Ming-Chi Kuo: MediaTek Expected to be a Key Collaborator in Musk's Terafab Project

Renowned Apple industry chain analyst Ming-Chi Kuo's latest research indicates that among several custom ASIC manufacturers, MediaTek is most likely to become a strategic partner for Elon Musk's Terafab project. MediaTek will assist in introducing Intel's 14A advanced process technology and advanced packaging, and is expected to start small-scale chip production for Musk's chip design team in 2028.

If this cooperation materializes, it would mean a strategic upgrade for MediaTek's AI ASIC business; for Terafab, it would be a key step in enhancing execution amidst a tight window of time.

Terafab Faces Triple Execution Pressure

Kuo's analysis suggests that while Musk's semiconductor vision is grand, his integrated circuit design team is facing severe challenges in terms of scope, time, and manpower.

Regarding scope, Terafab must simultaneously advance cooperation with TSMC, Samsung, and Intel on three advanced process technologies, covering both terrestrial inference and space-specific chip product lines, with over six parallel projects and a design cycle target of just 9 months, significantly faster than the industry standard of 18 to 24 months.

The pressure of time is equally pressing. Intel's 14A PDK 0.9 is expected to be open to external customers only by October 2026, and if Terafab fails to get involved at the first opportunity, they might miss the 2028 small-scale production window. Kuo also pointed out that Terafab has offered procurement prices significantly higher than market rates to equipment manufacturers, indirectly confirming the urgency of their time constraints.

On the human resource front, compared to the size of Apple's silicon engineering team, the combined workforce of SpaceX and Tesla's chip teams is only a fraction, or even a tenth, of the former, yet they need to execute broader tasks within a shorter timeframe.

Why MediaTek is the Best Choice

Kuo believes MediaTek has unique advantages in three dimensions. Firstly, MediaTek has actual experience with Intel's front-end process technology and advanced packaging EMIB-T, which can help Terafab quickly enter the design and manufacturing process after the 14A PDK release.

Secondly, MediaTek's cooperation with Google on TPU has exceeded expectations. According to Kuo's revelations, the TPU 8t involving MediaTek will go into mass production in the fourth quarter of 2026, and the Humufish project will go into mass production in the second half of 2027, demonstrating the Semi-COT collaboration model and large-scale mass production capability, which is exactly what Terafab urgently needs right now.

Moreover, MediaTek is already a Wi-Fi chip supplier for SpaceX Starlink's user terminals, and there is an existing commercial trust between the two parties, which is conducive to accelerating cooperation under high-pressure timelines.

Kuo also mentions an easily overlooked factor; MediaTek's value lies not only in the company's capabilities but also in the accelerated R&D model of the Taiwanese semiconductor ecosystem behind it. He cites TSMC's "Night Owl Project" which operates a round-the-clock R&D mechanism as an example, indicating that if Terafab collaborates with MediaTek, it could achieve transcontinental R&D relay between the US and Taiwan, further enhancing execution efficiency.