Last updated: 27 September 2026
Short answer: making a chip involves a chain of specialists, and several links depend on just one or two companies. Apple, NVIDIA and AMD design chips; Synopsys and Cadence supply the design software; ASML is the only maker of the EUV lithography machines needed for leading-edge chips; TSMC manufactures about 72 percent of the world's contract-made chips; three companies make almost all high-bandwidth memory; and advanced packaging is a bottleneck of its own. In 2026, AI demand is stretching memory and packaging hardest, which is why graphics cards, RAM and laptops have become more expensive.
What are the stages of making a chip?
Most people picture a chip being "made" in one factory. In reality it passes through at least six specialist stages, often in different countries:
| Stage | What happens | Leading companies | Main locations |
|---|---|---|---|
| Design | Engineers design the chip's logic | Apple, NVIDIA, AMD, Qualcomm, MediaTek | US, Taiwan |
| Design software (EDA) | Tools to design and verify the chip | Synopsys, Cadence | US |
| Equipment | Machines that pattern, deposit, etch and inspect | ASML, Applied Materials, Lam Research, Tokyo Electron, KLA | Netherlands, US, Japan |
| Fabrication (foundry) | Chips built on silicon wafers | TSMC, Samsung, SMIC, UMC, GlobalFoundries | Taiwan, Korea, China, US |
| Memory | DRAM, flash and high-bandwidth memory | SK hynix, Samsung, Micron | Korea, US, Japan, Taiwan |
| Packaging and test | Chips mounted, connected and tested | TSMC (CoWoS), ASE, Amkor | Taiwan, Southeast Asia |
Materials run underneath all of it: silicon wafers, specialty gases and chemicals, many supplied by Japanese companies. A problem at any one stage can hold up the whole chain.
Who dominates chip manufacturing?
TSMC, by a wide margin. According to TrendForce's figures for the second quarter of 2026, global foundry revenue approached $53.5 billion, and TSMC took about 72.5 percent of it. Samsung came second with 5.9 percent, with China's SMIC close behind at 5.4 percent and narrowing the gap.
SMIC's rise is notable because export controls prevent it from buying EUV lithography machines, so it has grown using older-generation equipment and by serving strong demand from Chinese customers. That limits how advanced its processes can become, but it shows how much of the market runs on mature technology.
That concentration is even sharper at the leading edge. The most advanced processes, the ones inside new iPhones and AI accelerators, are almost all made by TSMC, which began volume production of its 2nm process at the end of 2025. I explained what yield on that process means for chip prices in what TSMC's 2nm yield actually means for the chips you'll buy.
Why is ASML so important?
Because every leading-edge chip depends on machines only ASML makes. Extreme ultraviolet (EUV) lithography prints the finest features on today's most advanced chips, and ASML is the sole supplier of EUV systems.
Its newest generation, High-NA EUV, is even more concentrated. ASML's TWINSCAN EXE:5200 machines are reported to cost around $380 million to $400 million each, and fewer than a dozen were installed worldwide in 2026, at Intel, SK hynix, Samsung and the research centre imec among others. A single company building a single class of machine sits upstream of the whole advanced chip industry.
Why has memory become the bottleneck?
AI accelerators need huge amounts of fast memory next to the processor, a type called high-bandwidth memory (HBM). Only three companies make it at scale: SK hynix, Samsung and Micron. SK hynix has held the largest share for most of the AI boom, with Samsung gaining ground in 2026 as it ramped its newest HBM4 products.
HBM is more profitable than ordinary memory, so manufacturers have shifted capacity towards it. The knock-on effect reaches ordinary buyers: DRAM for PCs and phones has become scarcer and far more expensive, and PC makers have passed those costs on in 2026 laptop prices. Graphics cards have been hit hardest, because AI buyers compete for the same parts.
What is advanced packaging, and why does it limit AI chips?
Modern AI chips are not one piece of silicon. A processor and several stacks of HBM are placed side by side on a shared base and wired together at very high density. TSMC's version of this is called CoWoS, and it has been one of the tightest constraints on AI accelerator supply, with TSMC's packaging plants in Taiwan running at full capacity through 2025.
This is why "we have the wafers" does not always mean "we have the chips". A finished accelerator needs a fabricated processor, enough HBM and a packaging slot, and a shortage of any one holds up the rest.
Is the supply chain moving out of Asia?
Slowly, and partly. TSMC's first fab in Arizona entered high-volume production of its N4 process in late 2024, with yields reported as comparable to its Taiwan plants. TSMC plans six fabs, two advanced packaging plants and a research centre there, with construction of the first US packaging plants starting in 2026 and production expected around 2028.
Arizona is still a small part of TSMC's output, and the most advanced processes start in Taiwan first. Memory production remains concentrated in Korea, and much of packaging and materials stays in Asia. Diversification is real, but it is measured in years and hundreds of billions of dollars, not months.
What does this mean for businesses and buyers?
- Expect higher prices for memory-heavy products while AI demand keeps HBM and DRAM tight. If you need a laptop or workstation with lots of RAM, buying sooner has been cheaper than waiting for most of 2026.
- Watch lead times, not just prices. When one stage is constrained, deliveries slip even for products that do not use leading-edge chips.
- For product designers, single sources are a risk. Where you can, design with parts that have second sources, and check a component's expected production life before committing.
- Mature chips matter too. Most chips in cars, machines and appliances are made on older, cheaper processes. They depend on the same equipment and materials chain, so disruptions still reach them.
If you want to see how chip designers keep development costs down despite all this, I wrote about TSMC's shared prototyping runs in TSMC's 2nm CyberShuttle explained.
FAQ
Why can't other companies just build EUV machines?
EUV lithography took decades and a global network of specialist suppliers to develop, from ultra-precise mirrors to high-power light sources. ASML is the only company that completed that path, and a competitor would need years and enormous investment to catch up.
Why is TSMC so dominant?
It has led in manufacturing technology for several generations, invests heavily every year, and only manufactures chips for customers rather than competing with them by selling its own. That combination has made it the default choice for the most advanced designs.
Why did RAM and graphics card prices rise so much in 2026?
AI data centres are buying huge volumes of memory and GPUs. Memory makers have shifted capacity to high-bandwidth memory for AI, which squeezed the supply of ordinary DRAM and pushed up prices for PCs, laptops and graphics cards.
Will chip prices come down?
Some prices may ease as new memory and packaging capacity comes online, but new fabs take years to build. Analysts widely expect memory to stay tight into 2027. Treat any specific forecast with caution.
