The entrepreneur has already achieved brilliant success in difficult industries (automotive and space) where he had no experience. But his latest adventure, the Terafab project, could well be the most difficult of all.
Etching transistors on the scale of a few atoms, maintaining precision of the order of nanometers on entire surfaces, all in clean rooms where the air must be maintained at a level 100,000 times purer than in a surgical operating room, a single grain of dust on a 30,000 euro wafer being able to destroy dozens of chips. And all while managing complex and constantly tense supply chains, involving machines produced by one or two companies maximum around the world and raw materials concentrated in a few specific areas.
These are all reasons why cutting-edge semiconductors, used to drive and operate AI, are considered the hardest thing to produce on earth. A process that only a handful of companies in the world have mastered, led by the Taiwanese TSMC, the undisputed market leader. A domination that Elon Musk now intends to challenge with his new Terafab project. The objective: to free itself from TSMC to produce itself the chips equipping its Teslas, its Optimus robots and its data centers specializing in AI (some of which will be positioned in space).
Reasonable bet or egomaniacal delusion?
Musk’s announcement was greeted with disbelief and skepticism by some industry experts, such as Mike Demler, an American specialist in the semiconductor market. “First, I don’t think $20 billion is enough to build a state-of-the-art factory,” he begins. “Then you only have to look at Intel to see how financially and technically impractical it has become to be an integrated semiconductor manufacturer (which controls each step, from design to production, as opposed to Nvidia’s fabless model, for example, editor’s note). Other customers are needed to justify the expense, and to provide a variety of designs that feed data into improving manufacturing performance. Then, you have to work with the ecosystem of intellectual property providers and electronic CAD tool publishers to create process design kits, develop and test the design flow….” For the expert, “Musk’s plan to build a ‘Terafab’ is just one more example of his seemingly limitless arrogance.”
Others are less categorical, like Antoine Chkaiban, consultant at New Street Research. “Elon Musk is accustomed to grandiose announcements which take time to materialize, or which do not materialize, like his $25,000 vehicles, or even these battery factories which were to supply one hundred gigawatt hours of 4680 cells by 2022. And semiconductors are much more complex to produce than batteries…”, he recalls. However, “there are many opportunities for optimizing semiconductor production. Musk notably spoke of further integrating the production process, with wafers which would be entirely produced on the same equipment, which would make it possible to do without clean rooms. However, integration is precisely what will make it possible to increase productivity in this market.”
The fact that he is starting from scratch could also work to his advantage. “He is an engineer at heart, capable of taking a problem with the clean sheet approach, who is not bothered by the dilemma of the established player, already has production lines in place, etc. He can therefore hope to make more efficient, cheaper chips, with shorter innovation cycles and a better cost structure.”
The crux of the matter: men… or money?
If Elon Musk has achieved several feats in the past, the comparison with TSMC gives an idea of the scale of the Herculean task he has set for himself. “Two-nanometer manufacturing is arguably the most complex manufacturing process humanity has ever attempted on a large scale. TSMC has invested $165 billion and four decades of industrial know-how to achieve it. Terafab’s timetable involves doing the equivalent in a few years, with a team starting from scratch,” says David Cao, an American investor and entrepreneur.
This talent issue is particularly thorny in a semiconductor industry already facing a shortage of human capital. The United States lacks engineers, which is already slowing down the country’s reindustrialization, and semiconductor specialists are in particular demand. When setting up its first factories in the United States under the Chips Act, TSMC had to import its own engineers from Taiwan to fill the gap in local skilled labor. David Cao estimates that Musk will need at least 6,000 engineers specializing in semiconductors to launch his project: finding them will be a challenge, especially since he will be in competition with Samsung, which is also setting up in Texas to produce chips there.
There is of course also the question of financing. “Tesla’s 2026 investment plan does not fully account for the Terafab project, fueling market speculation that a large and dilutive capital raise is on the horizon. How the financing is structured, and whether investors view it as a Tesla capital expenditure or a separate entity, will determine how Wall Street values its companies,” said David Cao. Musk will finally also have to source machines from ASML, whose order book is already full.
Elon Musk’s dream
The entrepreneur said he wanted to ensure that he had more chips for his future needs: the million wafers that he intends to produce once his factory is running at full capacity (which should take at least ten years), and which according to him will be necessary to meet the needs of his companies in terms of cars, robots and data centers, would in fact account for two-thirds of TSMC’s annual production of two-nanometer chips.
If successful, this project would also allow it to considerably strengthen control over its semiconductor production chain, an objective that many players have been pursuing since the disruptions of Covid and the geopolitical turbulence leading in particular to fears of a Chinese invasion of Taiwan, the world cradle of the production of advanced semiconductors.
Finally, a semiconductor factory would admirably complement the ecosystem that Elon Musk is building around data, which according to a now widespread analogy, constitutes the gold of the 20th century. Indeed, Musk already has assets that constantly capture data on the real world through his Teslas, and soon his Optimus robots, while X gives him access to data published by users on the social network. And with xAI, it has the algorithms capable of processing this data.
Starlink provides it with the satellites which manage the internet layer to connect and make everything interact, satellites whose orbiting it controls thanks to SpaceX. Finally, Neuralink also opens up the prospect of bringing computer data into dialogue with that of the human brain. Terafab would constitute an additional brick within this ecosystem, and not the least since it would allow it to control a precious asset without which the algorithms for processing masses of data cannot function.
This is undoubtedly the ecosystem that Musk dreams of: a mix of infrastructure, hardware and software offering him almost total control over the collection and processing of data, allowing him in return to offer high value-added services, to refine the functioning of his cars and his robots, to improve his AI algorithms, and to invent new uses through the control of space and brain-machine interfaces. The vision lacks neither ambition nor elegance, but the path to achieving it promises to be strewn with pitfalls.
