AI · Asia · Factories | A Chip’s Journey Across Asia: Who Makes It, Who Packages It and Who Installs It

By Jiuri in Tokyo

The raw material used to make semiconductor chips is ultra-high-purity silicon with a purity level of 99.9999999%. That silicon ultimately comes from sand. In that sense, the journey of an AI chip begins with sand.

But how far must a chip travel before sand is transformed into an AI server?

First Stop: Japan — Materials and Equipment

Turning sand into a silicon wafer requires multiple stages, including purification, crystal growth, slicing and polishing. Japan’s Shin-Etsu Chemical and SUMCO together account for approximately half of global silicon-wafer supplies.

A wafer, however, is only a blank canvas. To create nanometre-scale circuits on its surface, manufacturers need photoresists, ultra-high-purity chemicals, precision inspection instruments and other specialised “paints and brushes.” These are among Japan’s most important competitive strengths.

Of the 19 core materials used in global semiconductor manufacturing, Japanese companies rank first worldwide in 14 categories. In key equipment segments such as coating and developing systems, wafer dicing and precision cutting, Japanese manufacturers hold global market shares of between 70% and 90%.

Without these materials and machines, none of the production stages that follow could begin.

Second Stop: Taiwan — Advanced Packaging, the Final Assembly Plant for AI Chips

Once circuits have been formed on the wafer, the chip is still not ready for use. An AI processor does not operate alone. It must work together with graphics processing units and high-bandwidth memory chips stacked and connected within the same system.

This process of stacking and interconnecting chips is known as advanced packaging.

TSMC’s Chip-on-Wafer-on-Substrate, or CoWoS, technology has become one of the world’s most widely used packaging solutions for AI chips. If wafer fabrication is comparable to constructing the shell of a building, CoWoS packaging is the interior fit-out. It integrates the building—the computing chip—with its utilities, including memory chips and electrical connections, so that the finished product can be delivered directly to the customer.

By the end of 2026, TSMC’s monthly CoWoS capacity is expected to reach between 120,000 and 140,000 wafers. NVIDIA alone is expected to take more than half of that capacity. Nearly every NVIDIA AI chip must pass through TSMC’s advanced packaging lines before it can be deployed, with Broadcom and AMD among the other major customers competing for capacity.

TSMC does not merely manufacture individual chip components. Through advanced packaging, it assembles them into functioning AI processors, enabling computing chips and memory chips to operate as a single system. In that sense, it serves as the final assembly workshop for AI computing power.

Third Stop: South Korea — HBM, the Blood Vessels of AI Chips

No matter how quickly a processor can calculate, its performance is limited if data cannot move in and out fast enough. When large AI models are running, GPUs must continuously read and process enormous volumes of data. The speed of that data transfer directly affects the actual performance of an AI chip.

High-bandwidth memory, or HBM, is designed specifically for this purpose. Installed in close proximity to the GPU, stacked HBM chips transmit data over extremely short distances and at exceptionally high bandwidth.

South Korean companies Samsung Electronics and SK hynix dominate the global HBM market. In the first quarter of 2026, SK hynix ranked first with a 58% share, while Samsung Electronics and Micron each accounted for 21%. The two South Korean companies together controlled approximately 79% of the global HBM market.

In a written response issued on July 13, the Bank of Korea stated: “The semiconductor industry is continuing to expand, primarily because rapid investment in AI infrastructure is driving sustained growth in semiconductor demand, while the pace of supply expansion continues to lag behind demand.”

Without South Korean HBM, even the fastest AI processor would struggle to move data efficiently. HBM functions like the blood vessels of an AI chip, determining how quickly its “blood”—data—can circulate.

Fourth Stop: Malaysia — Packaging and Testing, the Chip’s Final Inspection

Even after final assembly, a chip must pass through one last stage before delivery.

Packaging and testing involves checking completed chips for functionality, identifying defective units, sorting qualified products and preparing them for shipment. Every chip leaving a production line must pass through this quality-control checkpoint. Defective chips are removed, while qualified products are packaged and sent to server assembly plants.

Malaysia holds approximately 13% of the global semiconductor packaging and testing market. Often described as the “Silicon Valley of the East,” the country is one of the world’s most important centres for back-end semiconductor production. International manufacturers including Intel and Infineon operate major packaging and testing facilities there.

Singapore is also one of the region’s strongest front-end semiconductor manufacturing centres, while Vietnam is rapidly expanding its role in packaging, testing and assembly.

If Taiwan performs the sophisticated interior fit-out, Malaysia carries out the final completion inspection. Before a chip is delivered, it must undergo one last round of testing there.

Fifth Stop: Mainland China — Server Assembly, the Chip’s Final Destination

Once a chip has passed inspection and left the factory, where does it go?

The answer is an AI server.

Chips are transported to server assembly plants across mainland China. On production lines in cities such as Chengdu and Shenzhen, robotic arms and human workers operate side by side, integrating GPUs, CPUs, memory modules, motherboards, cooling systems and power-supply units into complete AI servers.

China is forecast to account for between 15% and 20% of global AI server shipments in 2026.

A single AI server may contain only a few high-performance chips or more than one hundred, depending on its configuration. Once assembled, the servers are shipped overnight to data centres across China and overseas. There, they support the training, inference and operation of large AI models.

At that point, the chip’s journey comes to an end.

Sand has completed its transformation into silicon intelligence. From a bottle of Japanese photoresist to a South Korean HBM chip, from a wafer processed in Taiwan to a final round of packaging and testing in Malaysia, and ultimately to a server assembled in mainland China, every AI chip is the product of a cross-border industrial relay.

Commentary

From Contract Manufacturing to Indispensability

In the past, Asia was often described as the contract-manufacturing base of the global semiconductor industry, responsible mainly for lower-value production processes. That role is now being redefined.

Japan controls critical upstream materials and equipment. South Korea dominates memory production. Taiwan leads in wafer fabrication and advanced packaging. Malaysia specialises in semiconductor packaging and testing, while mainland China carries out server assembly and system integration.

Asia has developed the world’s most complete and concentrated semiconductor supply chain.

The figures demonstrate how indispensable the region has become. In 2026, the combined capital expenditure of Asian semiconductor manufacturers exceeded US$136 billion, representing year-on-year growth of 25%. By 2029, 64 new wafer fabrication plants are expected to begin operations across Asia. Global semiconductor-device industry revenue could exceed US$1.5 trillion.

Morgan Stanley has noted that value creation in the global semiconductor industry is shifting away from chip-design capability alone and towards constraints on physical production capacity.

Wafers, advanced packaging, high-performance substrates, power components, electricity supplies and even cleanroom space have become hard constraints on the industry’s expansion. Asia controls critical links in almost all of these areas.

Without Asia, an AI chip would remain little more than a pile of sand.

Asia has moved from being a contract manufacturer to becoming indispensable.

Next in the Series

In the next report, “If the AI Boom Cools, How Long Can Asia’s Factories Stay Hot?”, we will move beyond the current expansion to examine the outlook more critically: Can the boom continue? Where are the risks? And is Asia prepared?