Foundry Model
The foundry model is a business structure where a semiconductor manufacturer sells wafer fabrication as a service rather than designing and selling its own chips — a foundry's customer is the one who owns the chip design, the architecture, and the end product; the foundry's job is purely to turn that design into working silicon, reliably, at scale, for a fee. This single structural choice reshaped the entire semiconductor industry, splitting what used to be one vertically-integrated business into a whole ecosystem of specialized companies.
What a foundry actually sells is a platform, not just factory capacity, and that distinction is the heart of the business model. A customer doesn't just buy wafer-processing time — they get a process design kit (PDK) defining exactly how to design for that specific process, a set of design rules the layout has to follow to manufacture correctly, device models that predict transistor behavior accurately enough to simulate before committing to silicon, pre-built standard-cell libraries and SRAM compilers so a design team doesn't have to design every basic building block from scratch, and ongoing yield data and manufacturing feedback as the design moves toward production. Removing any one of these pieces would leave customers unable to actually use the factory capacity productively — which is exactly why building and maintaining this platform is as central to a foundry's business as running the fab itself.
The foundry model exists because designing a chip and manufacturing one require such different capabilities and capital structures that few companies can excel at both simultaneously. Chip design demands deep expertise in architecture, verification, and software, with capital needs that scale with engineering headcount. Chip manufacturing demands enormous, continuous capital investment in fabs and equipment — tens of billions of dollars for a single leading-edge facility — plus deep process-engineering expertise that has almost nothing in common with chip design skills. Separating these into fabless design companies and dedicated foundries let each side specialize completely, rather than every company needing to be excellent at both a design business and a capital-intensive manufacturing business at once.
Trust is the foundry model's real competitive moat, more than raw manufacturing capability alone. A customer sending a chip design to a foundry is handing over architecture, performance targets, and strategic intent — effectively the customer's competitive advantage, in file form. A foundry that also designed and sold competing chips would create an unavoidable conflict of interest, which is exactly why the most successful standalone foundries deliberately stay out of designing and selling their own end-product chips, positioning themselves as a neutral manufacturing partner that every fabless customer, including direct competitors of each other, can trust equally.
The platform model creates a flywheel effect, where each new customer design makes the platform better for every other customer on the same process. Every chip that moves through a given process node generates yield data, reveals design-rule edge cases, and surfaces process issues that the foundry can feed back into improving that same process for everyone building on it. More customers on a mature, well-characterized process generally means a more robust, better-understood platform — which is a major reason established leading-edge nodes tend to attract disproportionately more business than newer, less-proven ones, even when the newer node offers meaningfully better raw transistor density.
Running the foundry model well requires balancing enormous fixed costs against unpredictable customer demand, which is the business's hardest ongoing challenge. A leading-edge fab represents one of the largest capital commitments in any industry, built years before it can generate revenue and only economical if it runs at consistently high utilization for years afterward. Foundries have to make capacity-allocation decisions among customers competing for the same limited wafer starts, often including direct competitors in the same end market, while simultaneously funding the next process generation's development — all without the demand visibility that a company selling its own branded end product would normally have, since a foundry's order book depends entirely on decisions its customers make about their own products.