How an AI Chip Is Made: From Design to Fab to Data Center
An AI chip is designed by one company, built on silicon wafers by another, joined to stacked memory in a packaging step and tested before it reaches a server rack. Here is each stage.
// chips & hardware
News and explainers on AI accelerators and the chip industry behind them: Nvidia, AMD and custom silicon, TSMC and rival foundries, advanced packaging and the memory shortage.
This section follows the chips that run artificial intelligence and the industry that builds them. That means AI accelerators, the specialized processors used to train and run AI models, from Nvidia, AMD and designers of custom silicon. It also covers the manufacturers that turn those designs into silicon, including TSMC, Samsung, Intel and Japan's Rapidus. Coverage extends to the advanced packaging that joins processors to memory, and to the high-bandwidth memory (HBM) at the center of a wider memory shortage.
The sums involved have grown quickly. World Semiconductor Trade Statistics, an industry statistics body, reported that global semiconductor sales reached $702 billion in the first half of 2026. It said that was up 102 percent from a year earlier, with memory sales up 305 percent. Nvidia reported revenue of $96.2 billion for its quarter ended July 26, 2026. It said $89.0 billion of that came from data center products.
Production is concentrated in a few companies. TrendForce is a market research firm. It estimated that TSMC held 72.5 percent of the contract chipmaking market by revenue in the second quarter of 2026, the first in which its 2-nanometer process contributed sales. In memory, the three biggest manufacturers are Samsung, SK hynix and Micron, according to IDC. TrendForce projected on September 30, 2026 that conventional DRAM contract prices would rise a further 10 to 15 percent in the fourth quarter. Explainers here focus on how the technology and its supply chain work, and on what the numbers mean for buyers.