Is it harder for Elon Musk to develop chips than to send Starship to Mars?

It is reported that in an interview with *The Korea Economic Daily*, industry insiders frankly stated that Elon Musk's Terafab fab project may be more difficult than sending Starship to Mars.

For practitioners who have worked in the semiconductor manufacturing industry for decades, tens of billions of dollars in capital can build the most advanced plants and purchase top-tier equipment, but cannot obtain stable high yield and mature process systems, which is exactly the biggest problem Terafab is currently facing.

After successively launching electric vehicles, Starlink, rockets and even large AI models, Elon Musk has set his sights on semiconductor manufacturing.

In March this year, Elon Musk announced the construction of the Terafab super fab in Texas, planning to build a semiconductor production system with an annual output of 1 terawatt of computing power, covering four major scenarios: autonomous driving, humanoid robots, space computing, and AI data centers.

This multi-billion-dollar project is jointly built by Tesla, SpaceX, and xAI. In addition to filling the chip supply gap for Musk's own industries, some people also regard it as a test of the global semiconductor manufacturing industry.

According to the plan, the project is built on the site of a decommissioned coal-fired power plant in Texas, with a first-phase investment of 16.8 billion US dollars and a planned area of more than 100 million square feet, equivalent to 1,400 standard football fields. The long-term total investment may reach as high as 1190 billion US dollars. If everything goes smoothly, the fab will complete capacity rollout and achieve large-scale mass production between 2027 and 2029, and finally build a full-process system covering chip design, lithography manufacturing, memory packaging, testing and verification.

At present, Tesla's autonomous driving, Optimus humanoid robot, SpaceX's space AI computing and xAI's large model training all require a huge number of dedicated high-end chips. If it relies on TSMC and Samsung for foundry services for a long time, it will not only face the risk of capacity being squeezed, but also cannot adapt to its own customized computing power needs. Therefore, Elon Musk wants to build an exclusive chip supply chain through Terafab, so as to keep the computing power in his own hands.

There have been many speculations in the industry about how the project will cooperate.

There were previous rumors that TSMC would participate deeply, but Elon Musk has explicitly refuted the rumors, stating that Terafab is fully responsible for construction and operation by Tesla and SpaceX. TSMC will only negotiate cooperation on plant and capacity leasing at most, will not participate in operational decision-making, and will not hold shares.

Of course, TSMC's own schedule is also very tight. NVIDIA and Apple have taken up most of the advanced process capacity, and AMD, Qualcomm, MediaTek and a large number of cloud vendors that develop self-developed AI chips all rely on TSMC for foundry services. In contrast, Intel's 14A advanced process exactly matches the mass production rhythm of Terafab. This 1.4nm-class process adopts cutting-edge technologies such as High-NA EUV lithography and backside power delivery, and will be the main solution for Terafab's initial mass production of logic chips.

In addition, choosing to cooperate with local US manufacturers can avoid geo-supply chain risks, and the two core businesses of aerospace and AI will be more secure. Intel also needs a heavyweight customer to verify the 14A process, and is willing to tilt technical resources to the project, so the two sides can meet each other's needs.

In addition, Elon Musk has chosen to bet on Free Electron Laser (FEL) technology for Terafab's lithography route, publicly claiming that FEL will win, trying to bypass ASML's traditional LPP-EUV lithography solution. In principle, this solution uses a particle accelerator to generate high-speed electron beams to produce EUV light sources with tunable wavelengths, with low pollution, energy consumption only about one-fifth of that of traditional solutions, and it can also adapt to more advanced processes in the future.

More importantly, FEL can be made into a centralized light source hub that supplies light to dozens of lithography machines in the whole plant, and the cost is much lower than buying a single EUV lithography machine.

However, FEL technology is currently only applied in the scientific research field, and it is conceivable how difficult it is to break through the technical ecosystem that ASML has accumulated over decades of deep cultivation in a short period of time.

A person familiar with TSMC's internal situation pointed out that building a fab itself is not a difficult task. The real challenge is how to continuously produce chips with stable quality and sufficiently high yield.

Advanced processes include hundreds of procedures such as lithography, deposition, etching, and cleaning. If there is a tiny deviation in any procedure or equipment fluctuation, the chip will fail.

A relevant person from Samsung said directly that a stable high-yield process system requires at least 30 years of experience accumulation. Even if you purchase top-tier equipment and build a standardized plant, you cannot replicate this accumulation in a short period of time.

In the early stage of mass production of new processes, many steps rely on long-term trial and error, and yield fluctuations between batches are the norm. Even with the support of Intel's process, equipment adaptation, process differences, and personnel systems in different plants still need to be re-run in. This cannot be solved simply by copying a production line, let alone the full-process implementation of chip design, manufacturing, storage and packaging.

Although this kind of integration can improve efficiency, it also raises the overall difficulty of the project to a higher level.

Conclusion

It is no secret that Elon Musk is keen to make grand promises, so the impact of Terafab on the stock prices of various companies may be more realistic.

For Intel, this is a key step in the IDM 2.0 strategy, which gives it the opportunity to get rid of the dilemma of having technology but no customers, and enter the first echelon of advanced process foundry services. For the entire semiconductor industry, the fact that a new tech giant is engaged in chip manufacturing is itself highly topical.

Competitors including Google, Amazon, and Microsoft are all accelerating the layout of their own chip production capacity. No matter whether it succeeds in the end or not, Terafab has already disrupted the competition pattern of global advanced semiconductor manufacturing, pushing this wave of global computing power autonomy to a new height.

This article is from the WeChat official account "Meku", author: Meku, published with authorization from 36Kr.