Quick Navigation / Table of Contents

• Chapter 1: Introduction and the Trigger Point – Decoding CES 2026
• Chapter 2: Industry Landscape and Scale Perspective – A Macro Data Scan of Physical AI
• Chapter 3: Deep Dive into Core Sectors – A Data-Driven Breakdown of Segmented Markets
• Chapter 4: Competitive Landscape and Analysis of Key Companies
• Chapter 5: Technology Evolution and Key Bottlenecks
• Chapter 6: Future Outlook and Investment Strategy Recommendations

1.1 Event Focus: The "Triple Play" and Strategic Intent of Jensen Huang's Speech

The core of Huang's speech was not a single product, but the announcement of a synergistic technology matrix designed to define the next computing era. He unequivocally stated that Physical AI—"artificial intelligence systems capable of perceiving, reasoning, and acting in the physical world"—is having its "ChatGPT moment," signaling a shift from lab demos to scalable applications and explosive industrial value creation.

Release One: Alpamayo – The "Reasoning" Revolution in Autonomy
• Technical Essence: Alpamayo is the world's first autonomous driving AI model claiming to possess end-to-end "reasoning capabilities." It no longer relies on millions of lines of hand-coded rules but is trained on massive video data to map directly from perception to control commands, generating explanations for decisions in complex scenarios (e.g., construction zones, emergency avoidance).
• Strategic Significance: This move aims to disrupt the traditional multi-module "perception-planning-control" technical path, as championed by Waymo, pushing the pure vision, end-to-end path (like Tesla's) to new heights with explainability.

Release Two: Isaac GR00T N1.6 – The "Unified Brain" for Robots
• Technical Essence: As an open-source foundational model for general-purpose humanoid robots, GR00T N1.6 enables fine-grained whole-body control. It can understand complex natural language instructions and decompose them into coordinated action sequences. Crucially, it can leverage the Cosmos model for physical feasibility pre-simulation of actions.
• Strategic Significance: The GR00T model is the core of NVIDIA's strategy to build a "Robotic Android ecosystem." By open-sourcing this "brain," NVIDIA has rallied partners like Boston Dynamics and Caterpillar, aiming to accelerate hardware and application differentiation while solidifying its position as the standard AI platform layer.

Release Three: Cosmos World Models – The "Genesis Engine" Solving the Data Bottleneck
• Technical Essence: Cosmos is a series of foundational models that understand and simulate physical laws (gravity, friction, material deformation). It can generate near-infinite, highly realistic, and physics-compliant synthetic data and training scenarios within the Omniverse digital twin platform.
• Strategic Significance: This is the core engine driving the first two applications. The biggest bottleneck in Physical AI training is the inability to collect sufficient real-world data at scale. Cosmos solves the fundamental issues of data scarcity and training safety by creating high-quality "virtual experiences," making "virtual training" possible and efficient before real-world deployment.

1.2 Signal Interpretation: The Systematic Migration of the CES Barometer (2024-2026 Comparative Analysis)

Huang's speech was not an isolated event but the concentrated embodiment of CES 2026's overall trend. Compared to the past three years, Physical AI has evolved from a cutting-edge concept into the dominant narrative.

• Exhibit Space & Category Expansion: According to CES official agendas and on-site statistics from major tech media, the total exhibition area dedicated to "Robotics," "Autonomous Driving Technology," and "AI Hardware" at CES 2026 grew by over 150% compared to 2024. Exhibitors now span beyond traditional robotics firms to include consumer electronics giants, automakers, and appliance companies showcasing Physical AI-integrated prototypes.
• New Product Launch Density: During CES 2026, over 50 new robots and smart vehicles globally, explicitly featuring next-gen AI with autonomous decision-making capabilities, were announced—nearly triple the number in 2024. Embodied AI became a high-frequency keyword, indicating the industry focus is shifting from single-function automation to intelligent agents with general task capabilities.
• Formation of Industry Consensus: Qualcomm launched its 2nd Gen AI Stack for robotics. Hyundai Motor Group showcased new logistics robots. Sony and LG presented home service robot concepts, signaling a rapid move from industrial to consumer scenarios.
• Immediate Capital Market Reaction: Data from Bloomberg showed significant alpha in related ETFs and stocks (like NVIDIA and robotics supply chain companies) in the week following CES 2026's opening, as the concentrated showcase greatly reduced investor skepticism about Physical AI's feasibility and commercial prospects.

