NVIDIA Overtakes Apple as TSMC’s Top Customer, Fueling World’s Largest Infrastructure

NVIDIA’s surge to become TSMC’s top revenue source reflects a broader shift in the semiconductor ecosystem toward AI‑centric workloads. According to Wccftech, the chipmaker now accounts for “more than 19 % of TSMC’s overall revenue,” a share that eclipses Apple’s long‑standing dominance in the foundry’s customer hierarchy. The uptick is driven by NVIDIA’s aggressive rollout of its Hopper‑based H100 GPUs and the upcoming Blackwell architecture, which demand the most advanced 5‑nm and 4‑nm process nodes that TSMC can supply at volume. Those nodes are critical for delivering the high transistor counts and power‑efficiency metrics required for large‑scale transformer training and inference, effectively turning NVIDIA into the primary engine of the world’s “largest AI infrastructure” as described by the outlet.

The scale of NVIDIA’s orders is reshaping TSMC’s capacity planning. Reuters reported that the Taiwanese foundry is projecting a 20 % revenue growth this year, explicitly attributing the boost to “AI demand.” In its Q4 outlook, TSMC expects a 27 % profit surge, with AI‑related business driving a “significant portion” of that increase. The company’s guidance includes a forecast of $22.4‑$23.2 billion in Q3 revenue, up from the prior quarter, and a Q2 net profit of T$247.8 billion—both figures surpassing market expectations. These numbers underscore how NVIDIA’s high‑volume, high‑margin GPU shipments are translating into fab‑level financial performance, effectively replacing the consumer‑electronics traffic that Apple historically supplied.

From a technical perspective, the transition is rooted in the differing silicon requirements of AI versus mobile computing. Apple’s A‑series and M‑series chips, while cutting‑edge, are optimized for power‑constrained, heterogeneous workloads that prioritize integrated graphics and CPU efficiency. In contrast, NVIDIA’s GPUs rely on massive parallelism, high‑bandwidth memory (HBM2e), and dense interconnects such as NVLink and PCIe 5.0 to sustain petaflop‑scale compute. To meet these specifications, TSMC must allocate a larger fraction of its most advanced process slots to NVIDIA, often employing its “chip‑on‑wafer‑level‑packaging” (CoWoS) and “InFO” technologies to stack dies and integrate HBM directly on the silicon. This shift also pressures TSMC’s supply chain, prompting the foundry to accelerate capital expenditures—Reuters notes TSMC’s capital spending is slated at roughly $2 billion for the year, aimed at expanding 3‑nm capacity and upgrading lithography equipment to keep pace with NVIDIA’s roadmap.

The strategic implications extend beyond pure volume. By securing NVIDIA as its premier customer, TSMC reinforces its position as the de‑facto backbone of the AI supercomputing market, a role that could attract further fabless AI startups seeking access to leading‑edge nodes. Analysts cited by Reuters have highlighted that TSMC’s refusal to enter a joint venture with U.S. partners signals confidence in its ability to service AI demand independently, leveraging its mature ecosystem of equipment suppliers and its deep expertise in advanced patterning. Moreover, the partnership may catalyze co‑development initiatives, such as custom process optimizations for NVIDIA’s upcoming tensor cores, which could further differentiate TSMC’s offering from rivals like Samsung and GlobalFoundries.

While NVIDIA’s ascendancy reshapes the revenue hierarchy, Apple remains a substantial contributor to TSMC’s overall business, especially in the 7‑nm and 5‑nm segments that power the iPhone 15 and M‑series Macs. However, the relative growth rates diverge sharply: Apple’s shipments are mature and predictable, whereas NVIDIA’s AI‑driven demand is expanding at double‑digit percentages year over year, as evidenced by the foundry’s 20 % revenue outlook. This divergence suggests that TSMC’s future earnings will be increasingly tied to the health of the AI market, making NVIDIA’s fortunes a bellwether for the entire semiconductor supply chain.