The Chip Chain: How the Semiconductor Supply Chain Fits Together
Most advanced chips pass through a short chain of companies. Apple, Nvidia, AMD and Broadcom design chips but don't build them. TSMC, based in Taiwan, builds them on machines that only ASML, in the Netherlands, can make. Memory from makers like Micron sits beside the chips. Each link depends on the next.
The chain starts with a drawing. Apple, Nvidia, AMD and Broadcom spend years designing chips, then pay someone else to build them. A chip company with no factory of its own is called "fabless" ("fab" is industry slang for a chip factory).
Their own annual reports say so. Nvidia uses outside chip factories "such as" TSMC and Samsung to make its chips. AMD has TSMC make every processor and graphics chip built at 7 nanometers or smaller, and uses GlobalFoundries for older, larger designs. Broadcom sends most of its chip production to outside factories, TSMC among them.
Chips are printed with light, a bit like a photo being exposed onto film, only with features far smaller than a virus. The most advanced chips need EUV machines: machines that use extreme ultraviolet light to print the tiniest details. ASML calls itself the world's only maker of them (What is EUV?).
ASML can't build those machines alone. Carl Zeiss SMT in Germany makes the mirrors and optics inside every one, and Zeiss's factory output limits how many machines ASML can ship. See the Zeiss card on ASML's map.
A foundry is a factory that builds chips designed by other companies. TSMC is the biggest. It prints many chips at once on a wafer, a thin disc of silicon about the size of a dinner plate, which is later cut into individual chips.
TSMC is headquartered in Hsinchu, Taiwan, and most of its factories are in Taiwan. It is adding factories abroad: Apple says TSMC's plant in Arizona is producing tens of millions of chips for it, and that Apple is the plant's first and largest customer.
An AI chip is useless without fast memory right beside it. HBM (high-bandwidth memory) is a stack of memory chips placed next to the processor so data moves quickly (What is HBM?). Nvidia says it buys memory from SK Hynix, Micron and Samsung.
Packaging is the step that joins the chip and its memory into one finished part. Nvidia's annual report names CoWoS, TSMC's packaging method, as the one it uses. So for Nvidia's AI chips, TSMC both builds the chip and packages it.
Finished chips go into servers, phones and cars. Nvidia uses contract manufacturers such as Hon Hai (Foxconn), Wistron and Fabrinet for assembly and testing. Apple says its products are made mostly by partners in China mainland, India, Japan, South Korea, Taiwan and Vietnam.
At the end of the chain are the buyers. Microsoft, Amazon, Alphabet and Meta fill their data centers with AI chips to rent out computing power or to run their own apps. Tesla buys chips for its cars and its self-driving computers.
The links that matter most, with the numbers the companies themselves report.
Nvidia relies on TSMC, and TSMC's biggest customer is Nvidia
Nvidia has TSMC make its chips, with Samsung also named in its annual report. In 2025 TSMC's largest customer paid it 19% of its revenue and the second largest 17%. TSMC doesn't name them; reports identify the largest as Nvidia, which passed Apple that year.
Apple was TSMC's biggest customer for about a decade before Nvidia overtook it. Apple now also gets tens of millions of chips from TSMC's Arizona plant, where it is the first and largest customer.
AMD's processors and graphics chips at 7 nanometers or smaller all come from TSMC. Broadcom outsources most of its chip production, with TSMC named as one of its factories. Together with Nvidia and Apple, that puts four of the seven chip maps on the same factory.
TSMC can't make its most advanced chips without EUV machines, and ASML is the only company that makes them. ASML in turn has a single supplier for the optics in those machines, Carl Zeiss SMT.
Nvidia buys its memory from SK Hynix, Micron and Samsung. Micron's own sales are spread more widely than Nvidia's: no single customer reached 10% of Micron's revenue in fiscal 2026, down from one customer at 17% the year before.
