Elon Musk revealed this week that his Terafab project will use Intel's next-generation 14A manufacturing process to produce AI chips, deepening a partnership that already represents one of the most ambitious semiconductor ventures in American history. The $25 billion joint venture between Tesla, SpaceX, and xAI aims to produce one terawatt of AI compute annually, and Intel's role as the primary foundry partner puts the chipmaker at the center of Musk's hardware ambitions.
From 18A to 14A
Intel initially signed on to the Terafab project in early April as the primary foundry partner, contributing its 18A process node, the company's most advanced logic manufacturing technology currently ramping to high-volume production in Arizona and Oregon. The revelation that Terafab will also use the even more advanced 14A process signals an escalation in the project's technical ambitions.
"We are building the most advanced chip manufacturing facility the world has ever seen," Musk said in a post on X. "Intel's 14A process will give us the edge density we need for next-generation AI inference and training workloads."
The 14A process node represents Intel's push into the angstrom era of semiconductor manufacturing, with transistor features measured in fractions of a nanometer. While details of the 14A technology remain limited, Intel has indicated it will deliver significant improvements in power efficiency and transistor density over 18A, which is itself competitive with TSMC's most advanced nodes.
Two Categories of Chips
Terafab is designed to produce two distinct categories of AI chips. The first is energy-efficient edge-inference processors for Tesla's Full Self-Driving systems, the Cybercab robotaxi fleet, and the Optimus humanoid robot program. The second category encompasses high-power, radiation-hardened chips for SpaceX satellites, xAI orbital data centers, and the Starshield defense constellation.
"The vertical integration story here is unprecedented," said Patrick Moorhead, CEO of Moor Insights and Strategy. "No other company in the world is attempting to design and manufacture AI chips for applications ranging from self-driving cars to orbital computing platforms."
SpaceX will be responsible for high-volume chip manufacturing operations, leveraging its experience with precision engineering and manufacturing at scale. The arrangement is likely structured as an Intel technology licensing deal, with SpaceX operating the fabrication lines using Intel's process recipes.
Why This Matters
The Terafab-Intel partnership represents a significant departure from the prevailing model of AI chip development, where companies like Nvidia design chips and TSMC manufactures them. By building vertically integrated chip production capabilities, Musk's ventures are attempting to control the entire AI hardware stack from design through fabrication.
For Intel, the partnership provides a high-profile customer for its most advanced process nodes at a time when the company is working to establish itself as a credible alternative to TSMC. The deal validates Intel's foundry strategy and could attract additional customers who have been hesitant to commit to Intel's manufacturing services.
The geopolitical implications are also significant. By building advanced chip manufacturing capacity in the United States, the Terafab project aligns with the CHIPS Act's goal of reducing American dependence on Asian semiconductor supply chains.
What to Watch
The timeline for 14A production remains unclear, with Intel not yet providing a firm date for the process node's readiness. Investors and industry watchers will be monitoring whether the Terafab project can hit its ambitious compute targets and whether the dual-chip strategy can deliver on the diverse performance requirements of its intended applications.
“The vertical integration story here is unprecedented. No other company in the world is attempting to design and manufacture AI chips for applications ranging from self-driving cars to orbital computing platforms.”— Patrick Moorhead, CEO, Moor Insights and Strategy