Musk Unveils US$25 Billion 'Terafab' to Build Proprietary Chips

Elon Musk has unveiled "Terafab," a vertical integration initiative aimed at manufacturing proprietary semiconductors to power Tesla’s autonomous systems, SpaceX’s orbital infrastructure, and xAI’s computational needs.

The project, announced on March 21 during an event at a decommissioned power station, marks a strategic pivot for Musk’s conglomerate of companies. Terafab will be located near Austin, Texas, and is designed to achieve a combined annual computing capacity of one terawatt (TW). According to industry benchmarks, one TW is equivalent to one trillion watts, a figure that approximates the total power generation capacity of the United States.

"We are very grateful to our current supply chain... but we need the chips, so we are going to build Terafab," Musk stated during the announcement.

Strategic Necessity and Vertical Integration

The move into semiconductor manufacturing is driven by a projected deficit in global supply. Musk indicated that the combined demand for chips across Tesla and SpaceX will significantly exceed the current production capacity of third-party global providers.

The project aims to achieve the most ambitious level of vertical integration in the history of the technology sector. Unlike the existing industry model, which relies on hyper-specialization, where different entities handle design, fabrication, packaging, and testing, Terafab intends to consolidate the entire lifecycle within a single facility.

"In Terafab, we intend to cover the entire process of development, manufacturing, packaging, testing, and improvement in the same facilities," Musk said.

If successful, this model would position Musk’s companies as direct competitors to established industry leaders such as TSMC, Samsung, and Intel, effectively eliminating reliance on external manufacturers for critical components.

Product Roadmap: Terrestrial and Orbital Silicon

Terafab’s production line is expected to focus on two distinct categories of semiconductors. The first category includes chips optimized for terrestrial applications, specifically targeting Tesla’s "Optimus" humanoid robots and the company’s fleet of autonomous "robotaxis." Musk noted that he expects the production rate of humanoid robots to eventually surpass that of vehicles by a factor of 10 to 100, necessitating a massive scale of silicon production.

The second category features a specialized processor designated as the "D3." These chips are designed specifically for aerospace environments, intended for use in low Earth orbit (LEO) satellites. Musk described the transition to space-based computing as a logical progression for the industry.

"It becomes an obvious decision. There will come a point where putting payload into orbit will be so cheap that hosting data centers in space will be cheaper than doing it on Earth because solar energy there is practically unlimited," Musk explained.

Operational Uncertainties and Industry Challenges

Despite the scale of the announcement, the project faces technical and logistical hurdles. Building a state-of-the-art semiconductor foundry requires Extreme Ultraviolet (EUV) photolithography equipment, a technology currently controlled exclusively by the Dutch firm ASML.

As of the announcement, Musk has not confirmed:

• A process partner or technology consultant.

• Formal agreements or purchase orders for ASML machinery.

• A specific initial investment budget, though media estimates place the figure between US$20 billion and US$25 billion.

• A definitive timeline or construction calendar for the Texas facility.

While Musk’s track record includes the successful disruption of the automotive and aerospace industries through Tesla and SpaceX, critics point to past projects that have faced delays or reduced scope, such as the Solar Roof, The Boring Company, and the Cybertruck.

Musk acknowledged the necessity of the venture despite the risks, framing it as the final component of his "master plan" to ensure the continued growth of his robotics and AI interests. "We need the chips," he reiterated, emphasizing that the current output from manufacturers like TSMC is insufficient to meet his long-term requirements for AI competition and satellite deployment.