In fiscal 2026, one direct customer paid Nvidia 22% of its $215.9 billion revenue and another paid 14%. That's 36% from two buyers, so a change in what a handful of cloud companies spend shows up quickly in Nvidia's results.
High-performance computing, which covers AI and data center chips, made up 58% of TSMC's 2025 revenue, up from 51% in 2024. Smartphone chips fell to 29% from 35%. TSMC's future now leans more on Nvidia, AMD and Broadcom than on phones.
A chain is only as strong as the links with no backup. Two kinds matter here: supplier concentration (one company or one place is the only option) and customer concentration (a few buyers pay most of the bills).
Only one company makes EUV machines, and only one makes their optics
If ASML can't ship machines, no chip factory can add capacity for the most advanced chips. If Zeiss can't make optics, ASML can't ship. There is no second source for either today.
Much of the most advanced chipmaking happens in Taiwan
TSMC's own annual report warns that most of its factories, and those of many of its suppliers and customers, are in areas exposed to natural disasters. It also notes that tension between China and Taiwan has weighed on Taiwanese companies in the past. Because Nvidia, Apple, AMD and Broadcom all use TSMC, one event there touches all four. See the Taiwan force on TSMC's map.
TSMC's ten largest customers made up about 78% of its 2025 revenue. Broadcom's five largest end customers were about 40% of its fiscal 2025 revenue. Nvidia's two largest direct customers were 36%. When a few buyers slow their orders, the effect runs back down the chain.
Export controls are government rules that say which chips and chipmaking machines can be sold to which countries. They touch every step of the chain at once.
Since August 2022, US rules have required licenses for Nvidia's A100 and H100 chips sold to China, and later rules widened the list. ASML can't sell its EUV machines to China, but Chinese customers still bought 29.1% of ASML's 2025 total net sales, mostly older machines. (Measured by system sales only, the figure is 33%, the number shown on ASML's map.) For the latest on each, see the outside forces on Nvidia's map and ASML's map.
Designers such as Apple, Nvidia, AMD and Broadcom draw the chips. A chip factory such as TSMC builds them, using machines from ASML. Memory makers supply the memory that sits beside each chip. Assembly companies put the finished products together, and cloud companies and device makers buy them. See the five steps
Who makes the machines that make chips?
For the most advanced chips, ASML in the Netherlands, which says it is the world's only maker of EUV machines. ASML depends on Carl Zeiss SMT in Germany for all of the optics inside them. ASML's map
Who makes the chips for Nvidia?
Mainly TSMC, with Samsung also named in Nvidia's annual report. Nvidia buys its memory from SK Hynix, Micron and Samsung, and uses Foxconn, Wistron and Fabrinet for assembly and testing. Nvidia's map
Who makes the chips for the iPhone?
Apple designs them and TSMC manufactures them. That includes tens of millions of chips at TSMC's Arizona plant, where Apple says it is the first and largest customer. Apple's map
What companies rely on TSMC?
Nvidia, Apple, AMD and Broadcom all have TSMC make their chips, according to their own filings. TSMC's ten largest customers made up about 78% of its 2025 revenue. TSMC's map
Where are most AI chips made?
Nvidia, AMD and Broadcom all have TSMC make their AI chips, and most of TSMC's factories are in Taiwan. TSMC has started making some chips in Arizona as well. TSMC's map
Why is Taiwan important for chips?
TSMC, the biggest maker of other companies' chips, is based there. So an earthquake, a power cut or rising tension near Taiwan could affect Nvidia, Apple, AMD and Broadcom at the same time. TSMC's outside forces
What are the biggest bottlenecks in the chip supply chain?
The places with only one supplier: ASML for EUV machines, and Zeiss for the optics inside them. AI memory has also been in short supply, which limits how many AI chips can be finished. Micron's map covers the memory shortage. Where the chain is thinnest
Figures come from the companies' own annual reports and releases where possible, and were last checked on Nov 21, 2026. Shares change every year, so check the dates. How we check connections.
StockTies is for information only. It is not investment advice. Spot something out of date? Tell us.