TSMC Q3 revenue climbed to NT$1.49 trillion (approximately $48.4 billion) in the three months ending September, up 51% year over year and beating the NT$1.46 trillion consensus compiled by analysts. The chipmaker reported its unaudited monthly figures on October 8, underscoring how demand from artificial intelligence customers continues to outpace even bullish Wall Street expectations. September alone contributed NT$511.86 billion, a 54.6% jump from a year earlier, narrowly missing August’s all-time monthly record of NT$514.8 billion. The print sets the stage for TSMC’s formal quarterly detail and fourth-quarter guidance, scheduled for the October 15 investor call.
TSMC Q3 revenue: Monthly Cadence Confirms Sustained AI Pull-Through
TSMC reports revenue on a monthly basis in Taiwan dollars, giving investors an early read on quarterly trends long before the official filing. The September figure reinforced that the second half of 2026 is tracking well above prior guidance, with the third-quarter aggregate easily clearing the high end of management’s own range. The 54.6% September year-over-year acceleration matched the July-August pattern, suggesting that capacity utilization remains tight across nearly every process node relevant to AI workloads. With CoWoS advanced packaging capacity fully allocated, TSMC is continuing to set shipment priorities rather than negotiate order volume, a posture that few foundries in the industry have been able to maintain for more than three consecutive quarters.
TSMC Q3 revenue: Nvidia Captures 60% of CoWoS, Toppling Apple as Largest Customer
The most significant structural change disclosed through the monthly data is the reshuffling of TSMC’s customer mix at the high end of its product stack. Nvidia is now estimated to consume roughly 60% of available CoWoS-L and CoWoS-S packaging throughput, securing enough advanced-packaging capacity to overtake Apple and become TSMC’s largest single customer by revenue. That milestone reflects how accelerated compute platforms, including the Blackwell and next-generation Rubin architectures, have come to dominate the foundry’s premium output. Apple’s reliance on TSMC for its A-series and M-series chips remains substantial, but its share of the most cutting-edge capacity has narrowed as Nvidia secured long-term wafer and packaging reservations. Analysts at multiple sell-side shops have already raised their 2026 and 2027 CoWoS forecast models on the back of the disclosure.
TSMC Q3 revenue: 2nm Wafer Pricing Climbs to $30,000 as 2028 Is Booked Out
TSMC’s most advanced nodes are now effectively sold out through 2028, according to industry sourcing familiar with reservation contracts. Early N2 wafers are reportedly changing hands at approximately $30,000 each, a meaningful premium to the $18,000-$20,000 price band commanded by N3 wafers during their equivalent early production phase. Apple is understood to have locked in roughly half of the initial 2nm output to power the A20 family in the iPhone 18 generation, with the remainder split among Nvidia, Qualcomm, AMD, and Broadcom. Pricing power at the leading edge remains in TSMC’s hands because no other foundry has yet demonstrated commercially viable 2nm gate-all-around production at comparable yield. Customers appear willing to absorb the higher per-die cost in exchange for guaranteed delivery slots and energy-efficiency gains.
High-Performance Computing Now 61% of Revenue Mix
The product-mix shift toward AI silicon is now mathematically visible in TSMC’s reported segment weights. High-performance computing, which includes data-center accelerators, networking chips, and AI training silicon, contributed 61% of total Q3 revenue, up from 53% in the year-ago period. Smartphone-related revenue fell to roughly the high-20s as a share, reflecting both Apple’s relative weighting and softer handset unit growth globally. The remaining slice was split across IoT, automotive, and digital consumer applications, each of which posted modest sequential gains but no breakout growth. Management has signaled that HPC’s share will likely climb further in 2027 as 2nm ramp combines with CoWoS expansion in Hsinchu, Kaohsiung, and Arizona.
Outlook: Pricing Discipline and Capacity Discipline Both Intact
With full-year 2026 revenue on track for well above the 30% growth range TSMC guided at the start of the year, attention is increasingly focused on gross-margin trajectory and capex absorption. Pricing discipline has held because the leading-edge node mix has expanded faster than the installed wafer base, letting each new shipment carry a higher contribution margin. Capacity discipline has held because TSMC has resisted pressure to convert mature-node fabs into advanced packaging lines at the expense of existing commitments. The October 15 investor call is expected to formalize fourth-quarter revenue guidance, refresh capex, and provide clarity on 2nm yield curves and CoWoS annual capacity through 2028. For now, the data confirms that TSMC Q3 revenue was not simply a print but a structural signal that the AI-driven order book is still accelerating into 2027.
Nvidia’s commanding grip on CoWoS capacity has direct implications for TSMC Q3 revenue sustainability, because advanced-packaging supply rather than wafer output is the binding constraint for AI-accelerator shipments. With Nvidia absorbing roughly 60% of available CoWoS lines, foundry customers competing for the remaining throughput — including AMD’s MI350 family, Broadcom’s custom ASIC partners, and several hyperscalers building in-house inference silicon — have been forced to redesign die layouts for larger interposers or accept longer lead times. Internal capacity expansion plans at Tainan’s AP6 and AP8 modules are expected to add meaningful incremental wafers by late 2026, but in the near term, allocation remains the primary lever shaping who can ship and at what margin.
Industry-watchers also note that TSMC’s stated capex trajectory is being underwritten by multi-quarter prepayments and volume commitments from its top customers, a structural departure from the cycle of just-in-time orders that defined prior semiconductor upturns. Officials have signaled that 2026 capex could approach the upper end of an already elevated range should Nvidia and adjacent AI clients convert their demand visibility into formal capacity reservations, effectively turning TSMC Q3 revenue strength into a forward-looking capital-build program rather than a spot-market phenomenon.
Source: Startup Fortune