Anthropic eyes Samsung partnership as it explores developing custom AI chips

Anthropic is weighing a potential collaboration with Samsung as it explores designing its own artificial intelligence chip, according to a new report, signalling that the battle among AI companies is increasingly shifting from software to the hardware powering advanced models.

The reported discussions come as leading AI developers seek greater control over the computing infrastructure required to train and deploy large language models. With demand for AI processors continuing to outpace supply, several companies are investing in specialised silicon that can improve efficiency while reducing reliance on third-party chipmakers.

Anthropic evaluates next step in compute strategy

According to a report by The Information, Anthropic has held talks with Samsung about working together on a custom AI processor. However, the project remains at an early stage. The report said the company has not yet finalised the chip's intended purpose, its role within server systems or the level of computing performance it aims to deliver.

More from Tech

The development follows an earlier Reuters report from April, which said Anthropic had begun considering the possibility of building its own AI chips in response to persistent shortages affecting access to high-end processors.

When contacted by TechCrunch, Anthropic did not confirm any partnership plans with Samsung. Instead, the company reiterated that its computing strategy will continue to rely on a mix of hardware supplied by Google, Amazon and Nvidia.

"A diversified hardware stack that includes chips from Google, Amazon, and Nvidia will continue to be pivotal to its compute strategy," Anthropic told the publication. Asked specifically about the reported Samsung discussions, the company declined to provide additional comment.

Even without an official confirmation, the reported move reflects a broader industry trend as AI companies look to tailor hardware for their own workloads while gaining greater independence from Nvidia, whose graphics processors remain the dominant choice for AI training and inference.

Custom silicon becomes the next competitive frontier

Anthropic's reported efforts come shortly after rival OpenAI unveiled its own custom inference chip developed in partnership with Broadcom. The processor, known internally as "Jalapeño", is designed to improve performance-per-watt compared with competing solutions, highlighting the industry's growing focus on energy-efficient AI infrastructure.

Large cloud providers have already invested heavily in proprietary silicon. Google has spent years expanding its Tensor Processing Unit (TPU) family to support AI workloads across its cloud platform, while Amazon has introduced its own custom chips to optimise machine learning services for customers.

Samsung, meanwhile, occupies an increasingly important position within the AI supply chain. The South Korean technology giant manufactures components used in AI hardware and works closely with Nvidia, producing chips that support the training and deployment of advanced AI models. Nvidia's software tools also play a role in Samsung's semiconductor manufacturing operations.

Beyond its relationship with Nvidia, Samsung has reportedly discussed semiconductor collaborations with Google and is involved in plans for an AI-focused chip manufacturing facility in South Korea.

If Anthropic eventually proceeds with its own processor, it would join a growing list of AI developers betting that proprietary silicon can offer long-term strategic advantages. As competition intensifies, ownership of both software and hardware is emerging as a key differentiator in the race to build more capable and cost-efficient AI systems.