Semiconductors & Semiconductor Equipment

Some companies create reusable intellectual property rather than manufacturing the final chip.

This can include:

• CPU architectures;
• processor cores;
• interface IP;
• memory-controller IP;
• physical-design IP;
• verification IP.

A chip designer can license these building blocks instead of developing every component internally.

How this role makes money

The business model can include:

• upfront license fees;
• annual access fees;
• per-unit royalties;
• support and related services.

Arm provides a clear example. Its filings describe a model in which customers license Arm IP and then typically pay royalties when chips containing that IP ship. [S8][S9]

Why this role exists

Modern chips are too complex for every company to reinvent every building block.

Reusable IP reduces:

• development time;
• engineering cost;
• technical risk.

Modern chips cannot realistically be designed by manually drawing billions of transistors.

EDA software provides tools for:

• circuit design;
• logic synthesis;
• physical layout;
• timing analysis;
• power analysis;
• verification;
• simulation;
• mask preparation.

Representative companies include:

• Synopsys;
• Cadence;
• Siemens EDA.

Cadence describes its business across core EDA, semiconductor IP and system design and analysis, while Synopsys similarly operates major EDA and semiconductor-IP businesses. [S7][S16]

How EDA makes money

Typical models include:

• software subscriptions;
• enterprise agreements;
• licenses;
• support;
• IP licensing.

Why EDA can have high switching costs

Chip-design flows are deeply integrated.

Changing tools can affect:

• engineering workflows;
• verification methods;
• design libraries;
• employee skills;
• relationships with foundry process-design kits.

That makes EDA different from ordinary productivity software.

A fabless semiconductor company designs chips but outsources most physical semiconductor fabrication.

Examples include:

• NVIDIA;
• AMD;
• Qualcomm;
• many other logic-chip designers.

NVIDIA's FY2026 filing explicitly describes a fabless and contract-manufacturing strategy. Qualcomm similarly describes a primarily fabless production model, and AMD identifies itself as a fabless semiconductor company. [S12][S13][S14]

What fabless companies concentrate on

They can focus resources on:

• architecture;
• circuit design;
• software;
• product definition;
• customer support;
• marketing;
• system design.

What they outsource

Depending on the company:

• wafer fabrication;
• packaging;
• assembly;
• testing;
• some raw-material procurement.

How they make money

They generally sell:

• chips;
• processors;
• modules;
• platforms;
• sometimes software or IP.

The trade-off

Fabless firms avoid the enormous capital cost of owning leading-edge fabs.

But they become dependent on external capacity.

That means supply agreements, foundry access, packaging capacity and manufacturing relationships become strategically important.

A semiconductor foundry manufactures wafers for customers.

The customer provides a chip design.

The foundry turns that design into physical silicon.

Representative foundries include:

• TSMC;
• Samsung Foundry;
• GlobalFoundries.

TSMC describes itself as a pure-play foundry whose mission is to support customers with technology and manufacturing capacity. [S4]

Samsung Foundry describes a business that manufactures system semiconductors for global fabless customers and supports them with process technologies, design infrastructure and packaging. [S15]

How a foundry makes money

Customers pay for manufacturing services.

Economics depend heavily on:

• wafer volume;
• process node;
• yield;
• utilization;
• manufacturing complexity;
• long-term capacity planning.

Why leading-edge foundries are difficult to build

The business requires:

• enormous capital investment;
• sophisticated process technology;
• highly specialized equipment;
• yield expertise;
• years of process development;
• customer trust.

Intel's FY2025 filing notes that leading-edge semiconductor manufacturing has become increasingly capital intensive, especially at advanced nodes using EUV lithography. [S10]

An IDM both designs chips and owns meaningful manufacturing capability.

Examples include companies such as:

• Intel;
• Texas Instruments;
• Micron;
• Samsung.

Not every IDM has the same degree of vertical integration.

Some may outsource parts of production.

Others may also manufacture chips for external customers.

Intel illustrates how the line can blur: it designs its own processors while also building an external foundry business for third-party customers. [S10]

Why the IDM model exists

Potential advantages include:

• tighter design-manufacturing coordination;
• control over process technology;
• internal capacity;
• optimization between product and fab.

The trade-off

Owning manufacturing requires enormous capital and creates more fixed costs.

When demand weakens, expensive fabs still exist.

That makes utilization important.

Fabs cannot manufacture modern semiconductors without highly specialized equipment.

Key equipment categories include:

• lithography;
• deposition;
• etch;
• inspection;
• metrology;
• cleaning;
• process control.

Representative companies include:

• ASML;
• Applied Materials;
• Lam Research;
• KLA;
• Tokyo Electron.

ASML describes lithography as a core part of semiconductor scaling and supplies systems that help customers manufacture increasingly advanced microchips. [S5]

How equipment companies make money

They sell:

• manufacturing tools;
• upgrades;
• spare parts;
• software;
• service agreements.

Why this layer can be powerful

A cutting-edge fab may have alternatives in some tool categories but very few qualified suppliers in others.

Equipment can take years to:

• develop;
• qualify;
• integrate into a process.

Semiconductor manufacturing also depends on specialized materials.

Examples include:

• silicon wafers;
• photoresists;
• specialty gases;
• chemicals;
• substrates;
• metals;
• packaging materials.

These materials may appear mundane compared with processors, but small purity or consistency problems can destroy yield.

The manufacturing process therefore depends on highly controlled material supply chains.

Memory is large enough to be treated as both a product category and a major industry role.

Representative manufacturers include:

• Samsung;
• SK hynix;
• Micron.

How memory differs from much of logic

Memory products can be more exposed to:

• large capacity cycles;
• inventory corrections;
• price swings.

At the same time, specialized memory such as HBM can become strategically differentiated when demand grows faster than capacity.

Why AI changed the memory conversation

Modern AI accelerators need to move enormous amounts of data.

That increases the importance of memory bandwidth.

Micron's filings describe AI servers as requiring increasing amounts of DRAM, including HBM. [S11]

After wafer fabrication, individual dies still need to become usable semiconductor products.

Back-end work includes:

• wafer bumping;
• die separation;
• packaging;
• assembly;
• testing.

Outsourced providers are commonly called OSATs: Outsourced Semiconductor Assembly and Test companies.

Representative companies include:

• ASE;
• Amkor;
• JCET.

Amkor describes itself as an outsourced semiconductor packaging and test provider, while SIA describes OSATs as contractors that provide assembly, packaging and test services to fabless firms and IDMs. [S3][S17]

Why packaging is becoming more important

Traditional packaging mainly protected a chip and connected it to a board.

Advanced packaging can now combine:

• multiple logic dies;
• chiplets;
• HBM;
• high-density interconnects.

TSMC's CoWoS products, for example, integrate logic and high-bandwidth memory for high-performance computing. [S6]

This means packaging increasingly affects:

• performance;
• bandwidth;
• power efficiency;
• system architecture.

Packaging is no longer just the last low-value step.