By announcing the construction of a chip factory in Austin on Saturday evening, Elon Musk reveals the growing concern of technology giants over their dependence on computing. With this new project, he intends to build, in Austin, a semiconductor factory capable of producing advanced chips for the specific needs of Tesla, SpaceX and its artificial intelligence subsidiary, xAI.
The project, called Terafabis presented as a joint operation of the two industrial groups, and as a direct response to the dependence of large technology companies on a handful of chip manufacturers whose capacities, despite massive investments, are struggling to keep up with soaring demand.
The rise of large artificial intelligence models has upset the balance of semiconductor manufacturers, where the industry previously managed a predictable progression of needs. It must today face colossal surges in demand, generated by a few players capable, overnight, of ordering thousands of graphics processors and requesting tens of thousands more in the following months.
A logic of vertical integration pushed to its conclusion
To understand Terafab, we must place the project in the industrial architecture that Musk has methodically built for a decade. Tesla already designs its own chips for autonomous driving, the famous H.W. which have equipped its vehicles since 2019, breaking with the dependence on Nvidia which had prevailed until then. SpaceX produces its own onboard systems, from the guidance processor to the atmospheric reentry computer. xAI, finally, consumes considerable volumes of calculation to train its Grok models, with needs that do not suggest any slowdown.
What Terafab is proposing is to internalize the last stage of the chain, namely manufacturing itself, which is today in the hands of a geographically concentrated oligopoly, exposed to geopolitical hazards and tensions in the Taiwan Strait that no one can ignore.
Controlling your chip supply chain means controlling your ability to exist in the artificial intelligence economy.
Dizzying ambitions, monumental obstacles
As usual, the figures mentioned by Elon Musk are dizzying, and the planned factory should be capable of producing chips on advanced nodes, where TSMC is still having difficulty controlling its increase in efficiency over this generation.
Producing a 2 nanometer chip requires mastering extreme ultraviolet lithography (EUV), a technology whose machines are manufactured almost exclusively by the Dutch ASML, each copy of which costs between 150 and 400 million euros. Today a state-of-the-art manufacturing site (a fab) requires several tens of billions of euros of initial investment, thousands of specialized engineers and several years of ramp-up before achieving acceptable returns.
TSMC, which today manufactures most of the most advanced chips on the planet, for Apple, Nvidia, AMD and Qualcomm, took forty years to acquire its technological lead. Intel, which has colossal resources and a century of experience in the semiconductor industry, has still not caught up on the finest nodes despite investments of several tens of billions of dollars.
The obstacles are the measure of the ambition. Building a state-of-the-art fab is one of the most complex and capital-intensive industrial projects there is. It requires lithography equipment in a quasi-monopoly situation, fine control of chemical processes, thousands of engineers trained in techniques that can only be acquired through decades of experience, and incompressible execution times that can be counted in years. In this area, TSMC, Samsung and the few established players have an advantage which is not only technological: it is institutional, human, cultural.
Austin, Texas and the geopolitical recomposition of the sector
The planned establishment in Austin is part of the industrial relocation movement that the American federal authorities have actively encouraged since CHIPS and Science Act of 2022, which mobilized more than $50 billion to repatriate part of semiconductor production to American soil. TSMC is already building two factories in Arizona. Samsung is investing heavily in Texas. Intel is increasing its projects in Ohio and Oregon.
The Texas ecosystem is consolidating around Austin, where Tesla moved its headquarters in 2021, where SpaceX has significant infrastructure and where the density of engineers trained in the technology industry continues to grow. Terafab is part of this logic and reinforces it; this proximity between players is not anecdotal in an industry where know-how circulates as much through individuals as through patents.
The spatial hypothesis, between vision and speculation
Among the perspectives mentioned by Musk, one falls outside the usual framework of industrial planning. SpaceX is exploring the possibility of putting satellites into orbit capable of hosting computing infrastructures, powered by solar energy available in abundance outside the Earth’s atmosphere. The initial units envisaged would be in the order of 100 kilowatts, with a gradual increase to megawatt levels.
If the logic is attractive with solar energy in space approximately eight times denser than on the ground, day-night cycles non-existent, and the heat dissipated by radiation in a vacuum avoids some of the cooling problems which burden terrestrial data centers, the constraints remain formidable: costs of putting into orbit, transmission latency, impossible maintenance in the event of failure.
The symptom of a deeper shift
Basically, what Terafab reveals is a shift that affects the entire technology sector. After a decade dominated by software, platforms and economic models based on capturing attention, the competition is moving towards physical assets: data centers, energy infrastructures, semiconductor production capacities. Artificial intelligence, as a technology, now anchors its challenges in a heavy industrial economy, where access to resources directly conditions performance and strategic survival.
Google, Microsoft, Amazon and Meta have all started designing their own chips. Apple has been doing this for years with obvious success. The question is no longer whether big tech companies will internalize semiconductor design, but how far down the chain they will go, and whether any will dare cross the threshold into manufacturing itself, where the barriers to entry are the highest in the entire global industry.
