Intel Korea Seeks CPU Priority for Local Server Makers on Tight 2027 Supply

Intel Korea launched the K-AI Full Stack Alliance with 23 initial participants at its Intel AI Summit Seoul at COEX on Nov. 12, and its president said the company is discussing priority CPU supply for the Korean server makers in it because supply will stay tight in 2027.

Bae Tae-won, president of Intel Korea, told the summit's media Q&A that server CPU supply is expected to stay difficult next year, according to Newspim and Newsworks. He said Intel Korea is working closely with its headquarters on how to prioritize demand and supply for the domestic server makers in the alliance, and is also talking with the government. No volumes, allocation shares or delivery dates were given.

The alliance is centered on domestically manufactured servers built on Intel processors, TheElec reported. Initial participants named across Korean reports include the server maker KTNF, AI chip firms FuriosaAI and DEEPX, CXL memory firms Panmnesia and Exina, and AI software and cloud firms Upstage, Nota AI, Lablup, Megazone Cloud, Hancom, Polaris Office and Selvas AI, plus Dnotitia, Metanet T Platform, Vuno and JLK. Selvas AI and its affiliate Mediana said separately on Nov. 12 that they had joined.

Why CPU supply is the story

Intel's Anil Nanduri, head of AI products and go-to-market for its data center AI group, said agentic AI shifts more work onto CPUs: sandboxing, tool calls and task execution. He told reporters that data center CPU demand could grow to about five times today's level within a few years, according to Newspim and Newsworks. If that holds, server availability for Korean buyers depends on allocation as well as price.

This fits what Intel has told investors. As Voltgrid reported, Intel's CFO said Oregon 18A wafer capacity is being swapped for Intel 3 data center output, and the company also plans for about 10% fewer PCs in 2027 as memory limits demand. In Seoul, Intel Korea applied the same supply squeeze to a named group of server customers.

What Intel showed in Seoul

Intel also presented its inference GPU Crescent Island, which inews24 said supports up to 480 GB of LPDDR5X memory at 350 W. It said Intel will build a reference architecture for autonomous factories with Daim Technology, and signed a memorandum of understanding with Seoul AI Hub to help Korean AI startups develop and test products. Bae said a recent financial AI project with Mesacure Company and NongHyup Bank runs agentic AI locally on Intel's B70 graphics card.

Korea's science ministry said in a video address that domestic AI chips entered mass production and commercialization this year, inews24 reported. The government is pushing heterogeneous AI infrastructure that mixes CPUs, GPUs and domestic NPUs, Newspim reported.

What is not confirmed yet

Intel has not published server counts, CPU volumes, prices or deployment dates for the alliance, and no partner has disclosed an order. Intel's own release was not found, so the account rests on the Korean outlets cited here. Reports differ on the date (TheElec's English article says Nov. 11, Korean outlets say Nov. 12) and on the roster; Newspim counts Intel among the 23. Newspim and Newsworks also differ on the size of the demand figure they quote for Nanduri, so it is left out. The priority arrangement is under discussion, not agreed.