Looking ahead: Elon Musk's artificial intelligence startup, xAI, is ramping up its efforts to develop a new generation of AI systems known as "world models" – technologies designed to interpret and construct representations of the physical world. The initiative places xAI alongside major competitors such as Google DeepMind and Meta in the emerging race to build AI that can reason about real-world environments, not just text or images.
People familiar with the company's plans told the Financial Times that xAI recently hired Nvidia researchers Zeeshan Patel and Ethan He, both of whom have experience developing world models and simulation-based AI.
Nvidia, one of the largest suppliers of computing hardware for AI research, has invested heavily in its Omniverse platform, a suite of tools for building and operating industrial-scale simulations. Some of that expertise is now shifting, in part, to Musk's venture.
World models represent one of the most ambitious frontiers in artificial intelligence research. Unlike large language models that generate text based on statistical patterns in data, world models are designed to learn how physical systems behave. They train on vast collections of video, sensor, and robotic data to build a causal understanding of objects, motion, and environmental dynamics.
By modeling the physics of real or simulated environments, these systems could enable AI to act more intelligently and safely in the physical world, an essential step toward autonomous machines and advanced robotics.
xAI's push into world models is expected to extend beyond research and into commercial applications. Two people familiar with the project said the company sees potential for the technology in video game development, using it to automatically generate interactive 3D environments.
One of the sources added that xAI also views the models as essential for future robotic systems capable of navigating complex spaces. Such capabilities could move AI beyond purely software-based tasks, establishing a foundation for physically embodied intelligence.
Musk has outlined some of xAI's near-term goals, including plans to release what he described as a "great AI-generated game" before the end of next year. The project would align with the company's broader effort to expand its generative tools beyond text and images.
Last week, xAI unveiled its latest image and video generation model, describing it as a major technical upgrade and making it freely available to the public. While video creation platforms such as OpenAI's Sora produce image sequences through statistical prediction, xAI and other researchers are attempting to train AI systems that understand cause and effect, or how one action leads to another in real time.
The company has also begun advertising new technical roles in image and video generation for what it calls its "omni team," which describes its mission as creating "AI experiences beyond text" across visual and audio formats. Salaries for these positions range from $180,000 to $440,000, according to job postings. xAI is additionally recruiting a "video games tutor" to train its chatbot, Grok, in game design, offering hourly rates between $45 and $100.
Genuinely what this industry needs is not more mathematically produced, psychologically trained gameplay loops, rather more expressions of worlds that folks are engaged with, or want to engage with. AI has its place as a tool, but we have all the tools in the world and they... https://t.co/eL98XeLGW8
– Very AFK (@Cromwelp) October 6, 2025
The broader promise of world models has drawn increasing attention across the AI industry. Nvidia told the Financial Times last month that the eventual market for these systems could approach the scale of the global economy, given their potential applications in manufacturing, robotics, and simulation-based training. Still, developing them remains a daunting technical challenge. Building accurate representations of the real world requires vast amounts of multimodal data and immense computing power, both of which are highly resource-intensive hurdles.
Skepticism persists among segments of the gaming community, where world models are seen as unlikely to replace human creativity. Michael Douse, head of publishing at Baldur's Gate 3 developer Larian Studios, wrote on X that AI tools will not fix what he called the industry's "big problem," pointing instead to a lack of "leadership [and] vision." He argued that the future of games lies not in "mathematically produced, psychologically trained gameplay loops" but in artistic engagement and world-building.