Top 10 Core Trends of AI in 2025

In December 2025, artificial intelligence (AI) is undergoing a critical transition from the "Tool Age" to the "Partner Age." 

According to the "Top 10 AI Trends Report 2025" released by Quantum Bits Think Tank at the "MEET2026 Smart Future Conference," the core trends of current AI development can be summarized into ten main directions, covering multiple dimensions including technology, industry, society, and national strategy:

Top 10 Core Trends of AI in 2025

Infrastructure of Computing Power
Computing power becomes a strategic resource for countries and enterprises.
Super-large data centers enter the era of "computing factories."
National projects such as "East Data West Computing" and space supercomputers accelerate the construction of green, efficient integrated computing networks.
AI-ization of Chips
The dominant position of GPUs is challenged; NPUs are widely used on the edge side, with ASIC/FPGA growing rapidly.
China's "chip + SDK + framework" full-stack solutions have been verified in large model training worth hundreds of billions (such as Ascend chips + DeepSeek models).
China is accelerating the construction of an independently controllable AI computing ecosystem.
Pre-training Determines Large Model Hierarchies
Hybrid expert (MoE) architecture has become mainstream: "large parameters, small activations," balancing performance and cost.
New architectures like linear attention and sparse attention break through Transformer's O(n²) bottleneck, promoting improvements in long text and video understanding capabilities.
Entering the post-Transformer era, architectural innovations are entering a concentrated breakthrough period.
Large Models Enter the "Inference Time"
Focus shifts from training to inference: multi-modal deep inference, adaptive inference, and edge inference acceleration become key.
Inference efficiency, real-time capability, and energy consumption ratio become core metrics for model deployment.
Information AI Application Period vs Physical AI R&D Period, Embodied Intelligence Becomes a Convergence Hotspot
Information-based AI (e.g., content generation, customer service) has been commercialized on a large scale.
Physical AI (e.g., robotics, autonomous driving) enters a deep R&D phase.
Embodied AI develops rapidly relying on VLA (vision-language-action) frameworks, with China's "Great Wall" model achieving end-to-end high-precision operations.
AI Reshapes Traffic Entry Points: Rise of Agentic Internet
AI Agents shift from "people seeking services" to "services finding people."
Operating systems evolve towards "super agents," gradually replacing traditional apps.
Personalized knowledge bases (e.g., Tencent ima) make AI users' "second brain."
Multimodality Becomes Key for AI Application Deployment
Sequential release of productivity in multimodal capabilities like video, 3D, code, etc.
Cross-modal understanding and generation capabilities support complex scenarios (e.g., healthcare, industry, creative design).
Multimodal AI is transitioning from a "competitive advantage" to a "survival necessity."
AI Hardware Flourishes
Lightweight models + edge computing promote AI penetration into terminals like smartphones, cars, glasses, toys, etc.
Edge-side AI addresses three major pain points: privacy, latency, cost, realizing local real-time intelligent services.
AI for Science Accelerates AGI Realization
AI upgrades from a scientific research tool to a subject of scientific research, initiating a new paradigm of "autonomous scientific discovery."
In fields like materials, biology, healthcare, AI can:
Predict protein structures
Design new materials
Predict gene mutations from pathological images (e.g., Tencent DeepGEM model predicts lung cancer mutations within one minute)
AI's complex reasoning capabilities have reached doctoral level and are seen as a key path towards Artificial General Intelligence (AGI).
Open Source AI Enters the "Chinese Era," AGI Has a "Chinese Route"
China transitions from an AI "participant" to a "leader."
The global influence of open-source models like DeepSeek and Qwen soars, ranking among the top downloads.
National strategies drive synergy between "open source + independent chips + computing networks + industry-academia-research," crafting a Chinese route towards AGI.

AI is becoming social infrastructure.

AI in 2025 is no longer just a technology or product but a fundamental force reshaping economic structures, societal operating logic, and human lifestyles. Its core features include:
Transitioning from passive response to active service (Agent-ization),
From virtual to physical worlds (embodied intelligence, robots),
From general capabilities to personalized deep collaboration (knowledge enhancement, personal AI),
From reliance on imports to independent innovation (domestic computing power, open-source ecosystems).

These trends suggest that in 2026 and beyond, AI will become a "default existence" in human society, much like electricity and the internet.