Anthropic begins talks with Samsung to manufacture its first custom AI chip on 2nm process

Summary

Anthropic is in early-stage discussions with South Korea's Samsung Electronics to design and produce its first proprietary AI chip, according to a report from The Information published July 2, 2026, and confirmed by Bloomberg and TechCrunch. The chip would be manufactured using Samsung's advanced 2-nanometer Gate-All-Around foundry process and leverage Samsung's packaging facilities. Anthropic has not yet determined the chip's intended use, power specifications or server integration, making the project nascent rather than advanced. The company hired Clive Chan, an early member of OpenAI's custom chip team, as a signal of its intent to build dedicated hardware capacity. Samsung, SK Hynix and Micron all invested in Anthropic's US$65 billion Series H funding round in May, creating a financial alignment ahead of the manufacturing relationship. Anthropic's post-Series H valuation reached US$965 billion and the company confidentially filed for an IPO on June 1.

The split

US technology press covered the story primarily through the lens of Nvidia competition: Anthropic's chip, if realised, would reduce the company's dependence on Nvidia's H100 and B200 accelerators and give it direct control over inference cost. South Korean business media treated the talks as a strategic lift for Samsung Foundry, which has struggled to accelerate its 3nm ramp against TSMC's comparable offerings. No Chinese coverage was identified, which is notable given that any Anthropic custom chip manufactured at Samsung (rather than TSMC) sidesteps Taiwan Strait supply-chain risk from Beijing's perspective. Industry analysts quoted in the coverage uniformly noted that the gap between "talks about a chip" and a "deployed chip" is typically three to five years for a first-generation design.

By the numbers

• 2nm, Samsung process node under discussion, using Gate-All-Around transistor architecture
• US$65 billion, Anthropic Series H round in May 2026, in which Samsung, SK Hynix and Micron all invested
• US$965 billion, Anthropic's post-Series H valuation
• 3-5 years, typical time-to-production for a first-generation custom AI chip from initial discussions
• 4, major AI labs that have or are developing proprietary silicon: Google (TPU), Amazon (Trainium), OpenAI (Jalapeño/Broadcom), and now Anthropic

Why it matters

Every major frontier AI lab having its own chip is the structural endgame of the current AI infrastructure cycle. Proprietary silicon lets labs control inference cost, latency and data-center architecture without depending on Nvidia's roadmap or pricing. For Samsung Foundry, winning Anthropic's business would be one of the highest-profile client acquisitions since Apple chose TSMC for iPhone chips, and would validate Samsung's 2nm process as competitive for the most demanding AI workloads. For the broader compute supply chain, Anthropic's chip aligns with a long-term trend toward custom silicon that makes Nvidia's dominance of the inference market more contested.

What to watch

• Whether the talks progress to a formal design agreement, which would be a concrete signal that Anthropic is committed to in-house silicon
• The foundry competition: whether TSMC enters the picture as an alternative manufacturer, given its superior yield track record
• Anthropic's IPO process and whether custom chip development is cited as a cost-control strategy in the S-1
• Timeline to any first silicon: a 2nm chip moving from discussion to tapeout typically requires 18-24 months of design work before manufacturing begins