In January 1950 the magazine Time dedicated the cover to Mark III, a US Navy calculator, accompanying it with a question: “Can man build a superman?”. At the time, the idea of a computer capable of surpassing the intellectual capabilities of a human being belonged more to science fiction than to reality. Not anymore today, so much so that its implications and applications dominated the media discussion throughout 2025, a year that ended with another cover of Timewhich chose “architects of artificial intelligence” as its people of the year.
It is likely that if in a few decades we have to indicate the year in which AI software definitively entered people’s lives, we will choose 2025. It was the year in which the photos and videos created by AI began to really seem real, and in which even those professional and productive sectors that had not done so until now had to deal with the changes brought about by these technologies. But 2025 was also the year in which it became normal for many people to turn to AI chatbots for daily, practical but also psychological issues.
In December 2024, according to OpenAI, ChatGPT had 300 million weekly users: in November 2025 it had almost triple that, 810 million. If in recent years AI software had already replaced the previous work tools for many professionals, it is in the last twelve months that ChatGPT, Gemini and similar chatbots have become a mass tool, like browsers for surfing the internet at the end of the nineties.
More and more people have started to replace search engines like Google with chatbots: despite the risk of errors and hallucinations, in fact, the possibility of receiving complete answers represents for some a better experience than the usual online search. AIs are particularly widespread among younger people because they are widely used at school, both for doing homework (or having the AI do it) and as study support. According to recent research, one third of US teenagers interact with a chatbot daily.
Instead, programs such as the Microsoft 365 package, which the company is counting on to promote its Copilot chatbot, contribute to spreading AI among adults, especially at work. These technologies have already had an impact on the employment front, especially in sectors such as computer programming: according to the site TechCrunchIn fact, in 2025 alone, US technology companies fired around 150 thousand employees, often justifying themselves with the optimization made possible by AI.
– Read also: AIs have learned to generate images that look like bad photos
2025 was also the year of Google’s comeback, which after having lagged behind for a long time in recent months made flagship products such as Nano Banana Pro (a very realistic image and video generator) and the Gemini 3 language model, considered the most advanced in the world, available online. A similar comeback has not been achieved by Apple, which still does not have a clear strategy for AI. Also for this reason, during 2025, the company lost managers and qualified personnel, especially due to Meta, which last summer spent tens of billions of dollars to acquire the Scale AI startup and hire engineers from the competition.
But the massive investments involved dozens of companies, so much so that throughout the year the discussion on new models and their applications was accompanied by the increasingly widespread fear of a bubble, caused by excessive investments in data centers, which represented 40 percent of the economic growth of the United States in 2025. As the Financial Times«America is now making a big bet on AI», fueled by competition at home and with China.
Anxiety about China’s technological progress has been a constant throughout the year. In fact, 2025 began with a traumatic event for the entire technology sector, when DeepSeek, a Chinese company unknown to most at the time, presented AI models comparable to Western ones but less expensive to develop, causing losses of one trillion dollars on the US stock market.
Despite everything, the emergency passed within a few days and the sector continues to be dominated by US companies and startups, such as OpenAI, Google and Anthropic. To these must also be added Meta and services such as Cursor, designed for the generation of computer code. However, the Chinese advance in the sector has not stopped with DeepSeek, as demonstrated by the models of Z.AI, Xiaomi and Kimi, some of the most powerful models in the world, while the only European company in the AI field remains Mistral (albeit with minimal market percentages).
In all of this, ChatGPT still remains the most used chatbot in the world today, even if the alternatives are increasingly numerous and are used in the most disparate ways, not just for work or study. 2025 was the year that stories of chatbot relationships gone very badly became a literary genre in newspapers. In one of the most relaunched, last August, the New York Times had reported the case of a sixteen-year-old who committed suicide in the United States following a series of mental health problems. During the last months of his life, the boy had begun to chat more and more with ChatGPT-4o, an OpenAI model, which he initially used to do his homework and had subsequently collected his confidences without alerting anyone of his declared intentions.
ChatGPT-4o was one of the clearest examples of a well-established trend in the industry: chatbots designed to please users in almost every circumstance, to the point that OpenAI chief Sam Altman called them “lickers.” To solve this problem, OpenAI equipped its new model, GPT-5, released in August, with a drier and less accommodating tone, to protect the mental health of the most fragile users and avoid the so-called “AI psychosis”. In the following months, however, the company backtracked, so much so that Altman announced that starting from 2026, adult users will be able to discuss erotic and sensitive topics with the chatbot.
– Read also: What happened to the caution in the artificial intelligence sector?
Even sectors that in the past had had more wary and combative attitudes towards AI, over the course of the year, decided to adapt: in particular there was a radical change in approach towards the sector by the cultural industry. In previous years, in fact, OpenAI had been sued for copyright infringement by authors and publishers such as New York Timeswhile the startups Udio and Suno, which generate music from textual input, had been sued by many record companies, including Universal, Sony and Warner.
In recent months the scenario has changed. Suno and Udio are now used by some professional musicians and songs generated with AI have entered some charts, such as country in the United States. But above all, Universal and Suno have closed the dispute with a licensing agreement that allows Suno to use the Universal catalog by paying the necessary royalties, and Warner and Udio have also signed a similar agreement.
The same happened with videos. One of the most discussed products of the year was Sora 2, OpenAI’s new model for generating videos, supported by Sora, a social network similar to TikTok dedicated to generated content. The year closed with a billion-dollar investment by Disney in OpenAI, which will make its characters and stories available to Sora users to generate videos.
Meta and Google have also presented products similar to Sora, which are already having major repercussions on social networks, increasingly crowded with junk generated with AI. The pervasiveness of this content on YouTube, Instagram, Facebook and TikTok has popularized the derogatory expression slop (“slime”), chosen by the Merriam-Webster dictionary as the word of the year for 2025.
These cheap applications of AI by the public were not part of the original promise of the sector, which had instead ensured the advent of an «AGI» (Artificial General Intelligence), a supposedly higher level of AI capable of surpassing human AI and changing society forever, of which however there is still no trace. Another promise of the sector were the “agents”, AI programs capable of carrying out online actions and searches autonomously on behalf of the user, which would revolutionize the way we use the web and our devices. The agents were the basis of new browsers designed for AI – such as ChatGPT Atlas, introduced by OpenAI this year – which so far have not lived up to the hype.
Net of the enormous hype It therefore remains difficult to take stock of the sector: AI users are growing but doubts remain about the sustainability of investments and the lack of a business model for these companies. At the center of these fears is the company that has gained more than any other from the AI race, Nvidia, which has a market share of more than 90 percent in GPUs, the powerful chips fundamental to the development of these technologies. Over the course of the year, Nvidia has invested in many companies in the sector (such as OpenAI and CoreWeave, which deals with data centers), which in turn have Nvidia as their main supplier: it is also this type of “circular” agreements between the same companies that worries many financial analysts.
– Read also: When does an AI become too powerful?
Despite everything, however, Nvidia ended the year with a huge concession from Trump, which allowed the company to return to selling its most advanced GPUs in China, where they had long been banned in an attempt to hinder the progress of the local AI sector. Trump’s decision was criticized by many: the Wall Street Journal he compared GPUs to early supercomputers (like the Mark III itself), recalling that during the Cold War no US company could have sold them to the Soviet Union.
