Musk’s xAI taps Nvidia talent to build ‘world models’ AI system

xAI, the AI startup founded by Elon Musk, is developing advanced AI systems known as “world models,” the Financial Times reported.

These models are designed to help AI navigate and design physical environments, aiming to move beyond current large language models like Grok and ChatGPT.

To support this effort, xAI hired Zeeshan Patel and Ethan He, both AI researchers from Nvidia with experience in training AI using video and robotics data.

Nvidia has been active in this area with its Omniverse simulation platform.

xAI is also recruiting technical staff for its “omni team,” which works on AI capable of understanding and generating content in formats such as photos, videos, and audio.

The company has announced plans to release an AI-generated game before the end of next year.

🔗 Source: The Economic Times

🧠 Food for thought

Implications, context, and why it matters.

xAI lacks compute to match world model leaders

xAI hired Nvidia Omniverse veterans Zeeshan Patel and Ethan He 1. It ended talks with Oracle (a cloud infrastructure provider) on a $10 billion server deal over timing and power supply 1.
The company is building its own system with Nvidia H100 GPUs (high-end AI accelerator chips) 1. OpenAI says over one million GPUs will be online by year-end, and CEO Sam Altman joked about 100x that 2.
Real-time world models need compute (raw computational capacity) for video and robotics training. Without capacity commitments that, per SemiAnalysis, OpenAI and ByteDance (the owner of TikTok) have with Oracle 3, xAI will wait longer to match them.

Robotics simulators should partner with AI labs on physical products

NVIDIA Isaac Sim (a robotics simulation platform) supports generating synthetic data (computer-generated training data) for perception and manipulation models 4. FANUC offers ROBOGUIDE (its offline programming and simulation software) for virtual workcells (production-line robot setups) 5.
RoboDK (robot programming plus simulation software) works with more than 80 robot makers and integrates with CAD/CAM tools (Computer-Aided Design/Computer-Aided Manufacturing) 6, which could make it base infrastructure for AI groups training world models on robotic data.
Companies building 3D content pipelines plus synthetic data generation should pitch themselves to AI labs 3. Oracle brings a leasing strategy plus a ClusterMAX Gold rating (an industry benchmarking designation for large AI compute clusters) along with direct ODM relationships (Original Design Manufacturers that design and build servers to a buyer’s specifications) that let providers capture value in the AI supply chain before hardware deployment scales 3.