SK Group Chairman Says Anthropic Wants Its Own Chip Supply

When Chey Tae‑won took the stage beside Anthropic CEO Dario Amodei at an AI gathering in San Francisco, he didn’t just shake hands – he dropped a surprising detail. The South Korean mogul, who runs the conglomerate that owns SK Hynix, said Anthropic has formally asked the chipmaker for a steady stream of memory supplies to build its own custom silicon.

“It’s remarkable that an AI‑focused firm is already thinking about designing its own chips,” Chey said, pausing long enough for the audience to let the idea sink in. “If they need our memory, we’ll work with them.” The comment was part of a broader conversation about how AI startups are increasingly looking beyond off‑the‑shelf processors.

Anthropic, the research‑first AI lab best known for its Claude models, has kept its hardware strategy fairly hush‑hush until now. But the request to SK Hynix suggests the company is eyeing a tighter integration between software and hardware – a move that could boost performance while keeping costs in check.

The timing of Chey’s remark is interesting. It coincides with South Korean President Lee Jae‑Myung’s trip to Silicon Valley, where he hosted an AI summit that attracted the likes of Nvidia, Naver and several home‑grown SK Group units. Deals were announced on the fly – Nvidia teamed up with Naver to bring accelerated AI services to Korean users, and SK Group hinted at deeper collaborations across its semiconductor and telecom divisions.

For SK Hynix, the request from Anthropic is more than a sales opportunity; it’s a potential foothold in a market that could reshape the chip industry. Memory chips have traditionally been a commodity, but tying them to a specific AI model’s lifecycle could open new revenue streams and lock in long‑term partnerships.

Of course, the road to a custom chip is anything but smooth. Designing silicon that meets the unique demands of large‑scale language models requires massive R&D budgets, iterative testing, and a willingness to take on risk. Yet the fact that Anthropic is already seeking supply agreements hints that the company believes the payoff will be worth the gamble.

Whether this partnership will materialize into a full‑blown custom ASIC remains to be seen. What’s clear, however, is that AI developers are no longer content to rely solely on third‑party processors. The drive for specialized hardware is gaining momentum, and big‑chip manufacturers like SK Hynix are feeling the pressure to adapt.

As the AI summit in Silicon Valley wraps up, observers will be watching closely: will Anthropic’s chip ambitions spark a new wave of collaborations, or will they stay in the realm of talk? Only time – and perhaps a few silicon wafers – will tell.