Taiwan’s AI server makers delivered exceptional results in the first quarter of 2026, driven by surging shipments of Nvidia’s GB300 (Blackwell) servers to major hyperscale cloud providers, including Amazon, Google, Meta, Microsoft, and Oracle. These figures underscore that AI server assembly has emerged as the dominant growth engine in Taiwan’s technology sector.
Wistron led the charge with consolidated Q1 revenue of NT$846.3 billion (US$26.5 billion), soaring 144.3% year-on-year. This remarkable pace highlights the company’s expanding role as a key assembler of Nvidia’s next-generation rack-scale systems. Quanta Computer followed closely with NT$809.2 billion (US$25.3 billion), up 66.6%, while cloud specialist Wiwynn reported NT$276.5 billion (US$8.7 billion), a 62.0% increase. Inventec rounded out the major players at NT$200.3 billion (US$6.3 billion), growing 27.6%.
Foxconn, the world’s largest electronics manufacturer, posted Q1 revenue of NT$2.12 trillion (US$66.4 billion), up 29.7% year-on-year, with AI cloud servers serving as the primary driver. March revenues alone hit a record NT$803.7 billion (US$25.1 billion), surging 45.6% year-on-year, fueled by sustained AI infrastructure demand and post-Lunar New Year restocking. Looking ahead, Foxconn anticipates sequential and year-on-year revenue growth in Q2 2026, traditionally a slower season for ICT, while noting that its AI rack business should continue its strong trajectory, albeit amid ongoing geopolitical and economic uncertainties.
March figures for Wistron and Quanta were equally impressive. Wistron achieved NT$333.0 billion (US$10.4 billion) for the month, up 117.7% year-on-year, and Quanta reached NT$362.8 billion (US$11.4 billion), jumping 88.4%. These peaks reflect both seasonal recovery and accelerated GB300 shipments in anticipation of broader Q2 datacenter deployments.
For Taiwan’s server assemblers, this shift brings important commercial advantages. Inference orders deliver sustained demand visibility and multi-quarter commitments from hyperscalers, replacing the more lumpy, project-based cycles typical of training infrastructure. As a result, companies now enjoy longer planning horizons and more predictable revenue streams, even as they continue supporting training GPU deployments.
Yet this explosive growth is increasingly constrained by bottlenecks across the supply chain.
At the foundry level, TSMC’s 3nm process remains the world’s most oversubscribed advanced node, with Nvidia, AMD, Broadcom, Apple, and MediaTek all competing fiercely for capacity. The scarcity is severe enough to prompt some chipmakers to explore Samsung's competing process and to motivate Elon Musk's Terafab initiative, which aims to build captive chip manufacturing capacity in Austin for Tesla, SpaceX, and xAI.
Downstream, TSMC’s CoWoS advanced packaging continues to act as a critical second bottleneck, even for companies that have secured wafer allocations. Nvidia dominates CoWoS capacity, reportedly claiming over 50% of TSMC’s 2026 output to support both Blackwell and the upcoming Rubin platforms.
Further along the supply chain, ABF (Ajinomoto Build-up Film) substrate supply is tightening rapidly amid surging AI demand. Nvidia’s Blackwell GPU already requires roughly double the substrate area of its Hopper predecessor, owing to larger interposers and more complex multi-die designs. The forthcoming Rubin GPU is expected to drive another substantial increase in material intensity on the order of 75% or more relative to Blackwell.
Compounding these pressures is a persistent global shortage of T-glass fiberglass cloth, a critical low-CTE (Coefficient of Thermal Expansion) material essential for high-end substrates. Japan’s Nittobo commands an estimated 80-90% share of advanced grades, and the imbalance is projected to persist into the second half of 2027, despite ongoing capacity expansions at its Fukushima plant.
Consequently, CCL (Copper Clad Laminate) lead times have extended to 24 weeks or longer in key segments, while substrate prices continue rising 3–7% per quarter. Taiwan’s largest substrate maker, Unimicron, has warned that price hikes will accelerate further in Q2 2026 to offset upstream cost inflation, spurring customers to secure long-term agreements to lock in supply.
The AI server makers enter Q2 2026 with demand visibility extending well into 2027, but constrained at nearly every layer of the supply chain. From foundry capacity to advanced packaging, substrates, and raw materials, each link in the chain is operating at or near its limits. Their ability to navigate these bottlenecks will determine whether the current momentum can be sustained.