Samsung Electronics gave investors a reminder on Friday that the artificial-intelligence trade is no longer only about memory chips. It is also about who gets to manufacture the custom silicon that the largest AI developers may need as compute costs keep rising.
South Korean shares staged a sharp rebound on July 3 after reports that Anthropic, the AI company behind Claude, is in early talks with Samsung to develop and manufacture a custom AI chip. The Korea Composite Stock Price Index closed up 5.76% at 8,088.34, according to Korea Exchange market data. Samsung Electronics rose 8.22% to 309,500 won, while SK Hynix gained 10.88% to 2,425,000 won, recovering part of the prior session’s steep selloff in Korean chip stocks.
The rally should not be read as confirmation of a signed contract. The Information reported that Anthropic is discussing a potential chip project with Samsung, with attention on Samsung’s 2-nanometer process and advanced packaging capabilities. Other reports described the discussions as preliminary, with the chip design, testing and manufacturing path still unresolved. That distinction matters. A prospective customer conversation is not the same thing as revenue, volume commitments or yield-proven production.
Even so, the market reaction says something important about what investors are rewarding. Samsung is already one of the world’s most important memory suppliers, but its foundry business has faced a tougher credibility test. Taiwan Semiconductor Manufacturing Co. has dominated leading-edge outsourced chipmaking, while Samsung has had to convince customers that it can pair advanced process technology with reliable execution at scale. A serious Anthropic project, if it progresses, would give Samsung a higher-profile opportunity in AI application-specific chips, not just in the memory that surrounds them.
That is why the story landed with such force in Seoul. The previous session’s slide had exposed the fragility of a market increasingly concentrated in a handful of AI-linked names. Samsung and SK Hynix are central to the Kospi’s performance, and their sharp swings can quickly turn a national equity benchmark into a proxy for global memory pricing, AI server spending and investor appetite for high-beta technology exposure. Friday’s rebound did not erase that concentration risk. It underlined it.
The financial logic behind custom AI chips is clear enough. AI developers are spending heavily on computing capacity, and reliance on off-the-shelf accelerators leaves them exposed to supply constraints, vendor pricing and power-efficiency limits. In-house or semi-custom silicon can help large model companies tailor performance to their own workloads. For a company such as Anthropic, exploring a custom chip does not necessarily mean replacing the existing AI hardware ecosystem. It may instead be a way to diversify future capacity and gain more control over long-term infrastructure costs.
For Samsung, the prize would be broader than one customer. Advanced AI chips require not only leading-edge wafers but also packaging that can move data quickly between logic and memory. Samsung’s ability to offer memory, foundry and packaging capabilities gives it a strategic pitch that few companies can match. The investor question is whether that breadth can be converted into manufacturing wins with the kind of customers that validate a foundry road map.
There are still reasons for caution. The reported Anthropic discussions may not result in a final deal. A custom chip can take years to design, qualify and deploy. Even if Samsung wins work, margins will depend on yields, pricing, capacity allocation and whether the customer ultimately orders at scale. Investors also need to separate the memory cycle from the foundry opportunity. A stock can be lifted by both narratives at once, but the economics are not identical.
Friday’s move therefore looks less like a verdict than a test. Korea’s chip leaders remain powerful beneficiaries of the AI buildout, but their market valuations now depend on more than near-term memory demand. Samsung’s reported Anthropic talks put the harder question in front of investors: whether the company can turn technological ambition into a foundry franchise that matters to the next generation of AI infrastructure.