Gigafab
A gigafab is a semiconductor manufacturing site whose combined fab modules produce more than roughly 100,000 wafer starts per month, the informal industry threshold that separates a large single fab from a true gigafab-class campus.
The number, not the building, defines the term. A gigafab is almost never one building: it is a campus of several connected fab modules sharing one power substation, one ultrapure water plant, and one bulk-gas distribution system, with total output across every module crossing the 100,000 wafer starts per month mark that TSMC itself popularized when describing sites like Fab 15 and Fab 18.
Shared infrastructure is the entire economic argument. Once a site has paid for its substation, water reclamation plant, and chemical delivery backbone, adding a second or third fab module costs far less per wafer of added capacity than building an equivalent standalone fab elsewhere, because the most expensive fixed systems are already sunk and simply get used harder.
Three campuses define the current scale frontier. TSMC's Fab 18 in Tainan runs multiple phases of 3 nm production and is ramping 2 nm, with site-wide output now well past the gigafab line; Samsung's Pyeongtaek campus in South Korea is the largest single semiconductor fab site in the world by floor area, built in phases since 2015; Intel Foundry's Ocotillo campus in Chandler, Arizona pairs Fab 52 and Fab 62 in a roughly 30 billion dollar buildout aimed at Intel 18A.
Capacity arrives in phases, not all at once. A new site secures land, power allocation, and water rights first, brings one module online and ramps yield on it, then adds further modules onto the same shared backbone, each one incrementally cheaper to build than the last because the hardest infrastructure problem is already solved.
| Gigafab campus | Operator | Lead node | Scale |
|---|---|---|---|
| Fab 18, Tainan | TSMC | 3 nm, ramping 2 nm | multi-module, over 100,000 WSPM |
| Pyeongtaek campus | Samsung Foundry | 3 nm GAA | world's largest single fab site by floor area |
| Ocotillo campus (Fab 52/62) | Intel Foundry | Intel 18A | roughly 30 billion dollar buildout |
| Fab 21, Taichung | TSMC | 28 nm and mature nodes | high-volume mature-node specialty capacity |
| Clay, New York (planned) | Micron | memory | up to 100 billion dollar multi-decade buildout |
Gigafab status also changes supplier leverage. A site ordering dozens of ASML EUV scanners and hundreds of Applied Materials and Lam Research tools at gigafab scale negotiates pricing and delivery priority that a single mature-node fab simply cannot, which is part of why TSMC's capital efficiency per wafer keeps compounding ahead of smaller competitors.
Memory manufacturers are following the same playbook. Micron's planned Clay, New York campus is structured as a multi-decade, multi-module buildout with a headline budget as high as 100 billion dollars, explicitly designed to cross gigafab scale in phases rather than open as a single finished site.
Read a gigafab through a *shared-infrastructure* lens rather than a *building-size* lens: the 100,000 wafer starts per month threshold matters because it marks the point where a site's fixed utility backbone is being amortized across enough volume to out-compete smaller, standalone fabs on cost per wafer, the same compounding-scale logic that lets TSMC's Tainan campus and Samsung's Pyeongtaek site keep extending their lead over single-module competitors.