AI Server Racks Ramp Into 2027 as ODMs Hit 15 to 20 Percent Component Shortages

AI server original design manufacturers are running into widening component mismatches, with shortages reaching 15 to 20 percent, as Nvidia's Vera Rubin platform enters its production ramp, according to DIGITIMES. Server ODM shipments are still expected to ramp further and the group delivered strong operating results in August, but the constraint has shifted from demand to the availability of specific parts needed to complete a rack. For buyers, that means lead times and configuration risk now matter as much as price.

Wiwynn is a clear example of the volume ramp. The company reported August revenue of NT$144.312 billion, about 4.53 billion dollars, up 22.62 percent from July and 50.36 percent from a year earlier, a third straight monthly record, with January to August revenue up 43 percent to NT$816.658 billion. Management said AI server shipments will overtake general-purpose server shipments in the third quarter of 2026, with the two accounting for roughly half each, and expects an even stronger fourth quarter as ASIC momentum builds. Rising average selling prices for AI servers should make revenue growth outpace shipment growth.

Component strategy is changing alongside the ramp. Wiwynn said some customers agreed in April to move to a memory procurement model handled by the server maker as memory prices surged, a change that helped improve second-quarter gross margin. President William Lin said the company is not currently facing a memory shortage, but CPU supply remains tighter and constraints may not ease until the first quarter of 2027. On capacity, Wiwynn opened its first plant in El Paso, Texas, with senior executives from Amazon Web Services and Microsoft attending; the first factory was completed in March 2026 and began mass-producing server racks within six months, while a second plant is scheduled to start mass production in 2027. The company is also expanding in Mexico.

The mix is shifting as well. Lin puts GPUs at about 70 percent of the AI server market and ASICs at roughly 27 to 28 percent, and expects GPU server growth to slow to 18 to 19 percent while ASIC growth could reach 35 percent, on the view that as AI moves from generative to agentic workloads more jobs will not require GPUs. He expects AI server growth of more than 20 percent annually over the next three to five years. Rising anti-data-centre sentiment in the United States has not so far translated into meaningful resistance from customers or officials, which he links to the midterm election cycle and expects to fade.