Elon Musk has said that Tesla needs to build and operate a semiconductor manufacturing plant, which he has called "TeraFab". This massive project, which will cost billions of dollars, marks a further expansion of the company's business scope, no longer limited to its core electric vehicle business.
"To eliminate potential capacity bottlenecks in the next three to four years, we must build a Tesla TeraFab," Mr Musk said on the company's earnings call on Wednesday. "A huge factory that covers logic, memory, packaging and so on, and it will be built in the US." Tesla, the world's most valuable carmaker, has bet its future on artificial intelligence, autonomous driving and robotics. These projects have huge demand for chips, and the Austin-based company currently sources chips mainly from Samsung Electronics and TSMC.
Musk pointed out that existing suppliers, including TSMC, Samsung, and Micron Technology, are unable to meet Tesla's supply needs.
"This is crucial to ensure that we are protected from geopolitical risks," said Musk. "I think people may underestimate some geopolitical risks, which will become an important factor in a few years." The world's high dependence on Taiwan's TSMC and its domestic production capacity is directly related to the supply of cutting-edge chips.
In recent weeks, the world's richest man has hinted that Tesla may produce its own chips to address what he sees as a major bottleneck in the fiercely competitive AI race: insufficient chip supply.
"If we don't build a fab, we're going to hit a chip bottleneck," Mr. Musk said in a recent podcast with Peter Diamandis, the founder of the X Prize Foundation, referring to the industry term for a chip factory. "We have two choices: We hit a chip bottleneck or we build a fab." In November, Mr. Musk told Tesla shareholders that the company might need to build a "TeraFab" - a trillion-scale chip factory - saying: "I don't see any other way to get to the volume we need." The economics of chip making are brutal. Building a cutting-edge factory requires tens of billions of dollars in fixed costs and a long time to ramp up production from start-up to full operation.
This also requires the procurement of complex machinery and equipment from multiple suppliers, especially ASML Holding, a European company that holds a dominant market position in a crucial aspect of the manufacturing process.
Despite the significant challenges in constructing a chip factory, it aligns with Musk's consistent strategy of vertical integration. By incorporating key components into in-house production, his companies can operate more swiftly than relying on supply chains. The overlapping businesses of Tesla, SpaceX, xAI, Neuralink, and the Boring Company are also increasingly evident.
It is currently unclear which region of the United States the factory will be built in, as well as the specific construction schedule.
Tesla is expected to spend more than $20bn on capital expenditure at its existing factories this year. The funding for the construction of "infrastructure" such as solar cell manufacturing plants and chip factories remains to be seen.

"We have over $44bn in cash and investments on our books. So we will use internal resources, but we also have other avenues to raise funds," said Vaibhav Taneja, chief financial officer. "As long as the cash flow is stable, you can go to the banks," he added. "We have already had discussions with banks," Mr Taneja said. "We need to look further into how we can raise funds, whether through additional debt or other means." Mr Musk said Tesla would make "more important announcements" about the TeraFab project in the future.

Tesla CEO Elon Musk stated at the shareholders' meeting that in order to meet the rapidly growing demand for AI chips, Tesla is considering building its own "TeraFab" with a scale surpassing TSMC's "Gigafactory". However, chip manufacturing involves extremely high technical thresholds and substantial investment. NVIDIA CEO Jen-Hsun Huang cautioned that advanced manufacturing processes cannot be replicated simply by throwing money at them, and the difficulty is quite high.
Musk pointed out that in order to meet the company's huge semiconductor demand in the field of artificial intelligence (AI), it may directly invest in building its own chip production business, and a chip manufacturing base named "TeraFab" (Terafactory) needs to be established.
He compared it to TSMC's "Gigafab" (Gigafactory), which has a monthly production capacity of over 100,000 wafers, and emphasized that the scale of the new factory would be "much larger".
Currently, TSMC refers to its factories producing 30,000 to 100,000 wafers per month as "Megafab" (Mega-scale Factory), while those producing over 100,000 wafers per month are referred to as "Gigafab".
If Tesla completes the TeraFab, its monthly production capacity will far exceed 100,000 wafers, significantly surpassing today's mainstream wafer manufacturers and even positioning it as one of the world's largest chip manufacturers.
Taking the reference example, TSMC's Fab 21, located in Arizona, USA, with a total investment of $165 billion, is expected to become a Gigafab-level park in the future. However, Musk stated that Tesla's plan will be even more ambitious than this.
However, in response to Musk's vision, NVIDIA CEO Jen-Hsun Huang pointed out at a TSMC-related event a few days ago that the complexity of chip manufacturing is often underestimated by outsiders.
He bluntly stated, "Establishing advanced chip manufacturing capabilities is extremely difficult. Besides the factory itself, TSMC's accumulated engineering technology, scientific research, and process experience all pose high challenges."

