NVIDIA has quietly overtaken Apple to become TSMC's largest customer, marking a structural shift in the semiconductor giant's client base. TSMC posted consolidated revenue of NT$1.13 trillion (US$34.6–35.8 billion) for the first quarter of 2026, up 8.4% quarter-over-quarter and 35.1% year-over-year, within its prior guidance and an all-time high. March alone brought NT$415.2 billion, another monthly record.
AI Demand Defies Seasonal Lull
The first quarter is traditionally a slow season for semiconductors, but TSMC's sequential growth reached high single digits, proving AI demand can offset consumer seasonality. Monthly figures formed a V-shaped acceleration: January set a record at NT$401.3 billion, February eased to NT$317.7 billion due to Lunar New Year, then March surged to NT$415.2 billion as AI chip pull-ins accelerated.
Two parallel forces drove March's strength: NVIDIA's next-generation AI GPUs entering mass shipment, maxing out advanced process capacity; and ASIC custom chips from cloud providers scaling up alongside GPUs. Both demands converged on advanced nodes and advanced packaging, keeping TSMC's high-end fab utilization at elevated levels.
CoWoS: The Bottleneck That Measures Demand
CoWoS advanced packaging remains the tightest bottleneck in the AI chip supply chain. The technology integrates HBM memory with GPU dies on a single interposer, essential for NVIDIA's flagship AI accelerators and one of TSMC's least replaceable roles. TSMC continues expanding CoWoS capacity, but not fast enough to match end-demand growth. In a way, the severity of the CoWoS shortage itself serves as a gauge for AI server demand—only a truly hot market would create such supply tension.
What to Watch in Q2
Market consensus for TSMC's second quarter centers on sustained cloud capex and expanding AI server builds, with sequential growth potentially remaining in high single digits. But the real focus is on the April earnings call, where investors want clarity on: updated CoWoS expansion timelines; ASIC customer order visibility; and whether the rising concentration risk from NVIDIA is being managed. The numbers are strong; the question is how management frames the second half.