Musk’s TerraFab Aims 50x AI Chip Output
Tesla CEO Elon Musk announced the construction of a massive factory called ‘TerraFab’ capable of producing AI chips at an annual scale of 1 terawatt (TW). This decision stems from the assessment that current AI chip production volumes are insufficient to meet demand. The project will involve all three of Musk’s companies: Tesla, SpaceX, and xAI. However, critics argue the goal is unrealistic due to costs, time constraints, and technological challenges.
Musk unveiled the TerraFab project on the 21st (local time) at Tesla’s Giga Texas factory in Austin, Texas. The plan is to build TerraFab in Austin, Texas, where Tesla’s headquarters is located. The facility will produce chips for Tesla’s humanoid robot Optimus and specialized chips designed for use in space.
Musk stated,
We will establish a state-of-the-art factory equipped with all necessary equipment”
And added,
We aim to elevate the energy and computing capabilities of civilization by several levels.
The project will integrate the technological expertise of Tesla, SpaceX, and xAI. xAI, an AI startup, will control the cluster with its AI technology; Tesla will design and manufacture the chips; and SpaceX will handle space transportation and securing solar power.
The reason for TerraFab’s construction is the shortage of AI chips.
Musk explained,
While the semiconductor industry is increasing production, it is moving too slowly to catch up with the chip supply we require.
He added,
Even for humanoid robots alone, annual production could reach 1 billion to 10 billion units, far surpassing automobiles.
To address this, Musk ultimately aims to secure 1 TW of annual computing power. Since global computing power production by companies like TSMC and Samsung Electronics is approximately 20 gigawatts (GW) per year, this would require 50 times the current output. Musk emphasized, “That pace is far below what we need. Either we build TerraFab, or we will face a chip shortage—and since we need chips, we must build TerraFab.”
Musk also presented a blueprint for a ‘petawatt (PW) era,’ 1,000 times larger than the TW era. The plan includes establishing a massive forward base and factory on the Moon, relocating core AI computing chips requiring advanced technology to the lunar surface, and using local lunar resources to manufacture bulky and heavy components like solar panels, heat sinks, and server structures. Additionally, giant ‘AI satellites (space data centers)’ will be launched into orbits capable of receiving infinite solar energy using the Moon’s mass accelerator.
However, the semiconductor industry criticizes the project as unrealistic. Musk previously stated that TerraFab would produce 2-nanometer chips, but Tesla has no experience in mass-producing 2-nanometer chips or semiconductor manufacturing. Even if the technology exists, securing the necessary semiconductor equipment for 2-nanometer processes is globally challenging. Currently, Tesla designs chips but outsources manufacturing to foundries like Samsung. The space data center plan is also expected to require astronomical costs.
A source from the semiconductor industry remarked,
Without equipment, funds, or time, building TerraFab—let alone a ‘TSMC’—within years is nearly impossible.
Bloomberg News noted,
Musk has no prior experience in semiconductor production and has a history of making exaggerated promises about goals and timelines.
In response to such criticism, Musk previously, stated,
We will start with a small-scale factory and gradually expand.
READ the latest news shaping the AI Chips market at AI Chips News
Musk’s TerraFab Aims 50x AI Chip Output, source
Add comment