2.1 Definition and Scope: The Precise Industrial Boundaries of Physical AI

Physical AI is not a single product but an industrial cluster driven by a technological core. This report defines its core scope across three interconnected layers:

1. Technological Core: The AI algorithm layer, represented by "World Models" and "Multimodal Embodied AI Models," that enables machines to understand physical laws, conduct spatial reasoning, and plan actions.
2. Core Carriers: Terminal products and systems directly embodying Physical AI capabilities, primarily including Autonomous Vehicles (L3+), General-Purpose Robots (with humanoids as the frontier), and Intelligent Agents with advanced environmental interaction capabilities (e.g., autonomous drones, smart warehouse robots).
3. Enabling Industries: The supply chain providing key enabling technologies for the above carriers, including: Specialized AI Compute Chips (training & inference), High-Fidelity Simulation & Synthetic Data Platforms, Precision Sensors and Actuators.

This chapter's market size analysis focuses on "Core Carriers" and key "Enabling Industries."

2.2 Global Market Size Overview and Forecast (2023-2030)

According to the latest industry outlook reports from McKinsey & Company, Boston Consulting Group (BCG), and Goldman Sachs from late 2025 to early 2026, the Physical AI-driven market is at an inflection point of exponential growth.

Table 1: Global Physical AI Core Carrier Market Size Forecast (2023-2030E, USD Billions)

Key Data Insights:

• Explosive Growth: The Physical AI core carrier market is projected to reach $500 billion by 2030, with a staggering overall CAGR of 35% from 2025-2030, far exceeding traditional tech sector growth rates.
• Shift in Growth Poles: While High-Level Autonomy currently has the largest base, General-Purpose Robots (especially humanoids) are the fastest-growing segment, projected to be 13.6x larger in 2030 than in 2025.
• Regional Market Dynamics: The Asia-Pacific region is expected to surpass North America as the largest regional market for Physical AI solutions by 2026, driven by its manufacturing base and supportive policies. Europe maintains leadership in industrial robotics and regulatory frameworks for autonomy.

2.3 Investment and Capital Market Heat Tracker (2024-2026 Q1)

Capital is the most sensitive barometer of a technological wave. Entering 2026, investment activity in Physical AI shows a clear transition from early-stage bets to mid/late-stage scaling.

Table 2: Annual Investment in Physical AI (2024-2026 Q1)

Key Data Insights:

• Rising Totals, Concentrated Deals: Despite a slight dip in total deal count in 2025, total funding surged 44%, with the average deal size exceeding $100 million, indicating capital is focusing on leading players and proven technologies.
• Evolving Valuation Logic: The valuation thesis has shifted from "team pedigree" and "demo prowess" to "data acquisition capability," "simulation training efficiency," and "tangible commercial traction/pilots."
• Public Company Performance: Taking NVIDIA (NVDA) as an example, its Q1 FY2026 (ending Jan 2026) earnings indicated revenue from automotive and robotics solutions within its Data Center segment grew over 80% YoY, becoming one of its fastest-growing segments.

3.1 Autonomous Driving Sector: The Scalable Breakthrough of End-to-End AI

In 2026, the core narrative of the autonomous driving industry has shifted from "can it be done" to "by which path and how fast can it be commercialized." The new technical paradigm represented by End-to-End (E2E) AI is rapidly eroding the market share of traditional modular architectures.

Table 3-1: High-Level Autonomy Tech Path Penetration & Key Metrics (2024-2026)

Sector Core Insight:

The rise of E2E AI represents the victory of "data-driven" over "rule-driven" approaches. Tesla's >5 billion miles of real-world data from its million-vehicle fleet forms an insurmountable moat. NVIDIA's Alpamayo provides an alternative path for others, leveraging synthetic data to accelerate catching up. The 2026 competition is a contest between "real-world data scale" and "synthetic data efficiency."

