Danny Weber
04:56 24-04-2026
© D. Novikov
Elon Musk reveals TeraFab will use Intel's advanced 14A process for AI chip production, aiming for mass production by end of decade with Tesla and SpaceX collaboration.
Elon Musk has stated that his TeraFab project intends to use Intel's 14A manufacturing technology to produce AI chips. According to him, by the time TeraFab‘s own facilities come online at the end of the decade, this process will be mature enough for industrial use.
Musk noted that TeraFab sees 14A as the optimal solution and is counting on collaboration with Intel, though the details of any potential licensing deal are not yet disclosed. He also emphasized his respect for Intel's leadership and new team, hinting at a possible partnership.
During a conference call with analysts and investors, the Tesla CEO said that TeraFab plans to use the Intel 14A process, which is the most advanced but not yet fully refined. By the time TeraFab reaches its next level, the 14A process should be sufficiently developed and ready for use. 14A seems like the right choice, and TeraFab has excellent relations with Intel, respecting its CEO, CTO, and new team.
Roles within the future ecosystem are already defined: Tesla will handle the creation and operation of the pilot production line, while SpaceX will be responsible for large-scale mass production. In the near term, Tesla plans to build a research fab in Texas at a cost of around $3 billion. This facility will be capable of processing several thousand wafers per month and will serve to test new ideas and technologies under conditions that mimic real production.
The full transition to mass production is expected to require the construction of a separate large-scale facility, likely to be overseen by SpaceX. However, coordinating between the companies could prove challenging due to the need for aligning decisions and checking for conflicts of interest, which may slow the project's progress.
Licensing an existing technology is seen as a way to accelerate production rollout, since developing a proprietary process can take up to a decade. Such practices are not new in the industry: for example, GlobalFoundries previously licensed technology from Samsung, and Rapidus licensed IBM's developments.
Still, a key question remains: how feasible is it to adapt a cutting-edge process like 1.4nm for third-party fabs, given its much higher complexity compared to older-node solutions.