Anthropic is reportedly discussing a potential custom AI chip partnership with Samsung as it seeks more control over the computing hardware used to develop and operate Claude. The discussions remain at an early stage, and Anthropic has not decided the chip’s purpose, specifications or placement within AI servers.
The company began considering an internally designed processor earlier this year as demand for AI computing continued to rise. Anthropic currently relies on a mix of hardware from Nvidia, Google and Amazon rather than a single supplier.
Anthropic told TechCrunch that a diversified hardware system would remain central to its computing plans. It declined to provide further details about its reported discussions with Samsung.
Samsung Could Provide Manufacturing and Packaging
The talks reportedly involve Samsung’s 2-nanometer manufacturing process and advanced chip-packaging capabilities. However, no detailed design or manufacturing work has started, and Anthropic may still decide not to proceed with the project.
A custom processor could be designed for training AI models, running inference or handling another specific workload. Anthropic has not yet determined how powerful the chip would need to be or how it would connect with other components inside its servers.
Developing specialised hardware could help Anthropic reduce some dependence on Nvidia, which remains the leading supplier of processors used for AI training and inference. It could also allow the company to optimise its hardware for the requirements of Claude models.
Anthropic has already committed to using substantial computing capacity based on Amazon’s Trainium processors. It has also secured access to infrastructure supported by Google processors and Broadcom technology, showing that any Anthropic-designed chip would form part of a wider multi-supplier system.
AI Companies Develop Their Own Processors
Several major AI and cloud companies have invested in custom processors to lower costs and improve performance for specific workloads. Amazon offers Trainium chips, while Google provides its Tensor Processing Units through Google Cloud.
OpenAI has also announced a custom inference processor developed with Broadcom. These projects give AI developers alternatives to Nvidia hardware, although most companies continue to use Nvidia processors alongside their own chips.
Samsung already supplies memory and other semiconductor components used in AI systems. The company also works with Nvidia on an AI-enabled semiconductor manufacturing facility that will use more than 50,000 Nvidia GPUs to improve chip development and production.
Anthropic has not announced a timeline for deciding whether to develop the processor or enter a formal partnership with Samsung.
Featured image credits: Heute.at
For more stories like it, click the +Follow button at the top of this page to follow us.