Nvidia Overtakes Apple as TSMC's Top Customer
Nvidia has officially surpassed Apple as TSMC's largest customer, now accounting for over 19% of the foundry's business. This move cements Nvidia's dominance in the global AI infrastructure buildout. To further secure its supply chain, TSMC is also in advanced talks with Nvidia to manufacture AI chips at its new Arizona facility.
For over a decade, Apple held the top spot as TSMC's primary customer, at times accounting for over a quarter of the foundry's revenue. This dynamic has shifted, with Nvidia's 2025 orders reaching an estimated $23.4 billion, more than double its 2024 spending. This surge reflects the high cost and complexity of AI accelerators, which are larger and more advanced than the processors used in consumer devices. The shift is evident in TSMC's revenue breakdown, where high-performance computing (HPC) now constitutes 57% of revenue, up from 30% in early 2020. During the same period, the smartphone segment's share dropped from 49% to 30%. This highlights the massive capital investment in the global AI data center buildout, a market expected to see a 40% compound annual growth rate in the coming years. Nvidia's demand is heavily focused on TSMC's most advanced process nodes, including 3nm, with a clear roadmap to adopt 2nm and 1.6nm technologies in the future. This close partnership on cutting-edge manufacturing solidifies TSMC's central role in the AI supply chain. The collaboration extends to U.S. soil, with Nvidia's Blackwell chips already in production at TSMC's new Arizona facility. This move is part of a larger Nvidia initiative to invest up to $500 billion in U.S.-based AI infrastructure over the next four years, partnering with companies like Foxconn and Amkor for supercomputer manufacturing and chip packaging in states like Texas and Arizona. Hyperscalers are also navigating a "build vs. buy" dilemma for their AI compute needs. While building their own data centers offers long-term control, leasing from specialized "neo-cloud" providers like CoreWeave allows them to deploy GPU capacity faster to meet immediate demand. This strategy helps them avoid losing enterprise customers while their own multi-year data center construction projects are underway. The competitive landscape for AI chips is diversifying. While Nvidia leads in GPUs, the market for custom ASICs (Application-Specific Integrated Circuits) is projected to grow at a 27% compound annual rate, reaching $118 billion by 2033. Companies like Broadcom and Marvell are key players, designing custom chips for giants like Google, Meta, Amazon, and Microsoft, who are increasingly investing in their own silicon to optimize performance and cost.
