Anthropic is building an internal team to design custom artificial intelligence chips, the company confirmed on August 5. This initiative marks a significant step for the AI developer as it seeks to optimize the performance of its Claude models. The company stated it is hiring engineers with expertise in both hardware and software to co-design chips and AI models. This approach is intended to make Claude operate faster and more efficiently for its customers.
The move signifies Anthropic's commitment to vertical integration in its computing infrastructure. While the company will continue to utilize hardware from partners such as Amazon Web Services (AWS), Google, Nvidia, and AMD, developing its own silicon aims to address the growing demand for computational power and potential supply constraints. This strategy mirrors efforts by other leading AI firms, including OpenAI, which has also explored custom chip development.
Anthropic's custom silicon initiative is part of a broader "multi-chip strategy." The company indicated that its proprietary hardware will complement, rather than replace, existing partnerships. This approach allows Anthropic to leverage the strengths of various hardware providers while tailoring specific solutions for its AI models. The demand for specialized AI chips has surged as models like Claude become more complex and are deployed at a larger scale.
The company is actively recruiting engineers for this new division, with job listings indicating a salary range of $320,000 to $485,000. These roles require candidates with experience in semiconductor design, from concept to production. Anthropic's decision to invest in in-house chip design reflects a trend among AI developers to gain more control over their hardware supply chain and performance optimization. This control can lead to improved efficiency and potentially lower costs in the long run, by reducing reliance on external chip manufacturers and optimizing hardware specifically for their AI workloads.
Previously, Anthropic has relied on a mix of cloud computing platforms and hardware accelerators. In October 2025, the company announced plans to expand its use of Google Cloud infrastructure, aiming for up to one million Tensor Processing Units (TPUs) to support Claude's growing demands. This expansion represented tens of billions of dollars and over one gigawatt of additional computing capacity. Anthropic has also utilized AWS Trainium chips and NVIDIA GPUs. However, the complexities of managing and optimizing across multiple hardware platforms have presented challenges, as noted in past performance issues stemming from infrastructure bugs.
The development of custom AI chips is a costly and complex undertaking. However, the potential benefits, such as enhanced performance, greater efficiency, and reduced dependency on a limited number of chip suppliers like Nvidia, are driving this trend. By co-designing hardware and models, Anthropic aims to achieve a level of optimization not possible with off-the-shelf components. This integrated approach allows for fine-tuning hardware architecture to match the specific computational needs of Claude, potentially leading to significant performance gains.
Anthropic's move into chip design also occurs within a broader geopolitical and competitive context. The company has previously expressed concerns about the concentration of AI development capabilities and has advocated for policies to maintain a technological advantage for democratic nations, including strict export controls on advanced chips. Developing its own silicon could be seen as a further step in securing its technological independence and contributing to the broader national security landscape surrounding advanced AI.