The Montgomery Summit 2026• Mar 16, 2026• 43:07ConferencePanel

How Physical AI is Driving a New Era of Industrialization

From The Montgomery Summit · 2026

Youngjun Choi(Moderator, EY)•Ali Agha(Panelist, Field AI)•Jagdeep Singh(Panelist, Rhoda AI)•Junghee Ryu(Panelist, RLWRLD)•Amit Goel(Panelist, NVIDIA)

Executive Summary

Physical AI, defined as AI that controls objects in the physical world, is considered the next major technological wave after LLMs, with many investors believing it will become the largest technology market in history.
A critical challenge is the 'lab-to-real-world gap,' where robotic systems that perform well in controlled demos fail in dynamic, real-world environments due to data distribution shifts and long-tail events.
Unlike LLMs, which can leverage vast internet data, physical AI faces a significant data bottleneck.
Collecting sufficient, diverse, real-world data is difficult, and traditional methods like teleoperation are not scalable.
The path to success requires more than just data; it demands new model architectures that can incorporate physics, measure uncertainty, and ensure safety, as well as a robust ecosystem of hardware, software, and edge computing infrastructure.

10 quotes

Concerns Raised

The significant performance gap between lab demos and real-world deployment.
The immense difficulty and cost of acquiring sufficient, high-quality training data.
The industry may be overestimating the speed at which physical AI can be deployed at scale due to infrastructure requirements.
Current large model architectures are insufficient for robotics and need to be fundamentally re-architected to handle physics and uncertainty.

Opportunities Identified

The market for physical AI is projected to be the largest in technology history, driven by massive labor shortages.
Solving 'dull, dirty, and dangerous' jobs in sectors like manufacturing, logistics, construction, and energy.
The convergence of commoditized hardware, powerful edge compute (e.g., NVIDIA), and advances in foundation models is creating a perfect storm for innovation.
Developing the platform and ecosystem components (simulation, data pipelines, deployment UX) required to enable widespread adoption.

Key Themes

The Lab-to-Real-World Gap

A central theme is the significant performance drop when AI models for robotics move from controlled laboratory settings to unpredictable real-world deployments. This gap is attributed to the vast difference in data distributions and the inability of training data to capture the long tail of real-world scenarios.

This is the primary hurdle to scaling physical AI solutions. Companies that can effectively bridge this gap through better data, simulation, and more robust models will gain a significant competitive advantage and unlock commercial viability.

The Data Collection Conundrum

The discussion highlights that physical AI's data needs are fundamentally different and more challenging than those of LLMs. Data cannot be simply scraped from the internet; it must be collected from physical, industrial environments, which is costly and complex. Furthermore, methods like teleoperation are insufficient in quantity and diversity, and skilled workers often lack the ability to perform teleoperation effectively.

Data acquisition is a key strategic bottleneck and a source of competitive differentiation. Startups and investors must prioritize novel strategies for data collection, such as learning from video or direct observation, to build generalizable models.

Investor Excitement and Strategy

There is immense investor optimism, with top VCs viewing physical AI as the next frontier with a market potential larger than any previous technology. Investors are looking for companies with large addressable markets, differentiated technology, exceptional teams, and early customer validation. The consensus is that success will require an ecosystem approach, not a single company solving everything.

The high level of investment is accelerating innovation. However, it also creates a competitive landscape where companies must clearly articulate their unique value proposition and path to overcoming the core technical and data challenges to secure funding.

Hardware and Form Factor Pragmatism

While humanoid robots represent a potential long-term platform shift, the immediate commercial demand is for more practical, specialized form factors. Customers in manufacturing and logistics prefer solutions like wheeled mobile robots with arms that can be integrated into existing workflows without significant reconfiguration. Bipedal robots are often seen as impractical or even unsettling in current industrial settings.

This highlights a disconnect between long-term research visions and near-term market needs. Companies focused on delivering immediate ROI with practical, cost-effective hardware are more likely to gain commercial traction in the short to medium term.

Beyond Data: The Need for New Architectures

A key misconception is that more data is the sole solution to building robust physical AI. Speakers argue that the fundamental challenge lies in creating new model architectures. These architectures must be able to incorporate the laws of physics, measure uncertainty, and manage risk, as the cost of failure (or 'hallucination') is far higher in the physical world than in conversational AI.

This points to the deep technical challenges ahead. True progress will come from fundamental research in AI architectures, not just scaling existing models, emphasizing the need for deep technical expertise within founding teams.

Get started free

Topics

Physical AI Robotics Foundation Models Generalization Lab-to-Real-World Gap Data Collection Teleoperation Simulation Venture Capital Manufacturing Automation Logistics Automation Labor Shortage Humanoid Robots Edge Computing NVIDIA Model Architecture Uncertainty Quantification

Processed Apr 19, 2026 yt-dlp + mlx-whisper + Gemini