News at a Glance
NVIDIA’s official blog wrote on September 21 that physical AI is moving rapidly from research to large-scale deployment, projecting that the global fleet of L3-L5 autonomous vehicles will reach 49 million units by 2035, with roughly 60 million industrial robots deployed between 2026 and 2035. As these machines enter shared spaces such as roads, factories, and warehouses, safety must run through every layer of the system rather than be applied as an afterthought. This view reflects the industry’s widespread concern about accident risks after physical AI scales, and also signals the importance of safety design.
Background
Physical AI is a technical field that integrates artificial intelligence with physical systems. In recent years, it has matured faster thanks to breakthroughs in large models and robotics, with applications covering autonomous driving, industrial manufacturing, logistics, and warehousing. However, large-scale deployment means machines will leave controlled environments and become directly intertwined with everyday human activity. Previous safety discussions have mostly focused on the functional safety of a single vehicle or machine, while a systematic safety framework spanning devices and scenarios has yet to be established. NVIDIA’s emphasis on “safety at every layer” at this time reflects both the central role of computing platforms in physical AI and the fact that safety is becoming a prerequisite and competitive focus for AI deployment.
In-Depth Interpretation
Liu Gong believes that the safety challenges of physical AI will not be solved by any single “killer safety solution” but must be embedded in every layer from chip to cloud, from perception to execution. This is similar to the early days of cloud computing, when only application-layer vulnerabilities were defended and data breaches were frequent; if physical AI is discussed only at the algorithmic level, then when tens of millions of machines coexist with humans, a single small vulnerability could turn into a public incident. In the future, safety will no longer merely be a compliance cost, but a hard threshold for companies seeking market entry, and may even reshape the balance of power in the industry chain. What deserves attention next is whether unified cross-industry safety standards or third-party evaluation bodies will emerge, and how platform vendors such as NVIDIA turn safety capabilities into ecosystem stickiness. Whoever builds trust at the safety layer is likely to take a half-step lead in the physical AI race.
Perspectives
Further Thoughts
- Physical AI safety is shifting from algorithm-level functional safety to end-to-end system safety covering perception, decision-making, execution, and communication.
- The lack of unified cross-industry safety standards and evaluation systems could become the biggest obstacle to large-scale deployment of autonomous driving and robotics.
- By emphasizing full-layer safety as a platform player, NVIDIA may be seeking to define the safety ecosystem in the physical AI era.
Source and Original Article
This news item is sourced from NVIDIA Blog (published September 21, 2026, 16:00:46). This site provides Chinese summaries and commentary on overseas AI developments; the original article is copyrighted by its author.
Daily aggregation of first-hand overseas AI developments and in-depth perspectives. Bookmark this site so you don’t miss any important signal; return to the homepage for more.