Intel Joins Musk's Terafab Initiative to Scale AI Compute to Terawatt Levels

Intel has officially joined the Terafab project, a venture led by Elon Musk to secure a massive, proprietary supply of semiconductors for Tesla, SpaceX, and xAI. The project's objective is to produce one terawatt of compute per year, a scale that dwarfs the current global output of leading-edge AI chips, which stands at approximately 20 gigawatts. The partnership arrives as SpaceX merges with xAI in preparation for an upcoming initial public offering. To meet the extreme compute requirements envisioned by Musk, the Terafab initiative aims to overhaul existing fabrication processes to enable unprecedented scaling of high-performance silicon. A critical component of this collaboration is Intel's recent technological breakthrough in ultrathin gallium nitride (GaN) chiplets. Intel researchers have successfully developed a method to grow GaN directly on standard 300mm wafers using existing production equipment. By employing a 'stealth dicing before grinding' (SDBG) process, Intel created a GaN chiplet with a silicon base measuring only 19 microns thick. This innovation allows for the integration of high-voltage power electronics and silicon logic on the same chiplet via a process called layer transfer. Traditionally, these components were kept separate to avoid electrical noise and heat interference, which increased system size and led to current loss. Intel's new architecture places power and logic transistors in a more compact space, enhancing efficiency. For Intel, the partnership validates its foundry services and advanced packaging capabilities. For Musk's ventures, the technology provides a potential pathway to the compute density required for next-generation artificial intelligence and aerospace systems.