Elon Musk plans to build a large-scale chip factory. Can Tesla really manufacture its own chips?

Elon Musk Plans to Build His Own Chip Factory TeraFab

Phoenix Technology News, February 4th (Beijing Time) — Columnist June Yoon of The Financial Times in the UK wrote on Wednesday that Tesla’s $1.3 trillion market capitalization indicates that market expectations are increasingly relying on AI rather than electric vehicles for the company’s future. To meet this expectation, Tesla must gain greater control over the hardware that supports autonomous driving, robotics, and AI training.

Tesla CEO Elon Musk stated that the company needs to build and operate a so-called “TeraFab” large-scale chip factory to produce its own chips. But can Tesla really manufacture chips on its own?

This idea should be viewed separately from Musk’s more grandiose plans. Compared to ambitious projects like colonizing Mars or brain-computer interfaces, building a chip manufacturing plant relies on existing mature technology.

Not necessarily cutting-edge

Industry skepticism towards new entrants largely stems from a misconception: when people talk about chip manufacturing, they often think of TSMC producing NVIDIA’s flagship AI chips using 3nm and 5nm advanced process nodes. These high-end chips require the most advanced equipment and decades of trial-and-error experience. Judged by this standard, new entrants seem to have no chance.

However, this comparison itself is problematic. Chip manufacturing companies do not need to reach TSMC’s most advanced process levels to survive. There is actually a narrow middle ground in the chip industry: new entrants can meet technical requirements while having process performance sufficient to support AI workloads. If Tesla’s wafer plant is built, it would likely target older technology around 7nm. This process node is generally considered the last advanced chip generation before complexity and capital requirements sharply increase.

TSMC began mass production of 7nm chips in 2018, and this technology is still widely used in AI and data centers today. Even if Tesla lags by several years, it can theoretically produce AI chips suitable for autonomous driving and bionic robots.

Significant challenges

This means 7nm will be a suitable technical benchmark for Tesla, but it’s not an easy goal to achieve. It still requires ASML’s extreme ultraviolet (EUV) lithography machines, large-scale power, cooling, and water treatment systems, ultra-clean facilities, hundreds of precision equipment, and advanced chip packaging capabilities.

Most importantly, it needs hundreds of experienced engineers skilled in reducing chip defect rates, a scarce talent pool currently concentrated mainly at TSMC. Initial production typically takes three years or more, during which material loss rates are high, and a long trial-and-error process is necessary to achieve usable output.

TSMC

Even if Tesla can reach the technical threshold, relying solely on feasibility does not guarantee value creation. In the chip manufacturing field, the current benchmark is TSMC, which spent over $40 billion in capital expenditures last year. This expenditure is justified because its risk is spread across a large customer base and thousands of designs, helping absorb manufacturing losses and accelerate technological iteration.

Tesla cannot replicate this model. Musk has stated that Tesla does not plan to sell chips externally. Without global orders, the chip business becomes a fixed-cost operation, making it difficult to avoid structural losses. TSMC’s past investments in building factories in the U.S. show that each wafer plant costs at least $20 billion (even after subsidies). Unlike most industrial assets, chip fabs require continuous reinvestment to stay competitive. Even under the most optimistic assumptions, the payback period could be decades.

Aside from economic factors, execution risk remains a bigger challenge. Although Intel has decades of industry experience, its aggressive schedules and internal pressures have repeatedly delayed the transition to 10nm chips since the mid-2010s, resulting in years of postponements and a permanent loss of industry leadership.

There have been reports that Tesla’s electric vehicle quality is inconsistent, with issues like uneven body panel gaps and post-delivery repairs, but this does not necessarily equate to a lack of technical capability. However, it does reflect a tendency: rushing to deliver products before manufacturing processes are fully stable. For cars, such defects are mostly cosmetic and repairable. But in chip manufacturing, this approach simply does not work.

Failure cases

Another case worth learning from is GlobalFoundries in the U.S. After acquiring IBM’s loss-making chip business in 2015, the company concluded within three years that advanced chip manufacturing was economically unviable.

Tesla will face these two risks simultaneously: the cultural pressures Intel experienced, and the unfavorable economic realities that forced GlobalFoundries out of the market. Historical experience shows that this dual risk combination can particularly easily lead to value destruction, which often only becomes apparent after significant capital investment. (Author: Xiao Yu)