Foundry vs Fabless vs IDM

Designing a chip and manufacturing a chip are two very different jobs — one is closer to architecture and software, the other is closer to running a factory that costs tens of billions of dollars to build. Almost every company in the semiconductor industry has had to choose how much of that full journey — from RTL design to a finished, packaged chip — it wants to own itself. That choice sorts companies into three business models: IDM, fabless, and foundry.

Three ways to be in the chip business

IDM (Integrated Device Manufacturer) companies design their own chips and manufacture them in their own fabs. They own the entire chain, from the first sketch of a circuit to the finished wafer coming off the line. Intel is the textbook example — for decades it designed its own processors and ran its own fabs to build them, and Samsung’s semiconductor division still does both today.

Fabless companies design chips but own no factories at all. They do the architecture, the RTL-to-GDSII design flow, and the verification — then hand a finished design file to someone else’s factory to actually manufacture. Nvidia, Qualcomm, AMD, and Apple’s chip division are all fabless: they design (and often achieve enormous success designing) but never operate a wafer fab themselves.

Foundry companies do the opposite: they own and run fabs but generally don’t design their own products. A foundry’s customers are the fabless companies — the foundry takes a finished chip design and turns it into physical silicon, then hands the wafers back. TSMC is the world’s largest pure-play foundry, built entirely around manufacturing other companies’ designs.

🧩 Think of it like… running a restaurant. An IDM is like a restaurant that grows its own vegetables, raises its own livestock, *and* cooks the meal — total control, but you need farmland, barns, and a kitchen all at once. A fabless company is like a chef who designs an incredible menu but doesn't own a farm — they order ingredients from a specialized supplier and focus entirely on the recipe. A foundry is that supplier: a massive, hyper-efficient farm-and-butchery that doesn't design any menus itself, it just fulfills orders from many chefs at once, as precisely as possible.

Why the industry split apart

In the early decades of chip-making, IDM was close to the only model — building a fab was expensive, but not so expensive that a single company couldn’t design and manufacture together. That changed as transistors shrank and manufacturing became dramatically more complex and capital-intensive. A single modern fab, equipped with tools like EUV lithography machines that individually cost hundreds of millions of dollars, can cost tens of billions of dollars to build. Very few companies can afford to build cutting-edge fabs and keep up with chip design at the same time.

This economic pressure created an opening: if manufacturing was going to be brutally expensive regardless of who used it, why not build fabs that serve many different design companies, spreading that enormous cost across many customers’ orders? That’s the foundry model, and TSMC — founded in 1987 specifically to manufacture chips designed by others — proved it could work at scale. Once foundries existed, a company no longer needed billions of dollars in fab investment just to have a great chip idea; it could design a chip and simply have TSMC, Samsung Foundry, or GlobalFoundries manufacture it. That unlocked an explosion of fabless design companies from the 1990s onward.

    flowchart TB
    subgraph IDM["IDM"]
    direction TB
    I1["Design team"] --> I2["Own fab"] --> I3["Finished chip"]
    end

    subgraph Split["Fabless + Foundry"]
    direction TB
    F1["Fabless company<br/>designs chip"] -->|hands off GDSII file| F2["Foundry<br/>manufactures wafers"]
    F2 --> F3["Finished chip"]
    end

The lines have blurred somewhat over time. Some historically pure IDMs now also manufacture for outside customers (Intel Foundry offers manufacturing services to other companies), and some fabless companies own small specialty fabs for niche products. As of the mid-2020s, “IDM,” “fabless,” and “foundry” describe a company’s dominant strategy, not always an absolute, unmixed category.

Why the distinction matters

Knowing which model a company follows tells you a lot about how it competes and what risks it carries:

A fabless company’s competitive edge is design talent and architecture — but it depends entirely on a foundry partner’s manufacturing capability and available capacity, which it doesn’t control.
A foundry’s competitive edge is process technology and manufacturing execution — its success depends on serving many customers well and staying at the leading edge of process nodes.
An IDM controls its whole destiny end-to-end, but also bears the full capital burden of keeping fabs current, even in years when demand or design wins are weak.

This split is also why a handful of companies — the foundries and the specialized equipment makers that supply them — sit at a genuine chokepoint in the world’s electronics supply, even though hundreds of fabless companies compete freely on chip design.

Key takeaways

IDM companies both design and manufacture chips in-house (e.g., Samsung’s semiconductor business, historically Intel).
Fabless companies design chips but outsource manufacturing to a foundry (e.g., Nvidia, Qualcomm, AMD).
Foundry companies manufacture chips designed by others, typically without designing their own products (e.g., TSMC).
The split emerged because leading-edge fabs became too expensive for most companies to build alongside design work, so manufacturing consolidated into specialized foundries serving many fabless customers.
See Major Players Map for real companies in each category, and Design Flow: RTL to GDSII for what a fabless company actually hands off to a foundry.

Major Players Map