Anthropic is in early talks with Samsung about building its own AI chip, the clearest sign yet that no frontier lab wants to depend entirely on Nvidia anymore.
Anthropic has opened early talks with Samsung Electronics about building its own AI chip, and the target is Samsung's most advanced manufacturing line: the 2 nanometer process node, paired with its advanced packaging facilities. The Information first reported the discussions on July 2, and the story has since been confirmed by outlets including TechCrunch and Bloomberg. Nothing is close to finished. There is no prototype, no finalized specification, and no manufacturing timeline. Anthropic has not even decided what the chip is optimized for, how powerful it needs to be, or how it fits into a server rack.
That vagueness is the point. This is a company still defining the problem, not announcing a product. But the direction is unmistakable, and it follows a script every frontier AI lab with enough cash is now running.
Look at who Anthropic just hired. Clive Chan, an early member of OpenAI's own custom silicon team, joined Anthropic as part of what the company is calling a deliberate hardware buildout. You don't recruit a rival's chip architect for a side project. Anthropic wants engineers who have already shipped an AI accelerator from whiteboard to wafer, and Chan is one of a small number of people on earth who has done that.
He would know. OpenAI unveiled its own first chip, a Broadcom-built inference processor called Jalapeño, on June 24. Broadcom and OpenAI took it from initial design to manufacturing tape-out in nine months, which the companies describe as the fastest ASIC development cycle ever completed in advanced semiconductors. Jalapeño is built for inference specifically, not training, and it's targeting deployment by the end of 2026. Anthropic is now reaching for the same playbook, just with Samsung instead of Broadcom and Seoul instead of the usual Bay Area supply chain.
None of this replaces what Anthropic already has running. The company has been explicit that Nvidia GPUs, Google's tensor processing units, and Amazon's Trainium chips remain central to how it scales Claude. In October 2025, Google committed up to $40 billion in cash and TPU capacity to Anthropic, giving it access to as many as one million TPUs, according to CNBC. In April 2026, Amazon followed with up to $25 billion in investment and a commitment from Anthropic to spend more than $100 billion on AWS over the next decade, tied to deploying up to five gigawatts of Trainium capacity. Anthropic is also in discussions with Microsoft and the UK chip startup Fractile. A Samsung chip wouldn't sit above that stack. It would sit alongside it.
Samsung has its own reason to want this deal badly. Its foundry business trails TSMC by a wide margin in advanced node market share, and a marquee AI customer on the 2 nanometer line would be the kind of reference win that pulls in the next customer. Samsung, along with SK Hynix and Micron, also took part in Anthropic's $65 billion funding round in May, so the relationship and the capital were already in the room before any chip conversation started.
Here's the part worth sitting with. None of these labs actually distrust Nvidia's chips. They distrust Nvidia's pricing power and the fact that a single supplier controls the input every one of them needs most. Building your own silicon is expensive, slow, and only pays off at massive scale, which is exactly why only the best-funded labs can even consider it. For everyone else, it's still cheaper to keep buying from Nvidia and complain about the margin.
Frankly, this deal isn't going to dent Nvidia's revenue this year, or probably next. Anthropic itself says its own chip has no timeline. What it does confirm is that owning your own compute has become table stakes for any AI company with the balance sheet to attempt it, the same way owning a data center did a decade ago. Whether Samsung ends up being the manufacturer or just one option among several Anthropic is testing, the direction of travel for the industry is now set. Everyone with the money is building a hedge against Nvidia, and Samsung just got a chance to sell one.
Also read: Alibaba and Tencent Back Kling AI Instead of Racing to Beat It • Jersey Mike's Mentioned AI 22 Times In Its Billion Dollar IPO Filing • Blackstone's QTS Walks Away From Its Massive Virginia Data Center Project
