Anthropic Eyes Samsung to Manufacture Its First Custom AI Chip
Anthropic, the San Francisco-based AI company behind the Claude family of large language models, is reportedly in advanced talks with South Korean electronics giant Samsung to manufacture its first custom artificial intelligence chip. The move would represent a significant escalation in the AI industry’s push toward designing purpose-built hardware — and away from its near-total dependence on NVIDIA.
According to sources familiar with the discussions, Anthropic is exploring a partnership that would leverage Samsung’s advanced chip fabrication capabilities to produce specialised AI inference chips optimised specifically for running Claude-scale models. Unlike the general-purpose GPUs that dominate the market today, these custom chips would be designed from the ground up to handle the unique computational patterns of transformer-based language models.
Breaking Free from NVIDIA
The semiconductor industry has been reshaped by the AI boom, with NVIDIA capturing an estimated 80% or more of the data centre GPU market. That dominance has led to soaring prices, long wait times for hardware, and a growing desire among AI companies to diversify their supply chains. Anthropic’s move toward custom silicon mirrors similar efforts at OpenAI, Google, Microsoft, and Amazon — all of whom are developing proprietary AI chips.
“When you’re spending billions on compute, even a 20% efficiency gain from custom silicon translates into enormous savings,” explained one semiconductor industry analyst. “Every major AI lab is asking the same question: can we build something more efficient than an H100 for our specific workloads?”
Why Samsung?
Samsung’s foundry business, while smaller than TSMC’s, offers several advantages for an AI startup looking to move quickly. The company has invested heavily in advanced packaging technologies that are particularly well-suited to AI workloads, and it has been aggressively courting AI chip designers as part of its strategy to close the gap with TSMC.
For Samsung, landing Anthropic as a customer would be a major validation of its AI chip ambitions. The foundry has struggled to attract high-profile AI chip customers, with most of the industry’s biggest names — including NVIDIA, AMD, and Intel — relying primarily on TSMC for their most advanced chips.
A Growing Trend
Anthropic’s move toward custom hardware is part of a broader industry trend. OpenAI is developing its own AI chips in partnership with Broadcom and TSMC. Google’s Tensor Processing Units (TPUs) are already on their sixth generation. Microsoft recently unveiled its Maia AI accelerator. Even Meta is working on custom silicon for its recommendation and AI workloads.
The rise of custom AI chips signals a maturing industry — one that is moving beyond the one-size-fits-all approach of general-purpose GPUs toward specialised hardware optimised for the specific demands of frontier AI models. For Anthropic, which positions itself as a safety-focused AI company, controlling its own hardware stack could also offer additional benefits for security and interpretability research. As one industry watcher put it: “In AI, controlling the silicon increasingly means controlling your destiny.”
