Anthropic’s SK Hynix Chip Request Sits Alongside a Korean Data-Center LOI

Chey’s position makes his remarks important but also defines their limits. He is chairman of the industrial group connected to the prospective supplier, SK Hynix, rather than a spokesman for Anthropic. The underlying claim is therefore Anthropic’s reported request, relayed by Chey and reported through Bloomberg and Yonhap—not confirmation by the prospective customer.

SK hynix HBM4 package shown in the company’s HBM4 development and mass-production announcement. Source: SK hynix.

Existing supply ties narrow the implication

At the summit, Amodei said Anthropic had already signed supply deals with Samsung Electronics and SK Hynix, without disclosing their terms.

That existing relationship, together with the spokesman’s description of a range of chips, is the strongest check on the notion that a possible custom processor would free Anthropic from the established AI-hardware supply chain. It could be another route to capacity or performance for particular workloads, but the reporting does not say it would replace other chips or suppliers.

SK Hynix’s role is also more specific than “chips.” It is a memory supplier. High-bandwidth memory, or HBM, vertically connects multiple DRAM chips to raise data-processing speed versus conventional DRAM. The company said in a September 2025 announcement that it had completed HBM4 development and prepared mass production; those are company claims, not evidence that Anthropic has reserved any HBM4 volume. A custom processor still needs memory, packaging and deployed systems around it.

A Korean supply-chain push, not an Anthropic-only event

President Lee hosted the San Francisco summit while promoting South Korea as a global AI supply-chain center. He met separately with Amodei as well as the chief executives of Nvidia, OpenAI and Broadcom; roughly 150 executives, investors, researchers and startup founders attended, the summit account says.

The event’s wider deals put the Anthropic discussion in perspective. SK Group and Nvidia unveiled a $500 billion-plus AI-infrastructure initiative that includes a planned 2-gigawatt SK Telecom data center using Vera Rubin chips and SK Hynix HBM4, due online in 2027. That figure belongs to the Nvidia-SK initiative, not to Anthropic’s component request or its data-center LOI.

Chey also said customers including Nvidia, Anthropic, OpenAI and Broadcom were asking for far more memory than expected. He described Anthropic as checking future chip availability, while Broadcom’s Hock Tan had said demand could exceed available supply. Those remarks are not a measure of Anthropic’s orders, but they explain why a conversation with a memory supplier and an infrastructure partner can matter even without a confirmed in-house processor.

What would turn the report into a chip strategy

The decisive evidence would separate three still-unproven layers: whether Anthropic is designing a chip; who would manufacture and package it; and whether the design has secured memory and data-center capacity on commercial terms.

For the SK Telecom letter of intent, the missing evidence is equally practical: a site, power allocation, financing, capacity, timetable and the role Anthropic would play. Until either company supplies those details, the record supports a reported component inquiry and an early infrastructure relationship—not a completed custom-chip program or guaranteed computing capacity.