Anthropic eyes South Korea's Samsung for custom AI chip
· UPIJuly 3 (Asia Today) -- Anthropic, the developer of the Claude artificial intelligence model, is in early discussions with Samsung Electronics about manufacturing a custom AI chip, according to a U.S. technology news report.
The Information reported Thursday, citing multiple people familiar with the discussions, that Anthropic is considering using Samsung's 2-nanometer manufacturing process and advanced chip-packaging facilities.
The project remains at an early stage. Anthropic has not begun detailed chip design, testing or manufacturing, the report said.
Samsung's 2-nanometer process is among the most advanced semiconductor manufacturing technologies available. Smaller manufacturing nodes can allow more transistors to be placed on a chip, potentially improving computing performance and energy efficiency.
Advanced packaging places processors, high-bandwidth memory and other chip components closer together. The shorter distance can increase data-transfer speeds and reduce bottlenecks when running large AI models.
Anthropic is studying the functions and performance required for the chip as well as how it could be integrated into servers, people familiar with the matter said. The company is also reportedly holding discussions with several chip-design companies.
Anthropic is considering using processors developed by Microsoft and British chip startup Fractile as it evaluates different approaches to expanding its computing infrastructure.
The company hired Clive Chan in June. Chan was the second hardware engineer to join OpenAI's custom-chip program and worked on the project from its early stages.
Chan announced his departure from OpenAI and move to Anthropic in a June 7 post on the social media platform X. He said he was drawn by the opportunity to begin climbing a new technological mountain from the bottom.
The recruitment suggests Anthropic is building an internal team capable of designing specialized processors as competition with OpenAI expands from AI models into hardware and data-center infrastructure.
Anthropic raised $65 billion in a Series H investment round completed May 28, giving the company a post-investment valuation of $965 billion.
The funding was led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital. Samsung Electronics, SK hynix and Micron participated as strategic infrastructure partners.
Anthropic said the three semiconductor companies provide technologies that play important roles in supplying memory, storage and logic chips.
Samsung is the only one of the three companies that also operates a large contract chip-manufacturing business, raising expectations that its relationship with Anthropic could expand beyond memory supplies.
A manufacturing agreement with Anthropic would give Samsung another major AI customer as the South Korean chipmaker seeks to challenge Taiwan Semiconductor Manufacturing Co. in the market for advanced processors.
Samsung previously signed a $16.5 billion agreement to manufacture next-generation AI chips for Tesla. Google is also reportedly considering using Samsung to manufacture part of a future tensor processing unit.
The potential Anthropic contract could strengthen Samsung's position as demand for alternatives to Taiwan Semiconductor's manufacturing capacity increases.
Major technology companies are developing specialized processors to reduce computing costs, improve energy efficiency and gain greater control over their AI infrastructure.
Google has developed several generations of its tensor processing units, while Amazon Web Services operates its Trainium processors for AI training.
OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom inference processor, on June 24. Inference refers to the process through which a trained AI model generates responses to user requests.
OpenAI said the processor was developed from initial design to production in nine months. Early deployment is expected by the end of the year.
Broadcom Chief Executive Officer Hock Tan described Jalapeño as the beginning of a multigeneration processor roadmap. The companies plan to install the chips in large-scale data centers operated with partners including Microsoft.
Anthropic said its custom-chip work would not replace its existing relationships with hardware suppliers.
"Nvidia GPUs, Google TPUs and AWS Trainium chips will continue to play a central role in our computing resources," the company said in response to a request for comment from The Information.
South Korea on Monday unveiled a wider semiconductor investment plan under which Samsung and SK hynix are expected to invest about 800 trillion won ($523 billion) over the next decade.
The plan includes four new semiconductor fabrication plants and expanded production of high-bandwidth memory and advanced packaging technologies used in AI systems.
Samsung has faced yield problems in some previous advanced manufacturing processes. Yield refers to the percentage of usable chips produced from each semiconductor wafer.
The performance and production stability of Samsung's 2-nanometer process are therefore expected to be critical to its ability to compete with Taiwan Semiconductor for major AI-chip orders.
An industry official said Samsung has become more selective about accepting manufacturing orders, focusing resources on projects considered commercially and technologically viable.
Anthropic is entering the custom-chip competition later than several major AI and cloud-computing companies. However, rapidly rising demand for AI infrastructure is creating opportunities for specialized processors.
TrendForce projects that shipments of servers using cloud companies' custom application-specific integrated circuits will grow 44.6% in 2026. Shipments of servers using general-purpose graphics processors are expected to grow 16.1%.
Nvidia remains the dominant supplier of AI processors, but the development of chips by OpenAI, Google, Amazon and other technology companies could gradually reduce their reliance on its hardware.
-- Reported by Asia Today; translated by UPI
© Asia Today. Unauthorized reproduction or redistribution prohibited.
Original Korean report: https://www.asiatoday.co.kr/kn/view.php?key=20260703010001110