show index hide index
- The debate about a potential AI bubble is intensifying. Despite the widespread adoption of AI, its integration remains a challenge for many companies, with only 4% of professions using AI for the majority of their tasks. Studies highlight a decline in employment among young workers in sectors exposed to AI. Economic indicators such as inflation or the Fed’s decision in 2026 could influence this context. The potential bursting of an AI bubble is a crucial question, both for industry and for the global economy.
By 2025, the rapid evolution of Artificial Intelligence (AI) has transformed industrial and economic dynamics worldwide. Emerging trends are redefining how we interact with intelligent technologies, revealing the immense potential of reasoning models, the growing role of Chinese open-source models, American supremacy in computing power, and ongoing debates about the existence of a potential AI bubble. These developments have shaped a diverse technological landscape, stimulating innovation while raising new challenges for the future.By 2025, artificial intelligence is undergoing significant transformations, marked by advances in reasoning models, increased adoption of Chinese models, massive investments consolidating American computing power, and questions about a potential AI bubble. These trends are defining the AI landscape and shaping future innovations. This article explores these four horsemen of the future of AI. Model Evolution: The Rise of AI’s Agentic Capabilities In 2025, artificial intelligence reached a significant milestone with the rise of reasoning models. These models, going beyond traditional autoregressive predictors, leverage computing power to refine responses and explore various reasoning paths. This technological advancement enables the emergence of advanced agentic applications, ranging from deep research to autonomous code development.The increasing power of these models led to a rise in the share of reasoning tokens to 60% by the end of 2025. In benchmarks such as ARC-AGI-1, the GPT-5.2 model dramatically outperformed its predecessor. These developments support a more proactive AI, capable of interacting more effectively with humans. Adoption and Usage: The Rise of Chinese ModelsIn parallel, 2025 is marked by a significant surge in Chinese AI models Chinese labs such as DeepSeek and Alibaba with Qwen are establishing themselves on the global stage. At Airbnb, for example, Qwen usage surpasses that of ChatGPT, illustrating this trend.The share of tokens generated by Asian companies has increased considerably, driven by the adoption of their models by businesses. Faced with American controls, China is pursuing its technological independence, relying on domestic semiconductors, such as those produced by Huawei.
Investments: American Supremacy in Computing Power TeaUnited States maintains its dominance with 75% of the world’s computing capacity dedicated to AI. Investments are massively concentrated in the United States, representing 82% of global AI capital. This American dominance is also reflected in the colossal energy consumption of supercomputers. The US economy is becoming inextricably linked to advances in AI, with a substantial portion of its economic growth attributed to investments in data centers and AI-dedicated software, now representing 4% of GDP. Is there an AI bubble?
The debate about a potential AI bubble is intensifying. Despite the widespread adoption of AI, its integration remains a challenge for many companies, with only 4% of professions using AI for the majority of their tasks. Studies highlight a decline in employment among young workers in sectors exposed to AI. Economic indicators such as inflation or the Fed’s decision in 2026 could influence this context. The potential bursting of an AI bubble is a crucial question, both for industry and for the global economy.
To read Claude et le chantage : Anthropic lève le voile sur une révélation surprenante
