ASIC servers are emerging as the most closely watched segment of the server supply chain heading into 2026. Nvidia remains the dominant force in AI servers and continues to ship the largest volumes of GPU-based systems. However, as Nvidia tightens its control over the supply chain through high server prices and increasingly standardized specifications, gross margins across the ecosystem are coming under pressure. Competition for ASIC server orders is intensifying as suppliers seek to protect and improve profitability.
According to DIGITIMES Research, shipments of high-end ASIC accelerators are projected to reach 5.135 million units in 2025 and 7.234 million units in 2026. While those figures remain below shipments of high-end GPU accelerators, estimated at 6.516 million units in 2025 and 7.988 million units in 2026, the growth trajectory differs sharply.
DIGITIMES Research estimates that ASIC accelerator shipments will post annual growth of more than 40% in both 2025 and 2026, compared with just over 20% growth for GPU accelerators over the same period.
The gap is expected to widen further. GPU server shipment growth is forecast to slow from 29.6% in 2025 to 22.6% in 2026, while ASIC shipment growth continues to accelerate. DIGITIMES Research concludes that stronger growth in high-end AI ASIC shipments from 2025 to 2026 will become the next major driver of expansion across the AI server supply chain.
Push and pull reshape supply chain dynamics
These projections align with current sentiment across the supply chain. On the pull side, faster growth and greater appeal are drawing suppliers toward ASIC opportunities. On the push side, GPU servers led by Nvidia continue to ship in large volumes, but margin compression is intensifying. High GPU server pricing is weighing on profitability, prompting supply chain players to more aggressively target the ASIC market in 2026.
Pressure is expected to increase with Nvidia's next-generation Vera Rubin architecture servers, scheduled for launch in the second half of 2026. The platform is expected to adopt more standardized designs, further limiting the number of suppliers across both mechanical components and assembly. Whether a supplier is included or excluded, the signal is clear. Those left out face constrained revenue growth, while those included must contend with execution risk and continued margin pressure.
Standardization tightens margins
Nvidia CEO Jensen Huang said during his CES 2026 keynote that a single MGX server rack can contain as many as 80,000 individual components, making annual redesigns unnecessary. Standardization, he said, is intended to ease the burden on suppliers and support efficient mass production.
For the supply chain, however, standardized component specifications reduce design differentiation and push suppliers toward contract manufacturing roles with diminishing value. Although Nvidia's GPU servers remain the backbone of the AI server market, large volumes also strengthen customers' bargaining power. One supplier said it prefers not to be the primary supplier, citing concerns that margins would be eroded too aggressively.
Assembly concentration adds pricing pressure
Beyond component standardization, Nvidia is expected to further concentrate large cloud service provider assembly orders for its next-generation Vera Rubin platform. Three ODMs—Wistron, Foxconn, and Quanta—are expected to handle L10-level shipments, with other suppliers taking on L11 rack assembly or L12 cluster integration. This centralized approach is intended to streamline logistics and clarify responsibilities.
Greater concentration and scale also invite more aggressive price pressure from Nvidia. This dynamic is not unique to Nvidia but reflects a broader industry pattern in which larger volumes lead to tougher pricing negotiations. For ODMs, the challenge is compounded by buy-and-sell shipment models that inflate reported revenue while delivering limited profit growth, resulting in a visible decline in gross margins.
Revenue surges mask margin erosion
Both Wistron and Wiwynn posted record revenue in 2025. Wistron surpassed Quanta to become the world's second-largest ODM and EMS provider after Foxconn, with annual revenue reaching NT$2.187 trillion (US$69 billion), more than double the previous year. Wiwynn's 2025 revenue also doubled to NT$950.663 billion.
Margins, however, moved in the opposite direction. Wistron's gross margin fell to 6.12% in 2025, down 1.89pp year over year, while its operating margin declined by 0.12pp. Wiwynn's gross margin dropped to 8.3%, down 2.07pp, and its operating margin fell to 6.7%, down 1.09pp. In both cases, the decline was attributed to changes in product mix.
Wistron's margin pressure reflected a sharp increase in full-rack AI server orders from US brand customers, as well as the consolidation of Wiwynn's revenue and gross profit. Wiwynn's margin decline was driven by rising shipments of Nvidia GPU servers.
Article edited by Jerry Chen
