Intel announced on April 7 that it will participate in Elon Musk's Terafab initiative alongside SpaceX and Tesla. The goal: manufacture 1 terawatt annually of computational capacity to support AI training and robotics workloads. Intel's stock rose more than 2% on the news. CEO Lip-Bu Tan and Musk shook hands at Intel's campus the previous weekend.
The specific scope of Intel's contribution has not been fully disclosed. TechCrunch noted the announcement confirmed participation without detailing what Intel Foundry will actually produce. What is clear is that Terafab — announced in March as a $20 to $25 billion chip fab complex in Austin — now has a foundry partner with established 18A process technology and an existing global manufacturing footprint.
Intel Foundry operating loss in 2025: $10.32 billion. Revenue growth: 3%. The division remains deeply unprofitable. Intel's 18A process node, previously restricted to internal use, is now being offered to external clients. Terafab would be among the largest potential customers in the foundry's history.
Intel Foundry is the central bet of CEO Lip-Bu Tan's restructuring program. Without large external customers, the foundry cannot achieve the utilization rates it needs to cover its fixed costs — fabs are expensive whether or not they are running at capacity. Landing Terafab as an anchor customer would change the unit economics of the foundry business in a way that no amount of internal optimization can.
D.A. Davidson analyst Gil Luria called the Tesla partnership "an important step" in Intel's organizational transformation. The framing is correct: Intel needs to show it can win the most demanding AI semiconductor customers, and Terafab's stated ambitions — 1 terawatt of compute annually, spanning chips for humanoid robots and orbital data centers — is exactly the kind of customer that validates a foundry's capability at the leading edge.
One terawatt of compute annually is a number that requires translation into physical terms. A modern AI accelerator draws 700 watts to 1,400 watts. At those TDPs, 1 terawatt of installed capacity implies millions of chips. Those chips go into racks. Those racks need cooling. Liquid cooling, at these densities, is not optional.
Terafab's stated ambition includes supplying chips for two distinct categories: humanoid robots and data centers — including Musk's stated goal of building AI data centers in space. The space data center component has its own cooling logic entirely: passive radiative rejection in vacuum, no CDUs, no cooling towers. The terrestrial component does not. Every chip that Intel manufactures for earthbound Terafab deployment lands in a liquid-cooled rack. The demand signal runs directly from the foundry to the CDU supply chain.
Intel's 18A process is produced in the United States — Oregon and Arizona primarily. In the current tariff environment, domestic semiconductor fabrication carries a structural advantage over Asian fabs. TSMC's Arizona facility is the only credible alternative for US-produced leading-edge chips. Intel Foundry with Terafab as a customer would give the US a second domestically produced AI chip supply chain. That is a geopolitical outcome as much as a commercial one.
The federal government is Intel's largest shareholder after its CHIPS Act capital infusions. The alignment between Washington's objectives and Intel's commercial recovery is direct. Whether that alignment produces a functional foundry that ships 1 terawatt of compute is still to be determined. The announcement is the beginning of a production story, not the end of one.