Artificial intelligence generates mixed opinions: some see it as an unstoppable revolution, while others consider it a bubble about to burst. Ben Thompson, a well-known analyst, used to be among the skeptics, defending the idea of a "good bubble" that, even if it burst, would leave positive advancements. However, after the latest annual NVIDIA conference, Thompson has changed his stance: for him, AI is not a bubble. His reasoning is based on three major milestones. The first milestone was the launch of ChatGPT in November 2022, which showcased the potential of generative AI. Despite its impact, the model had two significant limitations: it frequently made mistakes and, when it didn't know an answer, it confidently invented one. This made it fascinating but unreliable, more of an experiment than a professional tool. The second leap came almost two years later, in September 2024, when OpenAI introduced its model o1. This model introduced unprecedented reasoning capabilities: before responding, it evaluated the validity of its answers and considered alternatives. The result was a much more accurate and useful AI, albeit at a cost: it required significantly more computing power, which drove up the demand for data centers. The third advancement is that of AI agents. By the end of 2025, tools like Claude Code and Codex demonstrated that agents could autonomously execute complex tasks, correcting their own mistakes without human intervention. This marked a radical difference from previous models and, according to Thompson, dismantles the idea of a bubble. Thompson argues that, in a bubble, investment exceeds real demand. In the case of AI, the opposite is true: the demand for computing exceeds supply, and giants like Microsoft, Google, Amazon, and Meta are investing record sums in infrastructure to meet that need. These investments are not speculative but a response to a demand that continues to grow. Another key point is that, unlike chatbots, AI agents do not require mass adoption to have economic impact. One person can manage thousands of agents, multiplying productivity without the need for millions of users. This opens the door to "one-person businesses" where an individual can coordinate the work of thousands of AI agents. Regarding the business model, Thompson notes that consumers are not willing to pay for AI, but companies are, as they seek productivity and efficiency. Agents allow small teams to achieve what previously required large workforces and complex management structures, eliminating hierarchical layers and reducing costs. However, Thompson warns that automation will lead to layoffs, although many of the current cuts are due to overhiring during the pandemic. The real challenge will be adapting to a "post-AI" world, where companies that do not optimize their structures will compete at a disadvantage against new, more agile, and efficient players thanks to AI. In conclusion, for Thompson, the demand for computing will continue to grow, and the supposed AI bubble, if it exists, is not close to bursting. Source: xataka.com