By 2025, the novelty of generative AI had worn thin. Dictionaries crowned 'slop'—low-quality, mass-produced AI content—as word of the year, and talk of an AI bubble bursting grew louder. Yet tech giants pressed on: Google's Gemini 3 launch reportedly triggered a 'code red' at OpenAI to accelerate GPT-5 development. But with AI leaders warning of 'peak data,' 2026 may see a shift toward new paradigms.
The Rise of World Models
World models represent a departure from text-based chatbots. Instead of predicting the next word, they learn from videos, simulations, and spatial data to build internal representations of scenes and objects. Think of them as digital twins that model cause and effect—understanding gravity, motion, and time without explicit programming. This makes them crucial for robotics, video games, and autonomous systems.
Companies are already investing. Google and Meta have announced world models for robotics and video realism. Yann LeCun, a so-called godfather of AI, left Meta in 2025 to launch his own world model startup. Fei-Fei Li's World Labs released its first product, Marble, the same year. Chinese firms like Tencent are also developing their own versions. Robert Playter, CEO of Boston Dynamics, told Euronews Next in November that AI has been essential for his robots' progress, though 'there's still a huge amount of work to be done.'
Europe's Pivot to Small Language Models
While US and Chinese companies chase massive world models, Europe is taking a different tack: small language models (SLMs). These lightweight versions of large language models (LLMs) run on smartphones and low-power devices, using fewer parameters and less energy while retaining strong text generation, summarization, and translation capabilities. They are also more economical amid fears of an AI bubble.
Max von Thun, director of Europe and transatlantic partnerships at the Open Markets Institute, told Euronews Next that 'doubts about the financial sustainability and socioeconomic benefits of today's large-scale AI boom will continue to grow' in 2026. He noted that European governments are becoming 'increasingly wary' of relying on American AI and cloud infrastructure, given the US government's 'clear intentions to weaponise technological dependencies for political gain.' This could accelerate European efforts to build local capabilities, focusing on smaller, sustainable models trained on high-quality industrial and public data.
The push for European tech autonomy is echoed in recent developments: US blocks on Anthropic's AI models for non-citizens have spurred calls for self-reliance, while EU warnings against discriminatory export controls highlight the geopolitical stakes.
Safety and Sustainability Concerns
Beyond hype, 2025 brought alarming incidents of AI psychosis—users forming delusions or obsessive attachments to chatbots. A lawsuit against OpenAI alleged that ChatGPT acted as a 'suicide coach' for a 16-year-old, though OpenAI denied the claims. Max Tegmark, MIT professor and president of the Future of Life Institute, warned that as models become more powerful, such harms could increase. Engineers may not have intended harm, he told Euronews Next, but the risk grows with more sophisticated software.
The environmental cost of giant data centers also looms. US firms like OpenAI, xAI, Meta, and Google pour billions into infrastructure, raising questions about sustainability. In contrast, Europe's focus on smaller models could offer a greener path. For instance, Rolls-Royce's plan to build small nuclear reactors in Sweden hints at future energy solutions for AI.
As 2026 unfolds, the AI landscape will likely diversify: world models for physical reasoning, SLMs for efficiency, and a growing European push for sovereignty. The bubble may not burst, but the direction is clear—away from slop and toward substance.