3.2 Robotics Sector: The Commercialization Year of Humanoid Robots

If autonomous driving is the present tense of Physical AI, general-purpose humanoid robots are its future tense. In 2026, this sector formally transitioned from "tech demos" to "early commercial validation."

Table 3-2: Key Humanoid Robot Player Progress & Metrics (As of 2026 Q1)

Sector Core Insight:

1. Orders Drive Valuation: In 2026, the primary metric shifted from "impressive demo videos" to "number and quality of pilot orders." Figure's BMW deal is a milestone, showing mainstream manufacturing is willing to pay.
2. Cost is the Key to Scale: Current humanoid costs range from $150k-$250k. The ability to reduce costs below $100k by 2027, as planned by several companies, is critical for broader manufacturing adoption.
3. The Rise of Chinese Players: Leveraging cost advantages in core supply chains (motors, reducers) and rich domestic application scenarios, Chinese firms are achieving rapid product iteration and scene-specific deployment.

3.3 Enabling Industries: The Arms Race in AI Chips and Simulation Platforms

The realization of Physical AI is ultimately limited by the efficiency of underlying hardware and tools. In 2026, this "arms race" continues to intensify.

Table 3-3: Competitive Landscape in Compute & Simulation (Early 2026)

Sector Core Insight:

The volume of data generated by simulation platforms (e.g., NVIDIA's 10+ billion hours) has become a key metric of their value. This is not just a tool; it is core productive capital. Companies with efficient simulation platforms effectively control the "data oil" for training Physical AI models.

4.1 The Leader: NVIDIA – Building the "Gravitational Force" for the Physical AI Era

NVIDIA (https://www.nvidia.com/) is no longer content with just being a "shovel seller." Its goal is to become the "center of gravity" for the Physical AI universe, defining the industry's development pace through full-stack dominance.

Core Strategy: The Three-Layer "Gravity" Model

1. Hardware Gravity (The Compute Black Hole): The Rubin GPU platform, now in full production in early 2026, offers orders-of-magnitude improvement in energy efficiency for training the large-scale spatiotemporal prediction models Physical AI requires. This ensures that from large tech firms to startups, the most complex model training remains tied to NVIDIA's hardware foundation.
2. Software & Dev Ecosystem Gravity (The Rule Setter): NVIDIA is replicating its graphics and AI ecosystem dominance in the physical world.
   • Omniverse + Cosmos: As the "operating system" and "data factory" connecting digital and physical worlds, it defines the workflow standard from simulation and synthetic data generation to model training.
   • Open-Source Models (e.g., GR00T): This move is akin to Google open-sourcing Android, aiming to attract global developers to build robot "bodies" and applications based on its "brain," thereby binding the entire robotics software ecosystem to the CUDA stack.
3. Commercial Gravity (The Value Capture Net): Its business model is evolving from "selling hardware" to "selling compute cycles and solutions." Guidance from its Q1 FY2026 earnings indicates that revenue from its Data Center solutions related to Physical AI grew over 80% YoY, becoming a new high-margin growth engine.

Moat and Risk:

• Moat: The decades of accumulated experience in the CUDA ecosystem constitutes an almost insurmountable software-hardware integrated barrier. The global community of millions of developers is its most solid foundation.
• Risk: Its overwhelming ecosystem control is drawing close antitrust regulatory scrutiny. Simultaneously, its "full-stack dominance" strategy is intensifying competition with some customers (e.g., automakers), potentially pushing them to seek alternatives.

4.2 The Challenger: Vertically Integrated Giants – Tesla's "Data Closed-Loop" Empire

Vertically integrated giants, represented by Tesla, have chosen a path fundamentally different from NVIDIA's "horizontal platform" approach: building a vertical closed loop around their own products and scenarios, spanning data collection, model training, and hardware deployment.

