Will Musk’s TeraFab Redefine Space Data Centre Chips?

Elon Musk has outlined plans for TeraFab, a large-scale chip manufacturing facility designed to expand compute capacity for AI systems and data centre infrastructure. The project brings together Tesla, xAI and SpaceX in a single initiative focused on producing more than one terawatt of compute annually.

Elon unveiled the project on 21 March at the former Seaholm Power Plant in Austin, Texas. The facility integrates the full semiconductor production process – from chip design and mask fabrication to manufacturing, testing and packaging – within one site. This vertically integrated approach reflects growing pressure across the data centre sector to secure reliable access to advanced compute hardware.

Elon describes the project as “the most epic chip-building exercise in history so far”, positioning it as a response to rising demand for AI processing capacity.

Scaling compute for AI data centres

The TeraFab concept aligns closely with the needs of hyperscale data centres, where access to high-performance chips determines the ability to train and run large AI models.

Elon states that current global semiconductor output falls short of what is required for this level of expansion: “All of the fabs on Earth only provide 2% of what we need for the TeraFab project […] we’re very grateful to our existing supply chain, we would like them to expand as quickly as they can and we will buy all of their chips.”

This shortfall highlights a broader issue across the data centre industry, where demand for GPUs and specialised AI chips continues to outpace supply. By bringing fabrication in-house, TeraFab aims to reduce dependency on external suppliers while accelerating production timelines.

The Advanced Technology Fab in Austin forms the starting point for this strategy. It is designed to include all the equipment required for chip manufacturing, including lithography mask creation – a critical stage in semiconductor production that defines circuit patterns on silicon wafers.

Elon states that the integrated setup “doesn’t exist anywhere else in the world”, underlining the ambition to streamline chip production for large-scale compute deployment.

Linking terrestrial and space-based infrastructure

A defining feature of TeraFab is its intended allocation of compute output. Around 80% of production is expected to support space-based operations, with the remaining 20% reserved for use on Earth.

This approach connects with emerging concepts of distributed data centre infrastructure, where compute resources extend beyond traditional facilities. Elon links this strategy to energy availability, arguing that space offers fewer constraints than terrestrial environments.

"The way to scale civilisation is to scale power in space," he says.

Looking to space addresses challenges around power and cooling. On Earth, facilities must contend with grid limitations, rising energy costs and environmental considerations. In orbit, solar energy can provide a continuous power source without the interruptions caused by day and night cycles or weather conditions.

Elon highlights that “increasing power on earth has become harder and more expensive over time”, suggesting that shifting compute workloads to space could offer a more scalable alternative.

Integrated manufacturing and data centre demand

The TeraFab facility is designed to address not only chip supply but also the speed at which new compute capacity can be deployed. By combining design, fabrication and testing within a single location, the project reduces the complexity of global supply chains that currently support data centre hardware.

This model mirrors trends in the data centre sector, where operators are seeking greater control over infrastructure components to improve reliability and deployment timelines.

Elon also frames the initiative within a broader technological ambition, presenting it as a way to “turn science fiction into science fact”. While the statement reflects a wider vision, the immediate implication for data centres is the need to support increasingly complex workloads at scale.

The chips produced through TeraFab are expected to serve multiple use cases, including edge and inference applications. These workloads are central to modern data centre operations, particularly as AI systems move from training environments into real-time deployment.

Supporting next-generation compute workloads

The growing use of AI across industries is placing sustained pressure on data centre infrastructure. Facilities must not only expand capacity but also ensure access to specialised hardware capable of handling advanced workloads.

TeraFab’s proposed output – more than 1TW of compute per year – represents a scale that aligns with hyperscale requirements. By targeting both terrestrial and space-based deployments, the project reflects a shift towards more distributed and energy-aware data centre architectures.

The Austin facility is positioned as the first step in this process, with further expansion expected as demand increases. The development highlights the increasing importance of semiconductor manufacturing in determining how quickly new capacity can be brought online.