The three major breakthroughs in AI models in 2025 revealed, Jensen Huang outlines new directions at the Davos Forum

Nvidia CEO Jensen Huang summarized the three major breakthroughs in AI model layers over the past year at the Davos Forum: the maturity of agentic AI, the prosperity of open-source model ecosystems, and significant progress in physical AI. These breakthroughs mark AI’s transition from proof-of-concept to widespread application and also reflect a profound transformation across the industry.

The Three Major Breakthroughs in AI Models

Agentic AI: From Hallucinations to Reliable Reasoning

Early AI models suffered from serious “hallucination” issues, but Jensen Huang pointed out that this problem has been significantly improved over the past year. More importantly, these models can now be applied in research fields, capable of reasoning, planning, and answering questions without being specifically trained in a particular domain. This is the core capability of agentic AI—the model can autonomously complete multi-step tasks, rather than just passively answering questions.

The significance of this breakthrough is that AI has upgraded from a tool to an assistant. It no longer requires human step-by-step guidance but can understand goals, formulate plans, and execute tasks.

Open-Source Models: The Turning Point for Democratizing AI

Huang described the launch of DeepSeek’s first open-source reasoning model as a “major event.” Following this, the ecosystem of open-source reasoning models began to flourish, with many companies, research institutions, and educational organizations leveraging open-source models for innovation.

What does this mean? AI is no longer exclusive to large corporations. Small and medium-sized enterprises, startups, research institutions, and even educators now have the opportunity to build their own applications based on open-source models. This greatly lowers the barriers to AI adoption and accelerates its penetration across various industries.

Physical AI: Extending from Virtual to Real-World Applications

The third breakthrough is physical AI, a relatively unfamiliar but highly promising field. Physical AI not only understands language but also comprehends the physical world—including biological proteins, chemistry, physics, and other scientific laws. In the physical domain, AI can understand fluid dynamics, particle physics, and quantum physics.

The significance of this breakthrough is that AI’s application scope has expanded from information processing to scientific research and engineering applications. Fields such as drug discovery, materials science, and climate modeling could all benefit from new tools enabled by physical AI.

Industry Realities Behind These Breakthroughs

Related reports indicate that GPU rental prices are rising sharply with increased demand. Huang also emphasized that AI requires trillions of dollars in infrastructure support. This reflects a phenomenon: the improvement of AI model capabilities directly drives the demand for computing resources.

All three breakthroughs point in the same direction—AI is moving from the laboratory to production environments. Agentic AI enhances models’ autonomous capabilities, open-source models lower the barriers to use, and physical AI broadens application fields. These combined advances are fueling a new wave of infrastructure investment.

Summary

The three major breakthroughs summarized by Jensen Huang reflect three key directions in AI industry development. Agentic AI addresses model reliability, elevating AI from a tool to an assistant; open-source models break the monopoly, enabling more participants to innovate; and physical AI extends AI applications from information to scientific and engineering fields.

These breakthroughs are not isolated but mutually reinforcing. Stronger model capabilities require more computing resources, and the prosperity of the open-source ecosystem will further stimulate demand for computing power. It is expected that by 2026, the focus of AI industry development will revolve around these three directions, with infrastructure construction and investment continuing to be key drivers of industry growth.