI-related infrastructure investment will flow through the global economy by 2028, with more than 80% of that spending still ahead. Physical AI and robotics represent a growing share of this capital allocation, as the technology shifts from what Morgan Stanley describes as "speculative tech spending" to "industrial build-out."

For public markets, the implications are already visible. AI adopters are seeing cash-flow margin expansion outpacing the global average by 2x. But the market has moved from pricing potential to demanding proof of monetization. NVIDIA posted revenue up 73% year over year in its most recent quarter, yet shares fell 5% on earnings day. The Magnificent 7 is down roughly 7% year to date. The era of rewarding any company with "AI" in its investor deck is giving way to one that demands evidence of defensible competitive position.

The robotics subsector may be less exposed to this repricing. Companies like Figure and Apptronik are generating real production data, signing commercial contracts, and deploying units that run for thousands of hours in industrial settings. That is fundamentally different from software companies projecting future AI benefits.

Educational Takeaway

Physical AI sits at the intersection of several research verticals we cover at AdValorem: AI and robotics, pre-IPO market dynamics, and the broader restructuring of capital allocation toward hard technology. The transition from prototype to production line is exactly the kind of inflection point that separates speculative narratives from measurable outcomes. Understanding where commercial traction exists, and where it does not, remains the central task for anyone studying this space.