Anthropic is in early talks with Samsung Electronics to manufacture a custom AI chip, The Information reported Thursday, a move that would give Samsung a high-profile customer for its advanced chipmaking and help Anthropic reduce its reliance on off-the-shelf processors as it scales AI infrastructure.
Anthropic, creator of the Claude large language model family, is considering Samsung's 2-nanometer (nm) manufacturing process and its advanced chip-packaging technologies, according to multiple people familiar with the discussions. The project remains at an early stage, with no detailed chip design, testing or manufacturing yet underway.
Samsung's 2 nm process and advanced packaging
The 2 nm node is among the most advanced semiconductor processes, shrinking transistor size to pack more performance and energy efficiency into each chip. Advanced packaging places processors, high-bandwidth memory and other components closer together, increasing data transfer speeds and reducing bottlenecks when running large AI models.
Anthropic is studying the functions and performance needed for the chip, as well as how it would integrate into servers, the sources said. The company is also talking with several chip-design firms.
Anthropic builds hardware muscle
Anthropic hired Clive Chan in June, the second hardware engineer from OpenAI's custom-chip program to join the company. Chan announced his departure from OpenAI on X, saying he was drawn to "begin climbing a new technological mountain from the bottom." The hire signals Anthropic is assembling an internal team to design specialized processors as its competition with OpenAI expands beyond AI models into data-center hardware.
Anthropic raised $65 billion in a Series H funding round that closed May 28, placing its post-money valuation at $965 billion. Samsung, SK hynix and Micron participated as strategic infrastructure partners. Samsung is the only one of the three that also operates a large contract chip-manufacturing business, raising expectations that its relationship with Anthropic could widen beyond memory supplies.
The custom-chip effort does not mean Anthropic is walking away from current suppliers. "Nvidia GPUs, Google TPUs and AWS Trainium chips will continue to play a central role in our computing resources," the company said in a statement to The Information.
Anthropic is also evaluating processors developed by Microsoft and British chip startup Fractile as it considers different ways to expand its infrastructure.
AI companies race to design their own silicon
Major technology companies have been pouring resources into custom processors to cut computing costs, improve energy efficiency and gain better control over their AI hardware stack. Google has built several generations of tensor processing units (TPUs), while Amazon Web Services operates its Trainium chips for AI training.
OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom inference processor, on June 24. The chip was developed from initial design to production in just nine months, with early deployment expected by year-end. Broadcom CEO Hock Tan called it the start of a multigeneration processor roadmap.
Anthropic is entering the custom-chip race later than some rivals, but growing demand for AI infrastructure is opening opportunities for specialized hardware. TrendForce projects shipments of servers using cloud companies' custom application-specific integrated circuits (ASICs) will grow 44.6% in 2026, compared to 16.1% growth for servers using general-purpose graphics processors. The push to design custom processors reflects the intense demand for efficient hardware to power generative AI and LLM workloads.
Samsung's chipmaking ambitions and challenges
A manufacturing deal with Anthropic would give Samsung another major AI chip customer as it challenges Taiwan Semiconductor Manufacturing Co. (TSMC). Samsung previously signed a $16.5 billion agreement to make next-generation AI chips for Tesla, and Google is reportedly considering Samsung for part of a future TPU.
Samsung's 2 nm production stability will be critical. The company has faced yield issues in earlier advanced processes, and an industry official said Samsung has become "more selective about accepting manufacturing orders, focusing resources on projects considered commercially and technologically viable."
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 fabrication plants and expanded production of high-bandwidth memory and advanced packaging.
Why this matters for IT, development, and research professionals
The shift toward custom AI silicon will reshape the hardware options available for running large-scale inference and training workloads. IT architects and developers working with AI may soon have more chip choices beyond Nvidia GPUs, potentially bringing down costs and enabling more efficient model serving. Researchers focused on performance optimization and hardware-software co-design will find new opportunities as companies like Anthropic, OpenAI, and others build in-house chip teams and seek talent that can bridge model requirements and silicon design.
Your membership also unlocks:
