SK Group Chair Discloses Anthropic’s Bid for Custom Chip Supplies

When the lights dimmed at a packed AI conference in San Francisco, the conversation drifted from model scaling to something far more tangible – silicon. Chey Tae‑won, the veteran chairman of South Korea’s SK Group, took the stage beside Anthropic chief executive Dario Amodei and a handful of tech leaders. Between jokes about coffee‑fuelled coding marathons, Chey dropped a surprise: Anthropic has formally requested chip supplies from SK Hynix, the memory‑chip titan that sits under his corporate umbrella.

It’s not every day an AI‑first startup announces ambitions to design its own silicon. Most of today’s AI labs still lean on off‑the‑shelf GPUs or partner with foundries that handle everything from wafer to package. But Anthropic, the creator of Claude‑series chatbots, apparently wants a tighter grip on the hardware that powers its models. The request, Chey said, is for SK Hynix to allocate a portion of its high‑bandwidth memory production to Anthropic’s upcoming custom ASICs.

“It’s remarkable, really,” Chey mused, his tone half‑admiring, half‑pragmatic. “An AI developer thinking about its own chip supply chain signals a maturing industry.” He added that the talks are still early, and no contracts have been signed yet, but the interest alone marks a shift in how AI firms view hardware strategy.

The timing of the announcement dovetails with a broader diplomatic and commercial push from Seoul. South Korean President Lee Jae‑Myung landed in Silicon Valley just days earlier, convening an AI summit that attracted heavyweights from Nvidia, Naver and, of course, SK Group. In that forum, Nvidia pledged collaborations with Naver and SK Group on next‑gen AI infrastructure, underscoring South Korea’s desire to be a central node in the global AI supply chain.

Industry observers note that Anthropic’s request could have ripple effects. If the startup secures a steady flow of SK Hynix memory, it may accelerate its chip‑design roadmap, potentially challenging the dominance of larger players like Nvidia and Google’s Tensor‑flow chips. For SK Hynix, the deal would diversify its customer base beyond traditional memory‑intensive devices, bringing AI‑centric revenue into its portfolio.

Of course, there are practical hurdles. Designing a custom AI processor is a multi‑year, capital‑intensive endeavor, and integrating SK Hynix’s memory products into a novel architecture requires close engineering collaboration. Yet the mere fact that Anthropic is reaching out suggests confidence in SK Hynix’s manufacturing prowess and the strategic alignment between the two companies.

As the AI conference wound down, Chey and Amodei exchanged a brief handshake – a symbolic nod to a partnership that is still in its infancy but could reshape the hardware landscape for next‑generation language models. Whether this dialogue matures into a formal supply agreement remains to be seen, but the conversation itself signals that the era of AI‑only software startups may be giving way to a new breed of vertically‑integrated players.