How the chip, electronics, and data center value chain works
Who designs, who makes, who packages, and who assembles; the economics of each step; and how data centers fit in.
Industry brief, with a one-minute summary: Semiconductors and electronics hardwareKey takeaways
- Making a chip takes many specialist companies in many countries. Design is expensive in people, manufacturing is expensive in factories and machines, and both need huge scale.
- A semiconductor (chip) is a tiny circuit made on a thin disc of silicon called a wafer.
- A data center is a building full of servers, storage, and network equipment, with power supply, cooling, and security.
Key idea
Making a chip takes many specialist companies in many countries. Design is expensive in people, manufacturing is expensive in factories and machines, and both need huge scale. That is why a few companies dominate each step.
A semiconductor (chip) is a tiny circuit made on a thin disc of silicon called a wafer. One wafer holds many chips (dies). Main chip types are logic (processors for computers and phones, graphics processors, and AI accelerators), memory (DRAM for working memory, NAND for storage, and high-bandwidth memory, HBM, stacked next to AI chips), analog and power chips (which manage electricity, common in cars and industry), and sensors.
- From design to device to data center
- Design software and IPElectronic design automation tools (Synopsys, Cadence) and licensed designs (Arm).
- Key: Chip designFabless designers such as Nvidia, AMD, Qualcomm, Broadcom, and MediaTek; integrated makers (IDMs) such as Intel, Samsung, Texas Instruments, and Infineon both design and make.
- Manufacturing equipment and materialsLithography from ASML (Netherlands); other tools from Applied Materials, Lam Research, KLA (US), and Tokyo Electron (Japan); wafers and chemicals.
- Key: Wafer fabrication (foundry)TSMC (Taiwan), Samsung Foundry (Korea), GlobalFoundries, SMIC (China), UMC.
- Assembly, test, and packagingCompanies such as ASE and Amkor; advanced packaging that joins AI chips with memory is a bottleneck.
- Electronics assemblyContract manufacturers such as Foxconn (Hon Hai) build phones, servers, and PCs, in China, India, Vietnam, Mexico, and elsewhere.
- Devices and data centersPhones, cars, industrial machines, and servers in data centers run by cloud firms and colocation companies.
Each step is concentrated in a few companies and countries.
| Model | What it does | Economics (approximate) |
|---|---|---|
| Fabless designer | Designs chips and pays a foundry to make them | Heavy R&D spending; gross margins often 50 to 75 percent for leaders with unique products |
| Foundry | Makes chips for many designers | Huge capex; a leading-edge fab costs tens of billions of USD; profit depends on high utilization |
| Integrated device maker (IDM) | Designs and makes its own chips | Controls supply but carries both R&D and factory costs |
| Memory maker | Makes DRAM and NAND in very high volume | Prices swing sharply with supply and demand; strong boom and bust cycles |
| Equipment maker | Sells the machines that make chips | Revenue follows chip makers' capital spending plans |
| Contract electronics manufacturer | Assembles finished devices | Low margins, often a few percent, high volume |
So-what
The ranges are rough. Fabs have very high fixed costs, so when demand drops, utilization falls and profit collapses; this makes the industry cyclical.
Supply chain geography
Chip design is led by US companies. Most leading-edge manufacturing is in Taiwan, and memory is concentrated in South Korea (Samsung, SK hynix), with Micron in the US and elsewhere. The most advanced lithography machines come from one company, ASML in the Netherlands, and Japan is strong in materials and tools. Assembly and testing are spread across Taiwan, China, Malaysia (especially Penang), Vietnam, and increasingly India. This concentration creates risk: a disruption in one place can stop production worldwide, which is why governments pay subsidies to build factories at home.
Data centers
A data center is a building full of servers, storage, and network equipment, with power supply, cooling, and security. Capacity is measured in megawatts (MW) of power for the IT equipment. Hyperscalers (very large cloud firms) build their own; colocation companies rent space and power to many customers, usually priced per kilowatt per month. The main costs are the building, electrical and cooling systems, the servers and chips inside (for AI, the largest cost by far), and electricity. PUE (power usage effectiveness) is total facility power divided by IT power: 1.0 would mean no power is used for cooling and other overheads; modern sites often run at about 1.1 to 1.4.
| Metric | Plain definition |
|---|---|
| Wafer starts and capacity | Wafers a fab can begin processing in a month |
| Utilization | Wafers actually processed divided by capacity |
| Yield | Share of chips on a wafer that work |
| Process node | The manufacturing generation (for example 3 nanometre); smaller nodes pack more transistors |
| Book-to-bill | New orders divided by shipments; above 1 means demand is growing |
| Days of inventory | Inventory divided by daily cost of sales; rising inventory warns of a downturn |
| IT load (MW) and PUE | Power used by servers, and total facility power divided by IT power |
So-what
In chips, watch utilization and inventory to see where the cycle is. In data centers, watch available power, not just land and buildings.
What is a fabless chip company?
Sources for this lesson (1)
- Recognized public explanations of case-interview concepts and frameworks
My notes on this lesson
0 of 5,000 characters. Saves automatically.
Try the 1 remaining check or drill above to complete this lesson (0 of 1 done).
Spotted something wrong or out of date? Report a mistake. We check every report and correct the page.