Global AI Server Shipments to Top 2.5 Million Units in 2026; Quanta, Foxconn and Wiwynn Enter Top Five Suppliers

Global AI server shipments are forecast to exceed 2.5 million units in 2026, with HBM-equipped high-end AI servers reaching 2.37 million units, up 56% year-over-year and surpassing 15% of total server shipments, according to a DIGITIMES report. Taiwan's Quanta (2382-TW), Foxconn (2317-TW) and Wiwynn (6669-TW) are all expected to rank among the top five suppliers. Analyst Peter Hsiao said smoother shipments of Nvidia's (NVDA-US) GB300 NVL72 will lift the GPU-based share of high-end AI servers to nearly 60% in 2026. Nvidia solutions will account for more than half of XPU vendor share, while Broadcom's (AVGO-US) share falls to 29% as Google (GOOG-US) shifts orders and other customers ramp slowly. NVL72 racks are projected to rise to 42% of AI rack shipments from 24% in 2025, overtaking TPU-based Fish racks. Among buyers, North American CSPs' share of high-end AI server purchases will fall to 68% from 78%, while Neo Cloud buyers climb to 17%. Google and Amazon rank first and second among the top 10 customers; Oracle (ORCL-US) and SpaceXAI rise to third and fourth, ahead of Microsoft (MSFT-US). Anthropic enters the ranking for the first time in eighth with 135,000 units, and Nebius grows fourfold. DIGITIMES expects the top four supplier rankings to hold: Google+Celestica, Quanta and Foxconn lead, with Dell (DELL-US) fourth at nearly 300,000 units, up more than 1.62 times year-over-year. Wiwynn rises to fifth on cloud customer growth, while Supermicro (SMCI-US) slips to sixth.