Anthropic Assembles Engineering Team to Design Custom AI Chips for Claude
The startup joins OpenAI in pursuing proprietary hardware while maintaining key supplier partnerships.
Anthropic PBC is building an internal hardware team to design custom artificial intelligence chips, joining a broader push among leading AI developers to reduce reliance on external hardware suppliers and lower the costs of running large-scale models.
The San Francisco-based startup confirmed the development team’s formation to Business Insider, marking its first public acknowledgment of custom silicon ambitions. The effort focuses on creating specialized processors tailored to train and execute Anthropic’s flagship Claude AI models faster and with greater energy efficiency.
While Anthropic is stepping into hardware design, the company said it will maintain a “multi-chip” strategy, preserving its existing supply relationships with Nvidia Corp., Advanced Micro Devices Inc., Amazon.com Inc., and Google. Amazon and Google have both invested billions of dollars into Anthropic, with Amazon serving as its primary cloud and training partner utilizing its own Trainium and Inferentia chips alongside Nvidia GPUs.
Anthropic, led by Chief Executive Officer Dario Amodei, has not revealed how far along the chip initiative is or who will manufacture the physical silicon. However, reports in early July indicated the company had approached Samsung Electronics Co. to explore manufacturing partnerships, specifically targeting Samsung’s 2-nanometer fabrication process. Samsung participated in a prior $65 billion financing round for Anthropic as the South Korean foundry seeks to gain ground on market leader Taiwan Semiconductor Manufacturing Co.
The move mirrors recent steps taken by principal rival OpenAI, which partnered with Broadcom Inc. to create “Jalapeño,” a specialized processor built specifically for OpenAI’s neural networks. That project progressed from initial design to fabrication in nine months, assisted by OpenAI’s own language models during the design phase.
Scaling computing capacity remains the primary operational bottleneck for frontier AI developers, where hardware expenditures account for the vast majority of capital deployment. Designing in-house silicon allows AI companies to optimize architectures specifically for their proprietary software stacks while gaining leverage in supplier negotiations.
