CES 2026 Marks the Shift to “Physical AI”

The 2026 Consumer Electronics Show (CES) showcased the transition from digital models to “Physical AI,” an ecosystem where autonomous systems interact with the material world through sensors and reasoning. This shift positions the automotive and robotics sectors as primary drivers of a semiconductor market projected to reach US$123 billion by 2032.

The acceleration of this trend stems from the capability of generative models to perform causal reasoning within physical constraints such as gravity, friction, and mass. “Just as ChatGPT showed the world the potential of linguistic AI, the moment for physical AI has arrived. AI now acquires a body and becomes a being that coexists with us, reasoning and acting in the real world,” says Jensen Huang, CEO, NVIDIA.

Physical AI refers to the ability of autonomous systems to perceive, understand, reason, and execute complex actions in the physical environment. According to Grand View Research, the AI market in robotics will grow 38.5% toward 2030. This growth is driven by the automotive industry, which represents an opportunity of US$123 billion for chip manufacturers, an 85% increase compared to 2023.

Las Vegas 2026 CES served as a platform for corporations to demonstrate how humanoid robots, vehicles, and chipsets converge. In previous years, AI remained confined to screens. In 2026, the technology transitions to hardware that manages variables where errors are not tolerated, for instance, high-level autonomous driving and industrial manufacturing.

This evolution requires a combination of computer vision, sensors, generative models, and specialized hardware capable of responding in real time. The focus of the industry has shifted from selling graphic processing units (GPUs) to expanding dominance into new technological layers that allow machines to operate without a constant connection to the cloud.

The development of physical AI in 2026 relies on three pillars: high-performance hardware, physical simulation platforms, and the application of models in engineering and robotics.

The Convergence of Hardware: NVIDIA, AMD, and Intel

The competition between semiconductor manufacturers has redefined the role of the personal computer and industrial servers. Corporations now prioritize trillion operations per second (TOPS) as the primary metric for performance.

NVIDIA has positioned itself as a provider of a complete platform for this technology. Huang says that the company developed Cosmos, a platform designed to train, simulate, and deploy AI systems that interact with the real world. This platform supports robotic arms, autonomous vehicles, and smart factories. To achieve this, the company utilizes three integrated components:

• Cosmos: For the generation of synthetic data and training scenarios.

• Omniverse: A physical simulation environment that replicates gravity and friction.

• Isaac: A software stack that translates virtual learning into real hardware control.

AMD is also strengthening its position in high-performance computing for end users. Lisa Su, CEO, AMD, says that the new Ryzen AI processors for laptops and PCs exceed 60 TOPS. This capacity allows creators and professionals to execute complex models locally, which reduces dependence on the cloud and improves energy efficiency.

Intel debuted the Core Ultra Series 3 processors manufactured with the 18A process. This technology is part of the plan to recover leadership in advanced manufacturing. The 18A chips also target 60 TOPS and integrate neural processing units (NPUs) to execute local AI. This strategy allows Intel to defend its traditional business while demonstrating the viability of its industrial plan to governments and partners in the United States and the European Union.

The Transformation of the Automotive Industry

The automotive sector serves as the primary laboratory for physical AI. NVIDIA introduced Alpamayo, a model that uses generative technology to navigate complex traffic situations. Unlike previous systems based on fixed rules, Alpamayo performs reasoning based on sensor inputs.

During a demonstration at CES 2026, the system identified pedestrians, monitored a cyclist on a sidewalk, and waited for a vehicle to complete a turn before proceeding. This system does not rely on human-supervised training alone; it uses synthetic data. Huang says that the Cosmos platform generates images of events that did not happen in reality, such as a child jumping from a car or a truck braking suddenly in the rain. These "false memories” allow the AI to experience millions of dangerous patterns without risks in the real world.

Major manufacturers have adopted these ecosystems. Mercedes-Benz will deploy vehicles equipped with Alpamayo on public roads by the end of 2026, Geely uses NVIDIA chips for a high-level autonomous driving system, and Ford says that by 2028 it will sell a system that allows drivers to operate vehicles without looking at the road.

“The central brain of the vehicle will now be hundreds of times larger, and that is what chip manufacturers are selling. They see a great future in these cars,” says Mark Wakefield, Global Automotive Lead, AlixPartners

Robotics and Manufacturing

The integration of AI into robotic bodies is a priority for 2026. Boston Dynamics, Google DeepMind, and Hyundai announced the deployment of humanoid robots in factories, such as the robot Atlas, developed by Boston Dynamics. The corporation expects to reach a production capacity of 30,000 units per year by 2028.

These robots use the same pipeline as autonomous vehicles to learn tasks such as sorting parts, washing dishes, or plowing fields. The ability to simulate physical interactions allows these machines to learn how to handle objects with different weights and textures. 

Meanwhile, P-1 AI, which manufactures industrial cooling systems for data centers, introduced Archie. Archie uses reinforcement learning and neural networks to model design variations and simulate the behavior of physical systems in milliseconds. Traditional simulation tools take hours to test a single design, but Archie uses graph neural networks to approximate those results. The company aims to create a general-purpose AI that can design EVs or aerospace systems, says Paul Eremenko, CEO, P-1 AI. P-1 AI raised US$23 million in seed funding.

Scientific Discovery and Environmental Monitoring

Physical AI is accelerating discoveries in biotechnology and earth sciences. Lila Sciences, a biotechnology corporation, developed autonomous labs that generated hypotheses and discovered new catalysts for green hydrogen production in four months. This process traditionally requires several years of empirical research.

IBM Research has developed models such as TerraMind for satellite imagery and Prithvi WxC in collaboration with NASA. Johannes Jakubik, Research Scientist, IBM, says that these models use transformer architectures to analyze hundreds of variables in parallel. 

 “We have discovered that training models with a deep understanding of the physics behind a problem leads to more stable and reliable predictions. It is not enough to scale the models; we must make them intelligent relative to the world in which they operate,” says Jakubik.

 Future Projections and Economic Impact

The shift toward physical AI suggests that the value of hardware and software is redistributing. Systems that operate without a constant connection to the cloud provide advantages in terms of latency and security. For corporations, this means that the infrastructure for creating physical AI is already in motion.

NVIDIA aims to monopolize this process by making Alpamayo available to manufacturers. This strategy encourages the adoption of the NVIDIA ecosystem, including Omniverse and Isaac, creating a dependency similar to the position of Intel and Microsoft during the personal computer era. Huang predicts that 1 billion vehicles will eventually be autonomous, and all will function on the technical foundations laid by these platforms.