# Which Companies Lead Humanoid Robot Deployment in 2026?

Chinese manufacturers captured more than four out of every five humanoid robot installations worldwide in 2025, according to data cited by the South China Morning Post — a figure that reframes the dominant Western narrative around this industry. [AGIBot](https://humanoidintel.ai/companies/agibot) alone commanded 30.4 percent of global installations, followed by [Unitree Robotics](https://humanoidintel.ai/companies/unitree-robotics) at 26.4 percent. Together, those two Shenzhen-area companies account for well over half the market before Western players even enter the count. Meanwhile, [Figure AI](https://humanoidintel.ai/companies/figure-ai)'s Figure 02 logged more than 1,250 operating hours on a live BMW production line and contributed to the assembly of more than 30,000 vehicles — the most specific commercial deployment metric any Western humanoid company has publicly disclosed. [UBTECH Robotics](https://humanoidintel.ai/companies/ubtech) shipped 1,079 full-size humanoids in 2025. [Apptronik](https://humanoidintel.ai/companies/apptronik) has raised more than $935 million. The installation data, the deployment hours, and the capital totals together tell a clearer story than any single headline: the humanoid industry is bifurcating along geographic lines, with China winning on volume and U.S. firms betting on capability and software differentiation.

---

## AgiBot and Unitree: China's Volume Play

Founded in Shanghai in 2023, [AGIBot](https://humanoidintel.ai/companies/agibot) has moved faster from founding to market leadership than any other humanoid company on record. Its 30.4 percent installation share in 2025 was built on mass production, an open-source platform strategy, and a portfolio spanning full-size humanoids, industrial machines, service robots, and quadrupeds — a breadth that lets it address manufacturing, logistics, entertainment, research, and customer-facing deployments simultaneously.

[Unitree Robotics](https://humanoidintel.ai/companies/unitree-robotics) took 26.4 percent of installations with a different but complementary playbook. The company originally built its reputation on agile quadrupeds before launching the H1 and G1 humanoids. Its G1 starts at approximately $13,500 — a price point that puts capable bipedal hardware within reach of university labs and AI research groups that previously couldn't afford meaningful embodied experiments. That accessibility is strategically important: it seeds the developer ecosystem that generates training data, open-source models, and eventual commercial applications.

The combined 56-plus percent share held by these two companies is difficult for Western startups to challenge on volume alone. The more plausible Western strategy is to out-compete on [physical AI](https://humanoidintel.ai/glossary/physical-ai) sophistication, reliability, and vertical-specific integration — which is exactly the bet the remaining companies are making.

---

## Figure AI: BMW Hours as the Real KPI

[Figure AI](https://humanoidintel.ai/companies/figure-ai) has distinguished itself by publishing deployment metrics that most competitors avoid sharing. Figure 02's 1,250-plus operating hours and 90,000-plus components loaded on an active BMW production line are numbers that matter precisely because they were generated in an uncontrolled, production environment — not a demonstration cell. More than 30,000 vehicles incorporated work from Figure 02 during that period.

The company's current-generation Figure 03 is powered by Helix, described as an embodied AI system that translates visual input and natural-language instructions into physical actions — a [vision-language-action model](https://humanoidintel.ai/glossary/vision-language-action-model) architecture that the broader industry is converging on. The stated roadmap moves from factories toward home deployment, though the BMW data suggests Figure is correctly sequencing structured industrial environments before tackling the chaos of domestic settings.

---

## Tesla Optimus: Manufacturing Advantage, Deployment Lag

[Tesla (Optimus Division)](https://humanoidintel.ai/companies/tesla-optimus) brings assets no pure-play robotics startup can replicate: at-scale manufacturing experience, an existing supply chain for motors, batteries, and power electronics, and computer vision systems refined through billions of vehicle-miles. The source material notes that Elon Musk has publicly acknowledged the initial production ramp will be "extremely slow" — an admission worth taking seriously given Tesla's historical optimism on timelines.

The strategic question for Tesla isn't whether Optimus will eventually be manufactured at scale; it's whether the software stack will reach the reliability threshold industrial customers require before Chinese competitors lock up the market on volume. The answer is not yet clear from available data.

---

## Apptronik Apollo: The NASA Lineage and $935M War Chest

[Apptronik](https://humanoidintel.ai/companies/apptronik) traces its technical origins to NASA's Valkyrie humanoid program — institutional knowledge in actuator design, whole-body control, and safety architecture that is genuinely hard to replicate quickly. Apollo has secured commercial agreements with Mercedes-Benz and logistics company GXO, and the company has integrated Gemini-based intelligence through a partnership with Google DeepMind.