As a company possessing both an AI supercomputer and substantial computing demands for automotive applications, Tesla has procured a large number of NVIDIA GPUs. Following the discontinuation of the Dojo project, Tesla is now promoting the development of its own AI5 processor for use in autonomous vehicles, robots, and data centers.
To ensure a stable supply, Tesla currently adopts a "dual-source foundry" approach with TSMC and Samsung. Musk also indicated that Intel (INTC-US) may also become a potential partner, but no agreement has been signed yet.
Musk emphasized that as Tesla's AI applications continue to expand, external supply will struggle to meet demand, necessitating consideration of becoming a vertically integrated manufacturer (IDM) akin to TSMC and Samsung.
"Even if we estimate the best-case scenario for supplier chip production, the future supply of chips will still be insufficient. Therefore, we may have to build a TeraFab, which is imperative," said Musk
To master advanced chip manufacturing, the capital and technology required far exceed what outsiders imagine. Based on the current industry standards, a wafer fab with a monthly production capacity of approximately 20,000 wafers and capable of mass producing cutting-edge processes often requires tens of billions of dollars in investment, and this does not include the subsequent costs for process development and mass production tuning.

As an example, Japanese startup chip maker Rapidus is attempting to challenge this high wall.
The company plans to establish mass production capabilities for 2 nm processes in the coming years and estimates that a commercially viable production facility will be completed by 2027, with an overall expenditure of approximately 5 trillion yen (approximately $32 billion).
While this goal certainly demonstrates a strong ambition, in today's highly mature global semiconductor competition, it remains uncertain whether a new player can successfully break into the most advanced nodes.
Analysis points out that the R&D process of advanced technology is inherently a lengthy and highly interdisciplinary challenge. Starting from the formulation of process routes, material and transistor architecture design, to verifying electrical properties, stress, and leakage behavior through extensive TCAD simulations, any deviation in any step may require the entire process to be redone. Therefore, just the "starting" phase often takes several years.
Although Rapidus has obtained the 2 nm GAA transistor architecture authorized by IBM and secured partial technical cooperation from imec in Belgium and CEA-Leti in France, the transistor architecture is merely the starting point of the entire R&D chain.
Next, the engineering team needs to design and fine-tune thousands of process steps, including front-end-of-line (FEOL) transistor formation, middle-of-line (MOL) contact layer, and back-end-of-line (BEOL) metal interconnect. The processes such as deposition, etching, lithography, and annealing all require atomic-level precision.
Each step involves a large number of parameters, requiring profound engineering experience and repeated experimentation to ensure mass production reliability and yield.
Even after the aforementioned process flow has been completed in series, it is still necessary to establish a PDK (Process Design Kit), SPICE model, and standard cell library to ensure that the chip design team can actually use this process for circuit design.
At this point, the factory side must also synchronously adjust the settings of the production line equipment to ensure stable output in a real mass production environment, which is also not a process that can be accelerated by throwing money at it.
Ultimately, the real test lies in "yield rate". Whether a new entrant can achieve a profitable high yield rate for advanced processes in a short period of time is the key to securing a foothold in the market. This often requires a long-term presence of a seasoned engineering team, repeated adjustments, and countless failures.
As for whether Rapidus can deliver results in 2027, the industry generally holds a wait-and-see attitude. The outcome still needs time to be verified.