Last Saturday, Elon Musk took the stage to announce his plan to enter semiconductor manufacturing, using highly exaggerated language. "I have a major announcement; this will be the most ambitious chip manufacturing project in history to date," he told a small audience in Austin, Texas. "This will truly push the industry to a new level, one that people haven't even conceived of yet. We are going to elevate the current landscape by several orders of magnitude." No one has ever proposed a plan on the scale of what Musk calls the "Terafab." As described, the project would be a massive factory dedicated to producing cutting-edge semiconductors for artificial intelligence, robotics, and space exploration. He aims not only to challenge the world's leading chipmaker, Taiwan Semiconductor Manufacturing, but also to produce at a scale far exceeding the industry's current capacity. The proposed scale is staggering. Analysts at Bernstein estimate the project would require capital expenditure of $5 trillion to $13 trillion, enabling the construction of 140 to 360 new factories, each with a monthly output of 50,000 wafers, to achieve Musk's stated goal of one terawatt of computing power annually. As the world's richest person, Musk has repeatedly accomplished what others deemed impossible: building a commercially viable rocket business with SpaceX, mainstreaming electric vehicles with Tesla Motors, and creating a space-based internet with Starlink. However, many doubt whether Musk can, or even truly intends to, build the factory he outlined in Austin. Analysts at Bernstein, including Stacy Rasgon, wrote, "In our view, actually building the Terafab is almost impractical." The required scale of computing power "is equivalent to the total current global installed semiconductor capacity, and producing 'relevant' AI chips would require several times the existing capacity." Patrick Moorhead of Moor Insights & Strategy added that Musk will most likely not build a chip fab at all. Instead, the true purpose of Musk's Terafab concept may be different: to highlight the increasingly tight supply of chip capacity or to pressure chip manufacturers to accelerate their expansion efforts; it could also be an attempt to boost market expectations for SpaceX, which is planning an IPO later this year. There is growing concern in Silicon Valley that the semiconductor industry's expansion pace cannot keep up with the chip supply needed for AI companies to realize their ambitious business plans. Hyperscale cloud providers like Amazon and Alphabet are projected to invest approximately $650 billion in data center infrastructure this year alone. This has already caused a severe shortage of memory chips, which is beginning to spread to AI accelerator chips.
Comments