Terafab

On March 21, 2026, Musk staged a kickoff ceremony at a defunct Austin power plant for "Terafab": a fab meant to stamp out 100-200 billion custom AI and memory chips a year and, at full scale, a million silicon wafers a month. The world's largest single fabs, TSMC's, run on the order of a hundred thousand wafers a month. So the pitch was to build something roughly ten times bigger than anything that exists — a carmaker out-fabbing the company that took decades and $165 billion to reach this class of chip.

Tesla's 2020 Battery Day promised a 10-gigawatt-hour battery ramp within a year; five years on it hit about two percent of that goal.

Then the story shrank. Two weeks later, on April 7, Intel joined as the manufacturing partner, and the full-scale plant would run Intel's own process — making Tesla the marquee first outside customer for Intel's struggling foundry. Several analysts now read Terafab not as a from-scratch Tesla fab but as an Intel plant expansion with Tesla, SpaceX and xAI as anchor buyers. The cost moved the same direction: the $25B announced in March became a $55B prototype and up to $119B across all phases in a May regulatory filing, first reported by CNBC.

What's left that's solid is smaller than the headline. Tesla's AI5 chip is real and shipping — but its volume production is contracted to TSMC and Samsung plants in Arizona and Texas, not to Terafab, whose own output is a pilot line years out. The million-wafer number is Tesla's projection, not a built capacity. The useful way to hold it: a giant vertical-integration announcement whose verifiable core is a foundry contract and a fast-climbing budget.