Jensen Huang ushers in a new era of AI: How Physical AI is reshaping the computing landscape and the Crypto ecosystem

At the recent Davos Forum, Jensen Huang’s remarks sparked widespread industry reflection. The NVIDIA leader is not only redefining the direction of artificial intelligence development but also subtly opening a new door of possibilities for the crypto asset space. With the concept of “Physical AI,” he announced the arrival of a new era.

From Training to Inference: Jensen Huang Declares a Computing Power Revolution

Huang pointed out that AI application layers are currently experiencing a full-scale explosion, but the focus of computing power demand is undergoing a fundamental shift. The resource-intensive phase relying on hardware stacking for model training has come to an end. Future competition will center on inference and Physical AI, where AI not only “thinks” but also “does.”

This viewpoint marks a groundbreaking shift. As the absolute winner of the GPU era, NVIDIA once supported enterprises with massive computational resources to train larger models through resource stacking. But Huang’s new statement indicates that pursuing parameter scale alone is no longer the winning strategy. The future AI race will shift toward application deployment, scenario implementation, and real value creation.

The Essence of Physical AI: Bridging Virtual and Reality

Large Language Models have completed comprehensive learning from internet text data, but that’s far from enough. A trained large model still cannot, like humans, precisely open a bottle cap or understand the weight and texture of objects. This is the core problem Physical AI aims to solve—bridging the gap between AI intelligence and real-world execution.

The key limitation of Physical AI lies in its extreme requirement for real-time responsiveness. When ChatGPT experiences a one-second delay, users might just feel the interface is a bit laggy. But if a bipedal robot pauses for a second due to network latency, it could fall down the stairs. This means Physical AI must rely on local computing and edge processing, not cloud-based remote handling.

Three Major Technical Challenges of Physical AI

Huang, in discussing this new field, implicitly pointed out three technical hurdles that must be overcome. These challenges not only define the industry’s development direction but also hint at new investment opportunities.

Spatial Intelligence: Robots’ Ability to Perceive the 3D World

Stanford professor Fei-Fei Li once stated that spatial intelligence is the next Polaris for AI evolution. Robots operating in the physical world must first truly understand their environment.

This is not just recognizing objects in images—“This is a chair”—but deeply understanding “the chair’s exact position in 3D space, its structural features, and how much force I need to move it safely.” Such understanding requires building on vast, real-time, comprehensive 3D environment data covering indoor and outdoor scenes. Currently, such data is still insufficient.

Virtual Training Grounds: Robots’ Simulation School

Huang specifically mentioned Omniverse, which represents a new training paradigm. Before entering the physical world, robots need to complete thousands of trial-and-error cycles in fully virtual environments. Just as learning to walk involves “falling down ten thousand times,” robots must experience massive failures in simulation to master real-world movements. This process is called “Sim-to-Real”—from simulation to reality.

Allowing robots to learn directly in real environments would cause hardware damage with each collision or fall, leading to astronomical repair and replacement costs. Virtual training grounds offer the advantage of large-scale learning at nearly zero hardware cost. The demand for physical simulation engines and rendering computing power grows exponentially in this process.

Tactile Data: Untapped Data Goldmine

To achieve human-like “touch” in Physical AI, sensors must perceive temperature, pressure, and material textures via electronic skin. These “tactile data” are being collected at an unprecedented scale for the first time in human history.

At CES, Ensuring showcased a breakthrough: their mass-produced electronic skin integrated 1,956 densely packed sensors on a robotic hand. Thanks to these tiny sensors, the robot can precisely manipulate eggs without breaking them. This tactile data collection capability represents a new category of data assets.

Hidden Opportunities in the Crypto Arena

After understanding these technological needs, many might think only large tech companies and hardware manufacturers can participate. But in fact, the decentralized application ecosystem—through DePIN, DeAI, DeData—can precisely fill the critical gaps of the Physical AI era.

Global 3D Environment Data Collection Networks

Google’s Street View cars can scan major streets worldwide, but they cannot access every alley, neighborhood, or basement corner. These “blind spots” in 3D environment data are crucial for real-world deployment of robots.

DePIN networks, through token incentives, can mobilize millions of users worldwide to continuously collect 3D environment data of these edge areas using their smartphones and other devices. Users contributing valid data earn corresponding tokens. This approach is far cheaper than relying solely on large corporate fleets and can help close the last mile of data coverage.

Distributed Edge Computing Networks

Physical AI’s demand for real-time computation means it cannot rely solely on cloud processing. This opens a big door for distributed computing networks. Billions of consumer devices—desktops, gaming consoles, high-performance mobile devices—are often idle or underutilized.

By leveraging decentralized distributed computing, these idle hardware resources can be pooled, allocated, and scheduled to handle compute-intensive rendering tasks in Sim-to-Real training. Inference calculations for machine learning models can also be distributed to edge nodes. This not only solves latency issues but also significantly reduces the operational costs of physical AI applications.

Privacy-Protection and Data Rights Token Economy

Tactile data involves highly sensitive privacy information. If users directly contribute this data to AI giants, privacy concerns will inevitably arise. But through blockchain-based data rights and revenue-sharing mechanisms, the situation changes dramatically.

Users can retain ownership of their data, know exactly how it is used, and receive economic rewards throughout its lifecycle. This “data as an asset” model, combined with token incentives and smart contracts for automatic distribution, can motivate large-scale user participation in privacy data contribution while safeguarding data security and personal privacy.

Strategic Significance of Crypto in the Physical AI Era

Huang’s remarks actually point the entire industry toward a direction. Physical AI is not only the second half of the Web 2.0 AI track but also a rare strategic window for Web 3.0 and crypto sectors.

Decentralized infrastructure and data applications—DePIN, DeAI, DeData—are no longer just theoretical concepts but essential components of this new era. From global distributed data collection to edge computing scheduling, from secure privacy data flow to fair value distribution, decentralized solutions demonstrate their true value in this new context.

Huang’s shift in perspective effectively lights a guiding beacon for the entire crypto asset industry. Against the backdrop of Physical AI, projects focused on infrastructure, data protocols, and incentive mechanisms may truly be standing at the cusp of history.