Introduction

Data from IT Juzi only scratches the surface: In 2025, the embodied intelligence industry secured 419.29 billion yuan in funding across 370 financing rounds.

What stands out more than the figures is the migration of talent—this sector has emerged as the primary destination for top-tier self-driving professionals.

Huawei’s Autonomous Driving Unit, DJI, Horizon Robotics, Momenta—these “incubators of talent” in the self-driving arena are systematically channeling founders, CTOs, and leading engineers into the robotics field.

When the “all-star team” of self-driving technology starts addressing challenges such as robotic precision tasks and mobility, we must pose the question:

Is this a deliberate “strategic technological shift” grounded in technological continuity, or a frantic pursuit where capital, sensing the self-driving narrative’s conclusion, desperately seeks the next “tech sensation”?

(For related reading, click: “Tencent’s Tang Daosheng: Self-Driving Experience Can Help Embodied AI ‘Avoid Pitfalls’! Tencent Cloud Fuels Over 40 Embodied AI Firms with Computing and Storage Power”)

I. The Context of the Migration: Self-Driving’s “Limit” vs. Embodied AI’s “Frontier”

Why now? The explanation lies in the differing stages of development between the two sectors.

For self-driving technology, particularly L4 Robotaxi, the industry is grappling with “convergence” challenges.

Debates over technical approaches (pure vision vs. multi-sensor fusion), business models (self-operated vs. technology-enabling), and applications (Robotaxi vs. trucking logistics) have persisted for years, with patterns gradually becoming clearer.

More critically, the “holy grail”—the large-scale commercialization of fully autonomous driving—has proven to require overcoming extended technical “long tails” and regulatory-ethical obstacles, with time and capital investments far surpassing initial expectations.

As the marginal benefits of innovation diminish, the brightest minds and most aggressive capital naturally seek new frontiers with higher potential and broader possibilities.

Embodied AI, in contrast, offers a vast “unexplored territory.”

Unlike self-driving’s focus on “mobility” as a singular task, embodied AI aims to equip physical entities (robots) with general intelligence capable of understanding and interacting with the physical world.

Its theoretical applications are limitless: spanning household services, industrial manufacturing, medical rehabilitation, and specialized operations. This “versatility” and “openness” provide immense opportunities for technological and capital advancement.

The core technologies refined by self-driving—perception (e.g., BEV bird’s-eye view, occupancy networks), prediction, planning, and control—are precisely the foundational skills robots require for autonomous action in complex, dynamic environments.

This high degree of technical overlap facilitates a seamless “transfer learning” path for self-driving professionals.

Thus, a talent migration driven by both “technology reuse” and “opportunity expansion” has naturally emerged.

II. Factional Dynamics: Strategic Variations Among “Self-Driving School” Robotics Firms

The incoming elite professionals are not blindly advancing; they bring deep expertise from their original fields, forming distinct entrepreneurial groups:

1. Huawei School (e.g., Tashihang, Zhiyuan Robotics): High-stakes, integrated software-hardware, ecosystem-focused.

Entrepreneurs from Huawei embody the “legion-style” and “full-stack self-research” philosophy.

Tashihang’s $120 million angel round set a record, with its “embroidery-capable robot” demonstrating extreme precision control.

Zhiyuan Robotics (founded by Zhihui Jun) is a standout, rapidly launching the humanoid robot “Farzoom A1” while investing upstream and establishing industrial funds to signal ecosystem-building ambitions.

Their strategy is clear: not content with being a mere robotics company, they aim to shape the “infrastructure” of the intelligent robotics era, mirroring Huawei’s success in communications and smart vehicles.

2. DJI School (e.g., Unitree, Strutt, Xuanji Power): Hardware-focused, product-driven.

Teams from DJI excel at integrating cutting-edge technology into stable, reliable, and aesthetically pleasing consumer products.

Unitree’s annual shipment of over 5,500 humanoid robots showcases its scalable manufacturing capabilities. Strutt’s $7,499 high-end intelligent mobility tool aims to create a new market category.

Xuanji Power emphasizes full-stack self-research of core components.

Their shared trait is a profound understanding of and mastery over hardware, costs, supply chains, and product design—critical for commercializing laboratory robots.

3. Horizon School (e.g., Digua Robotics, Vita Power): Bottom-layer focus, providing the “engine.”

As a chip and algorithm platform company, Horizon-affiliated startups naturally adopt an “enabling” approach.

Digua Robotics positions itself as a “universal robotics software-hardware base,” serving developers and universities to become the “Android” or “Horizon Journey Chip” of robotics.

Their strategy avoids direct competition in the complete machine market, instead seizing the “water seller” role in the supply chain—a lower-risk, higher-reward choice in the industry’s early growth phase.

4. Momenta School (e.g., Xinghaitu, Jianzhi Xinchuang): Algorithm- and data-driven, pursuing rapid iteration.

Momenta, renowned for its AI algorithms, extends this strength to its spin-offs.

Xinghaitu swiftly launched a humanoid robot, while Jianzhi Xinchuang focuses on “data infrastructure” for embodied AI.

Their rationale: self-driving succeeded through massive data loops, and robots must similarly address efficient data collection, labeling, and simulation to achieve general intelligence.

They aim to transplant Momenta’s proven data-driven methodology into robot learning.

III. The Extravaganza’s Undercurrent: 2026 Will Be the “Litmus Test” for Commercialization

The 419 billion yuan influx and top-tier talent have undoubtedly accelerated the industry’s progress.

Yet, all technological waves face the ultimate question: when, where, and how to generate revenue?

For embodied AI, 2026 will likely mark the critical transition from “tech demonstrations and funding hype” to “commercialization validation.”

Currently, the industry confronts core challenges:

1. The Scenario Dilemma:

Should firms bet on the distant future of “general-purpose humanoids” or first deeply cultivate “specialized wheeled/arm robots” for immediate needs?

Should they target price-sensitive, demanding household services or focus on industrial logistics, healthcare, or agriculture, where willingness to pay and task specificity are clearer?

Strategic choices outweigh mere effort.

2. The Cost Conundrum:

Core components (dexterous hands, joints, sensors) for agile, safe robotic interaction remain prohibitively expensive.

Reducing costs to commercially viable levels through innovation and scale is a prerequisite for mass production.

3. The Reliability Test:

Unlike self-driving cars operating on relatively structured roads, robots face far more unstructured physical environments and unpredictable interactions (with humans and objects).

Ensuring safety and stability under prolonged, high-load conditions poses engineering challenges far beyond flashy demonstrations.

Teams with self-driving “battlefield” experience may hold a true advantage here: they’ve navigated the harsh transition from demo to mass production, from technical marvels to stable products. They understand the art of balancing performance, cost, and safety in complex systems.

By 2026, embodied AI may enter a “rationalization phase,” but this won’t signal a bubble burst—rather, it’s a necessary step toward maturity.

Capital will shift from “betting on teams” to “betting on orders,” technology from “flashiness” to “practicality.” Entrepreneurs must answer: What problem does my robot solve that others cannot? Is the unit economics viable?

Firms rooted in real-world scenarios, refining products, and controlling costs will prevail in the market shakeout. Projects relying solely on concepts will face elimination.

In summary, the “second expedition” launched by self-driving veterans may not instantly create a robotics “Tesla.” Instead, their mission is to leverage elite engineering expertise to firmly “embed” AI from the virtual world of bits into the physical world of atoms—making intelligence truly “embodied” and machines genuinely useful to humanity.

Dawn has arrived, but the true test under the scorching sun has just begun.

What’s your take?

#SelfDrivingVehicles #AutonomousDriving #DriverlessTech #SelfDrivingCars