Anthropic is reportedly in talks with Samsung to build its own custom AI chip, as it looks to reduce costs and improve performance across AI training and inference.
According to a report by The Information, the plans are still in their early stages but appear to focus more on cost control than on outright performance gains. Anthropic has held talks with several chip firms, although discussions with Samsung are said to be further along.
It is the latest frontier AI model developer to look at custom silicon. OpenAI has announced its own custom inference chip, Jalapeño, with Broadcom, while the major cloud computing providers — Amazon, Microsoft, and Google — all have custom chips embedded in their data center infrastructure. Meta has also developed its own AI chips.
Anthropic and its slower AI infrastructure buildout
Anthropic has been slower than OpenAI to build out its own capacity, with the vast majority of its compute still coming from rented infrastructure. It has two data centers in Texas and New York, built with Fluidstack, and is expected to come online sometime this year. These would likely be among the first facilities to receive any custom Anthropic chip if it reaches production.
That lack of a major infrastructure buildout has kept Anthropic more cash-efficient than its rivals, with one report suggesting it may have reached profitability in the second quarter. However, it has also suffered from downtime more often than OpenAI or Google.
To improve capacity, Anthropic has signed compute deals with Amazon and Google in exchange for a small percentage of its shares. It has also reportedly signed a multi-year deal with SpaceX for access to its Colossus 1 data center, at a high price of $1.25 billion per month.
For Samsung, a partnership with Anthropic could be lucrative both financially and reputationally. The South Korean electronics giant is already benefiting from intense demand for memory, which has pushed up prices across parts of the consumer electronics market. Samsung is a market leader alongside SK Hynix and Micron and would be well-placed to develop a chipset with high-speed memory at its center.
More must-read AI coverage
- SS&C Intralinks DealCentre AI vs. Datasite: Which platform is built for the future of dealmaking?
- SS&C Intralinks FundCentre AI vs. Juniper Square: Which platform better supports modern private markets fund managers?
- Why Data, Not Models, Determines AI Success
- The Rise of the AI-Native Factory: How Physical AI Is Transforming Manufacturing
More power needed for Claude
Anthropic remains one of the most in-demand AI model makers, with Claude powering a growing share of enterprise AI workflows. That demand has led to bouts of instability and has also caught the attention of the US government, which banned exports of its most powerful model, Fable, a few weeks ago.
The government and Anthropic have been working on how to relaunch the model globally, although it may be some time before governments and companies in other countries get access.
In the meantime, Anthropic has launched a smaller, cheaper model, Sonnet, which should not face the same regulatory issues. It reportedly has stronger agentic features and additional safety protections and should be useful to many businesses deploying AI that are starting to worry about costs.
What this means for businesses using Claude
For companies building workflows around Claude, custom chips could eventually matter in a very practical way: availability, speed, and cost.
AI models are becoming part of customer support tools, coding assistants, internal search systems, document review workflows, and other business software. If Anthropic can lower the cost of running Claude and add more dedicated capacity, businesses may see fewer slowdowns, more reliable access, and pricing that is easier to sustain at scale.
That outcome is not guaranteed. Custom chips take time to design, test, manufacture, and deploy, and any Anthropic-Samsung project would still be early. But the direction is clear: as demand for AI grows, companies with greater control over their infrastructure may have an advantage in keeping services stable and affordable.
Also read: Banned Nvidia AI chips keep reaching China despite US export controls, as shell firms and intermediaries allegedly divert restricted hardware.