The more than $935 million raised positions Apptronik to invest in manufacturing scale without the revenue pressure that constrains smaller players. The Google DeepMind integration is analytically significant: it suggests Apollo's differentiation will hinge on AI capability rather than hardware novelty, consistent with where the industry's center of gravity is moving.

---

## Boston Dynamics Atlas: Decades of Dynamics, Now Commercializing

[Boston Dynamics](https://humanoidintel.ai/companies/boston-dynamics) spent years demonstrating that dynamic bipedal locomotion was physically possible long before anyone could deploy it commercially. The fully electric Atlas now combines that locomotion heritage with hands designed for [dexterous manipulation](https://humanoidintel.ai/glossary/dexterous-manipulation) in industrial environments. Hyundai — which owns Boston Dynamics — plans to introduce production versions of Atlas at its own factories, giving the company a captive first customer with serious manufacturing throughput requirements. A concurrent Google DeepMind partnership on AI capabilities mirrors Apptronik's strategy: hardware credibility plus frontier AI.

---

## UBTECH Walker S2: Volume with Caveats

[UBTECH Robotics](https://humanoidintel.ai/companies/ubtech) shipped 1,079 full-size humanoids in 2025, making it one of the few companies with a four-digit deployment figure. The Walker S2's ability to autonomously swap its own battery — eliminating the need for human intervention to resume operation — is a meaningful feature for continuous industrial shifts. However, a Reuters investigation cited in the source material found that humanoids in general remain too slow and error-prone for most industrial work. That finding should temper any interpretation of shipment numbers as evidence of validated economic value.

---

## The Structural Problem No Demo Solves

Every company profiled here can produce impressive demonstrations. The harder question — which a Reuters investigation explicitly raised — is whether any humanoid today delivers reliable economic value in unscripted, production-grade conditions. The BMW deployment data from Figure is the closest the industry has come to a public answer, and even that involves structured automotive assembly rather than general-purpose manipulation.

The bifurcation between Chinese volume leaders and Western capability bets will define the next 24 months. If Chinese manufacturers continue scaling while gradually improving software quality through sheer data volume, the window for differentiation on AI sophistication narrows. If Western companies can demonstrate superior [zero-shot generalization](https://humanoidintel.ai/glossary/zero-shot-generalization) and task reliability before Chinese platforms close the capability gap, the market could support a more fragmented competitive structure. Neither outcome is predetermined.

---

## Key Takeaways

- **Chinese manufacturers held more than 80% of global humanoid installations in 2025**, with AgiBot (30.4%) and Unitree (26.4%) leading.
- **Figure 02 logged 1,250+ operating hours** on a live BMW production line, the most specific Western deployment metric publicly available.
- **Unitree's G1 starts at ~$13,500**, the most accessible price point for a capable humanoid, seeding research and developer ecosystems.
- **UBTECH shipped 1,079 full-size humanoids** in 2025, but a Reuters investigation found humanoids generally remain too slow and error-prone for most industrial work.
- **Apptronik has raised more than $935 million**, with commercial agreements at Mercedes-Benz and GXO and a Google DeepMind AI integration.
- **Tesla Optimus has manufacturing advantages no startup can match**, but Musk has acknowledged the production ramp will be extremely slow.
- **The industry's core test is economic reliability**, not locomotion or demonstration performance.

---

## Frequently Asked Questions

**Which humanoid robot company has the largest market share in 2025?**
AgiBot, founded in Shanghai in 2023, led global humanoid installations with 30.4 percent market share in 2025, according to figures cited by the South China Morning Post. Unitree Robotics followed with 26.4 percent.

**What is the cheapest humanoid robot available in 2026?**
Unitree's G1 humanoid starts at approximately $13,500, making it the most accessible full-capability humanoid currently available for purchase, according to the source material.

**Has Tesla Optimus been commercially deployed?**
As of the available source data, Optimus has not achieved significant commercial deployment. Elon Musk has publicly acknowledged that its initial production ramp will be extremely slow.

**How many humanoid robots has Figure AI deployed?**
Figure AI's Figure 02 loaded more than 90,000 components and accumulated over 1,250 operating hours on a BMW production line, contributing to the production of more than 30,000 vehicles — though total units deployed is not specified in available source material.

**Are humanoid robots actually useful in factories today?**
Deployment is increasing, but a Reuters investigation found that humanoids generally remain too slow and error-prone for most industrial work. Structured automotive assembly represents the most validated use case to date.

**What is AgiBot's strategy for humanoid market dominance?**
AgiBot combines mass production, an open-source platform strategy, and a broad product portfolio covering manufacturing, logistics, research, and customer-facing applications — prioritizing installation volume over premium positioning.