Terafab: The Trillion-Watt Compute Factory by SpaceX, xAI, Tesla, and Intel

A joint venture between SpaceX, xAI, Tesla, and Intel is building what they call Terafab, a compute facility designed to consume one terawatt of power. If completed, it would be the largest single computing installation ever built, dwarfing existing hyperscale data centers by orders of magnitude.

What Terafab Is Designed For

Terafab is purpose-built for training next-generation AI models. Current frontier AI training runs consume hundreds of megawatts; Terafab is designed for workloads requiring 100x that capacity. The facility would house millions of GPUs and custom AI accelerators, connected by a networking fabric capable of moving exabytes of data between processors per second.

The “trillion-watt” label refers to peak electrical capacity, not continuous draw. Actual sustained consumption would be lower, but the infrastructure needs to handle peak loads during intensive training phases. For context, one terawatt is roughly equivalent to the total electrical generation capacity of Germany.

Why These Four Companies

Each partner brings a critical capability. xAI provides the AI models and training workloads that justify the facility’s existence. Tesla contributes energy infrastructure expertise, including Megapack battery storage and potentially solar generation capacity. SpaceX handles construction logistics (the company’s Starbase operations demonstrated massive-scale facility builds). Intel brings semiconductor manufacturing, potentially producing custom AI chips fabricated at Intel’s foundries.

Elon Musk’s involvement across three of the four partners raises governance questions that the companies have not yet addressed publicly. How compute access is allocated between competing interests within the Musk ecosystem remains unclear.

The Power Problem

Sourcing a terawatt of power is the project’s biggest challenge. No single grid connection can deliver that capacity. The current plan involves a combination of dedicated natural gas generation, large-scale solar installations, nuclear partnerships, and grid connections across multiple substations.

Environmental groups have already flagged concerns about the facility’s carbon footprint. Tesla’s involvement suggests a commitment to renewable energy, but the gap between renewable aspirations and the raw power demands of AI training is enormous. Even optimistic energy efficiency improvements cannot close it entirely at this scale.

Timeline and Location

Terafab is reportedly planned for a site in Texas, near existing SpaceX and Tesla operations. Construction would happen in phases, with initial capacity coming online in late 2027 and full build-out extending to 2030. The phased approach allows the consortium to begin training runs while expanding capacity.

The total investment is estimated between $50-100 billion, making it one of the largest private infrastructure projects in American history. Financing details have not been disclosed publicly, though all four companies have sufficient cash reserves and market access to fund their shares.

What This Means for the AI Industry

Terafab represents a bet that AI progress is limited primarily by compute. If bigger training runs produce proportionally better models, then the entity with the most compute wins. This philosophy, called the “scaling hypothesis,” has held true so far but faces skeptics who argue that algorithmic improvements matter more than raw scale.

For competitors, Terafab raises the barrier to entry in frontier AI development. If training the next generation of models requires terawatt-scale facilities, only a handful of organizations worldwide can participate.