Fabless vs foundry: what is the difference?
The split between design and manufacturing is the defining structure of today's chip industry. Understanding it explains who invests in what, where each kind of company makes its profit and why capacity at a few foundries matters so much.
The short answer
A fabless company designs and sells chips but owns no wafer fab; a foundry manufactures chips designed by other companies. Fabless firms such as NVIDIA, Qualcomm and AMD spend on engineers, EDA tools, IP and software instead of factories, while foundries such as TSMC carry the huge capital cost of leading-edge fabs and spread it across many customers. The two are complementary halves of one supply chain and are usually contrasted with the integrated device manufacturer (IDM), which both designs and manufactures, as Intel and Samsung do.
Side-by-side comparison
| Aspect | Fabless | Foundry |
|---|---|---|
| Role | Chip design and verification, marketing, software | Wafer manufacturing of chips designed by customers |
| Fabs | None | Owns and operates them; the core of the business |
| Main investment | Design engineers, EDA tools, IP licences, software | Process equipment and fab construction |
| Source of profit | Product performance, features and the ecosystem around them | Process technology lead, yield and fab utilisation |
| Own-brand chips | Yes | Pure-play foundries do not, to avoid competing with customers |
| Main risks | Securing capacity; dependence on a single foundry | Low utilisation in downturns; recovering investment in new nodes |
| Typical examples | NVIDIA, Qualcomm, AMD, Broadcom, MediaTek | TSMC, Samsung Foundry, GlobalFoundries, UMC, SMIC |
| Japanese examples | Socionext | Rapidus (planning 2 nm production), JASM (TSMC's majority-owned Kumamoto subsidiary) |
How the split emerged
Until the 1980s most chip companies were IDMs that designed and built their own products. As scaling advanced, the cost of each new fab rose, and fewer companies could afford to keep building leading-edge capacity on their own.
TSMC, founded in Taiwan in 1987, introduced the pure-play foundry model: it would make no products of its own and only manufacture for others. That gave companies without fabs access to advanced processes, and fabless start-ups multiplied. Established IDMs later followed; AMD spun off its manufacturing in 2009, and that operation became GlobalFoundries.
Different profit engines
A fabless company competes on what it designs. Its advantage often extends beyond the chip to the software and development tools that make it usable; NVIDIA's combination of GPUs and its software platform is the standard example. Without fabs, fixed costs are lighter and successful products can carry high margins.
A foundry competes on how it manufactures. It recovers its capital spending by filling fabs with orders from many customers and keeping utilisation high. Being first to ramp a new node with good yield gives pricing power, and that lead is a major reason TSMC has been able to sustain strong margins.
Mirror-image risks
Because fabless companies outsource production, they cannot ship if they fail to secure foundry capacity when demand surges. The concentration of leading-edge manufacturing in Taiwan is also discussed as a geopolitical risk for the whole fabless sector.
Foundries face the opposite exposure. When demand falls, utilisation drops and fixed costs weigh heavily, and investment in each new node must be committed years before the demand it is meant to serve. One side's risk is essentially the reverse of the other's.
IDMs and hybrid models
An IDM can co-optimise products and processes but must fund both. In memory, where process technology is itself the main competitive edge, IDMs such as Samsung, SK hynix, Micron and Kioxia dominate. Many companies sit between the categories: Intel remains an IDM while offering foundry services and outsourcing some of its own products to TSMC, and 'fab-lite' companies keep mature nodes in-house while sending leading-edge designs to foundries. In practice the three labels describe a spectrum rather than strict boxes.
FAQ
What is the difference between fabless and IDM?
A fabless company specialises in design and outsources manufacturing to foundries. An IDM designs, manufactures and sells its own chips. Intel, Samsung and Texas Instruments are well-known IDMs.
Why doesn't TSMC sell its own chips?
To avoid competing with its customers. If a foundry sold its own chips, customers would be handing their designs to a potential rival. TSMC has kept to this policy since its founding, which underpins the trust fabless companies place in it.
Is Apple a fabless company?
For chips, it works like one: Apple designs its A-series and M-series processors and has them manufactured by TSMC. Its core business, however, is building those chips into its own products rather than selling chips, so it differs from conventional fabless chip vendors.
Do foundries also do packaging?
Traditionally foundries made wafers and outsourced assembly and test (OSAT) companies handled packaging. As advanced packaging has become critical, TSMC also offers its own packaging and 3D integration services such as CoWoS and SoIC.
Terms compared
- Fabless, Foundry and IDMThe three main ways chip companies divide work: IDMs design and make their own chips, fabless firms only design, and foundries only manufacture.
- ASIC (Application-Specific Integrated Circuit)An integrated circuit designed for one specific application, giving the best performance and power efficiency for that task.
- MPW (Multi-Project Wafer)A prototyping service where many chip designs share one mask set and wafer run, splitting the cost; also called a shuttle.
Japanese companies involved
- Kioxia · NAND flash memory and SSDs
- Renesas Electronics · Microcontrollers, automotive SoCs, analog and power chips
- JASM (TSMC Kumamoto) · Foundry fab operated by a TSMC-majority joint venture
- Sony Semiconductor Solutions · CMOS image sensors
- Rapidus · Leading-edge logic foundry (2 nm, in development)
- Socionext · Fabless custom SoC design
More industry and supply chain comparisons
Sources
Semiconductor comparisons
- All semiconductor comparisons
- Tokyo Electron vs Applied Materials: what is the difference?
- EUV vs DUV lithography: what is the difference?
- High-NA EUV vs standard EUV: what is the difference?
- Single patterning vs multi-patterning: what is the difference?
- Chemically amplified resist vs metal oxide resist: what is the difference?
- ALD vs CVD: what is the difference?
- CVD vs PVD: what is the difference?
Industry and supply chain terms
Last updated: November 2, 2026