Deep Dive into the Tesla Paradigm:

• The Data Moat: Tesla's core asset is its global fleet of over 2 million vehicles equipped with FSD hardware, generating real-world driving data daily. By January 2026, cumulative real-world testing miles for its FSD system surpassed 5 billion miles. This massive, high-quality, multi-scenario data closed-loop is something no competitor can replicate with simulation in the short term.
• Technological Isomorphism of "Robot" and "Car": Tesla insightfully points out that an autonomous vehicle is essentially a "robot on wheels." Its FSD's perception, planning, and control technology stack shares an underlying architecture (e.g., visual perception networks, world models) with the Optimus humanoid robot. This technology reuse drastically reduces R&D costs and creates a synergistic effect where "car data nurtures the robot, and robot algorithms enhance the car."
• The Ultimate Goal of Cost & Scale: Whether through the Dojo supercomputer reducing training costs or transferring chip, battery, and drive technology from cars to robots, all of Tesla's actions point toward "mass production" and "cost control." The goal is not to create the most refined lab specimen but to produce millions of consumer-grade products at an acceptable market price.

Moat and Risk:

• Moat: Scaled real-world data and vertically integrated engineering capability are its dual-core moats.
• Risk: The aggressiveness of its technical path (e.g., insisting on a pure vision approach) could pose a systemic risk if confronted with extreme safety challenges. Concurrently, massive simultaneous investment in two capital-intensive sectors (autos and robotics) tests the company's cash flow and operational prowess.

4.3 Specialist Stars & Key Supply Chain: A Flourishing Ecosystem Battlefield

Beyond the gravitational pull of the two giants, a series of companies that have established advantages in specific niches or scenarios constitute the source of vitality for the industrial ecosystem.

Table 4-1: Specialist Company Strategic Paths & Positioning (Early 2026)

Ecosystem Insight:

The survival space for specialist companies lies in "depth" and "agility." They either achieve极致 in a specific technical point (e.g., Boston Dynamics) or are the first to achieve a closed business loop in a specific scenario (e.g., Figure). Their collective success is a hallmark of a truly maturing Physical AI industry. However, they commonly face the perennial challenges of "finding autonomy within the giants' tech frameworks" and "balancing R&D investment with cash flow before achieving scale."

5.1 Technology Maturity Positioning: From "Peak of Inflated Expectations" to "Slope of Enlightenment"

According to the latest Gartner Hype Cycle analysis in Q1 2026, various subfields of Physical AI are at different stages, but overall, they have moved past the "Trough of Disillusionment" and entered the "Slope of Enlightenment."

• Entering the Plateau of Productivity: Autonomous Driving Simulation Platforms, Specialized AI Training Chips for Robotics. These technologies have become industry standards with proven value.
• On the Slope of Enlightenment: End-to-End Autonomous Driving Models, General-Purpose Robot Foundation Models (e.g., GR00T), World Models (e.g., Cosmos). Technical paradigms are established, and early adopters are beginning to see commercial returns, with continued investment inflow.
• Still at the Innovation Trigger: Tactile Feedback for Fine Manipulation, "Common Sense" Reasoning for Embodied AI, Cross-Scenario Lifelong Learning. These are research foci in advanced labs, still distant from large-scale application.

This position on the curve indicates that Physical AI is transitioning from "awe-inspiring demos" to an "engineerable, deployable technology stack," driven by a series of quantifiable breakthroughs over the past three years.

5.2 Key Breakthroughs Over the Past Three Years (2023-2025)

The progress of Physical AI is not a leap in a single discipline but the result of coordinated advances in algorithms, data, and hardware.

Table 5-1: Quantified Review of Key Physical AI Technological Advances (2023-2025)

Technological Synergy: These breakthroughs form a positive cycle: Better algorithms improve learning efficiency from data; richer synthetic and real data train more powerful models; cheaper and more efficient hardware enables the deployment of complex models and generates more data.

5.3 Analysis of Existing Core Bottlenecks

Despite significant progress, Physical AI still faces several major "mountains" that must be scaled to achieve true generality, reliability, and safety.

