The Anthropic Samsung custom chip conversation is still in its earliest stages, but it signals something harder to ignore: one of the world’s best-funded AI labs is now serious about breaking its dependence on existing silicon suppliers. The Information reported that Anthropic has been in contact with Samsung to explore a collaboration on a bespoke chip, with the two companies yet to settle on what the chip will do, how it will fit into a server, or how much processing power it will need.
Anthropic, when asked for comment, was careful not to confirm or deny the Samsung talks. A diversified hardware stack that includes chips from Google, Amazon, and Nvidia will continue to be pivotal to its compute strategy, the company said. On the Samsung partnership question specifically, it had nothing further to add.
What an Anthropic Samsung Custom Chip Would Mean for the AI Hardware Market
The stakes of getting the design right are considerable. UPI reports that Anthropic is considering Samsung’s 2-nanometer manufacturing process and the company’s advanced chip-packaging facilities for the project. Moving to a 2nm node would put any resulting chip at the frontier of semiconductor fabrication, the same territory occupied by the most advanced processors currently in production.
Samsung brings more to the table than a fabrication line. The Korean manufacturer is already deeply embedded in the AI supply chain as a major partner of Nvidia, producing chips Nvidia needs to train and run its models. The two companies are now extending that relationship: Samsung’s AI Megafactory, built in collaboration with Nvidia, will deploy more than 50,000 Nvidia GPUs and integrate Nvidia’s Omniverse platform across Samsung’s entire manufacturing flow, spanning semiconductor design, process, equipment, operations, and quality control. Samsung has also held separate talks with Google about chip-making.
For Anthropic, a Samsung collaboration would represent a meaningful step beyond its current position as a customer of other companies’ silicon. The appeal is not merely supply security. Custom chips allow a lab to tailor hardware to its specific workloads, potentially wringing out efficiencies that general-purpose processors cannot match.
OpenAI’s Jalapeño Sets the Competitive Tempo
The backdrop to Anthropic’s hardware ambitions is OpenAI’s unveiling of Jalapeño, its first custom AI inference chip, built with Broadcom and announced at an event on 24 June 2026. OpenAI describes Jalapeño as the first chip in a multi-generation compute platform, designed with the flexibility to work across all large language models in the industry, not only OpenAI’s own. The partnership with Broadcom was first disclosed in October 2025, predating the formal chip reveal by several months.
The Jalapeño development timeline itself became part of the announcement’s substance. VentureBeat reported that the chip went from initial design to manufacturing tape-out in just nine months, a pace OpenAI believes is the fastest ever achieved for high-performance semiconductors. OpenAI’s own AI models were used to accelerate portions of the chip design process.
Celestica played a manufacturing role alongside Broadcom, helping industrialise the platform through chip implementation, board and rack system integration, high-performance networking, and scalable production. Broadcom’s press release frames the collaboration within a 10-gigawatt accelerator partnership, with OpenAI’s longer-term goal being gigawatt-scale datacentre deployments.
OpenAI’s chip also claims a performance-per-watt advantage over competing processors, a metric that carries weight as datacentre energy costs become a defining constraint for AI operators at scale.
The wider pattern is clear enough. Amazon and Google already offer custom tensor processing units as part of their cloud platforms. OpenAI now has Jalapeño. Anthropic, which runs its Claude models on a mix of Google, Amazon, and Nvidia hardware, is the most prominent frontier lab still without a chip of its own. Whether the Samsung conversation produces a finished product, or simply establishes what a finished product might eventually look like, will tell researchers and rivals a great deal about how seriously Anthropic intends to compete at the infrastructure layer.
The first concrete signal will come when, or if, Anthropic settles on a use case for the chip. Until then, the 2nm ambition sits somewhere between a credible roadmap and an open question.
