Anthropic-Samsung Chip Talks Jolt Seoul; AI Capex Hits $400B

Reports that Anthropic is in talks with Samsung to co-develop a custom AI chip moved markets in Seoul on Friday, pulling semiconductor stocks sharply higher after a brutal two-day selloff, while fresh data on hyperscaler spending underlines just how much capital is now flowing into AI infrastructure.

Key points

• Anthropic is reportedly in talks with Samsung to develop a custom AI chip, aiming to reduce dependence on external suppliers including Nvidia.
• The news helped lift Samsung Electronics 6.12% and contributed to a 2.1% bounce in the KOSPI on Friday morning, following a 7.89% plunge in the prior session dragged by semiconductor selling.
• Microsoft, Google, Meta, and Amazon are guiding combined 2026 capital expenditures above $400 billion, up from roughly $230 billion in 2024, with most of that flowing through chip designers, foundries, and memory.
• Value-focused Vltava Fund disclosed it sold its Lam Research (LRCX) position in Q2 2026, citing overvaluation.
• Google’s AI agent integration at Walmart is drawing fresh attention to the broader infrastructure value chain behind agentic commerce, with five public companies identified as central to that stack.

Anthropic joins the custom-chip crowd. What does it mean for Nvidia?

Anthropic’s reported Samsung talks follow a well-worn path. OpenAI, Amazon, Microsoft, and Meta have each moved to develop proprietary AI silicon in recent years, primarily to manage costs and reduce exposure to supply constraints. For Anthropic, which relies heavily on cloud-provided compute, a custom chip partnership would be a significant strategic step.

Samsung is a natural partner for the work. Its foundry business and HBM memory capabilities position it as a one-stop option, and the company has clear motivation to win design wins that could cement its role in the AI supply chain beyond commodity memory. The Seoul market’s immediate reaction, a jump in Samsung shares following days of sharp losses tied partly to semiconductor sector anxiety, suggests investors read the news as a meaningful demand signal.

The longer-term read for Nvidia is murkier. Custom inference chips from AI labs tend to handle specific workloads and do not replace Nvidia GPUs across the board, at least not in the near term. But the direction of travel across the industry is consistent: large AI buyers want more control over their silicon, and every custom chip that reaches production at scale is one less GPU order.

Does the $400B capex figure change the semiconductor ETF calculus?

Three years into the AI infrastructure buildout, hyperscaler spending has carried semiconductors from a cyclical play to the highest-beta vehicle in the equity market, according to one recent analysis tracking the AI capex cycle. Combined 2026 guidance from Microsoft, Google, Meta, and Amazon now exceeds $400 billion, nearly double the 2024 figure of roughly $230 billion.

That backdrop is what makes moves like Vltava Fund’s LRCX exit worth watching. Vltava is a self-described value shop, and its Q2 letter frames the sale as a valuation call rather than a view on semiconductor demand. The fund’s letter also explores how AI is reshaping which forms of human intelligence carry economic value. A value manager walking away from a semiconductor equipment name at this point in the capex cycle is a data point, not a verdict, but it flags that some disciplined buyers think prices have run ahead of fundamentals even with the spending wave intact.

The Philadelphia Semiconductor Index fell 5.4% in the most recent U.S. session, extending a two-day decline, while the Nasdaq dropped 0.8%. The Dow, by contrast, hit a record high after weaker-than-expected June jobs data reduced near-term rate-hike fears. That divergence, broad market strength alongside semiconductor-specific weakness, is worth tracking as the Q2 earnings season approaches.

Who captures value when AI agents do the shopping?

A separate thread gaining attention today involves Google’s AI integration at Walmart’s checkout. When an AI agent browses, recommends, and completes a purchase on a shopper’s behalf, the retailer is one node in a longer value chain. Five public companies sit at the center of that infrastructure, though the source is careful to note they are not equally positioned. The specific names are not detailed in available summaries, but the framing reflects a broader investor question: as agentic AI handles more consumer transactions, does value accrue to the model provider, the cloud layer, the payment rail, or the retailer itself? That question does not have a settled answer yet, and investors pricing AI-commerce themes should keep it open.

Nothing in this article is investment advice. Independent research and professional guidance are always recommended before making investment decisions.