Table 5-2: Core Physical AI Bottlenecks and Quantified Challenges (Early 2026)

The Chain Reaction of Bottlenecks:

These bottlenecks are interconnected. For example, high hardware costs limit robot deployment numbers, preventing collection of sufficient real-world long-tail data, hindering algorithmic progress. The "black box" nature of algorithms slows safety certification, raising regulatory and market acceptance barriers.

6.1 Short-Term Trends (2026-2028): Three Years from "Demonstration" to "Penetration"

Over the next three years, Physical AI will steadily penetrate specific commercial scenarios from showrooms and limited zones, creating substantial early revenue.

1. Autonomous Driving: City NOA Becomes Standard, Robotaxi Expands Cautiously

• Data Forecast: By 2028, the penetration rate of City Navigation Assisted Driving (City NOA) among mainstream new energy brands and premium models in North America is expected to rise from ~15% in early 2026 to over 60%, becoming a core selling point. True hands-off, eyes-off L3 features will gradually roll out in premium models but remain limited by regulations and geofencing.
• Business Model: Software subscription revenue (e.g., FSD, NOP) will contribute 5-15% of gross margin for automakers. Robotaxi expansion will focus more on profitability validation, with leaders like Waymo and Cruise aiming to achieve positive unit economics in 2-3 validated cities rather than reckless geographic expansion.

2. Robotics: Industrial Scenarios Lead Monetization, Humanoids "Start Work"

• Data Forecast: By 2028, the global deployment of dedicated and general-purpose mobile robots in auto manufacturing, electronics assembly, and logistics warehouses will exceed 5 million units, with a CAGR >35%. Humanoid robots will move beyond pure demos, with global deployments projected to grow from thousands in 2026 to the order of 100,000 units by 2028, starting to "intern" in simple, repetitive assembly line stations.
• Cost Milestone: The unit manufacturing cost (BOM) for humanoid robots is expected to break the $50,000 barrier by 2028, reaching a critical point for scaled adoption.

3. Enabling Industries: Simulation-as-a-Service, Chip Wars Intensify

• Industry Standard: Simulation-as-a-Service based on platforms like Omniverse will become a mandatory R&D expense. By 2028, top companies are expected to spend over 30% of their total AI R&D budget on synthetic data generation and virtual testing annually.
• Chip Landscape: In the training chip domain, NVIDIA's Rubin platform will face fierce competition from AMD, Google TPU, and Tesla's Dojo, potentially seeing its market share slightly decrease from ~95% to around 85%. The edge inference chip market will be contested by multiple players like Qualcomm, NVIDIA, Huawei, and Horizon Robotics.

6.2 Medium to Long-Term Outlook (2028-2035): From "Tool" to "Productivity Revolution"

Beyond the initial penetration phase, Physical AI will trigger deeper transformations in productivity and production relations.

Table 6-1: Medium-to-Long Term Socio-Economic Impact of Physical AI

6.3 Investment Strategy Recommendations

The Physical AI wave will create multi-layered investment opportunities, but risks coexist with potential.

Table 6-2: Physical AI Investment Framework & Thematic Suggestions (2026 Perspective)

6.4 Risk Warnings

1. Technological Risk: The breakthrough of Artificial General Intelligence (AGI) remains uncertain. The current Physical AI path based on big data and deep learning may hit a ceiling. The resolution speed of "long-tail problems" may be slower than anticipated.
2. Regulatory & Ethical Risk: Global AI regulatory frameworks are forming rapidly. Uncertainties around data privacy, algorithmic safety, and liability assignment may delay product launches and increase compliance costs.
3. Social Acceptance Risk: Large-scale machine replacement of human labor could trigger significant social employment restructuring and public backlash, requiring careful transitional policies and communication.
4. Geopolitical Risk: The security of core chip and software supply chains has become a national strategic issue. Tech embargoes and ecosystem fragmentation could disrupt global R&D collaboration and increase operational risks for companies.