Anthropic is forming a new team to accelerate chip development efforts specific to artificial intelligence models. By designing its own hardware, the company aims to enable the Claude family to work with higher performance and more efficiently. The development, which was first brought to the agenda by Business Insider, was later confirmed by Anthropic in a statement made to TechCrunch. Thus, the company is preparing to take a more direct role not only in the software side, but also in the hardware field that forms the artificial intelligence infrastructure.
According to Anthropic’s statement, the goal is to increase processing performance and make resource use more efficient by designing hardware and artificial intelligence models together. Nowadays, training large language models and especially inference processes require very high computational power. Therefore, simply providing access to more powerful GPUs is not considered sufficient. The company evaluates that special silicone solutions developed according to its own needs can provide advantages in terms of both cost and performance in the long term.
Anthropic is looking for engineers for its custom silicon team
The job postings published by Anthropic also show that these plans have become concrete. The company is looking for engineers with experience in chip design to work in its team called “custom silicon team”. This team will be responsible for developing hardware architectures specific to Anthropic’s artificial intelligence models. By optimizing the hardware and software together, Claude models are aimed to run with lower latency and use existing computing resources more efficiently.
In a news published by The Information last month, it was claimed that Anthropic was evaluating a possible collaboration with Samsung for the production of these chips. Although the company has not made an official confirmation on this issue, working with experienced semiconductor companies on the production side is seen as an expected approach for the industry. Today, many companies developing special artificial intelligence chips do the design in-house, while leaving the production to foundries such as TSMC or Samsung.
This step by Anthropic comes after increasing demand for Claude models. Companies operating in the artificial intelligence sector are making multibillion-dollar agreements with large technology companies to access high-performance computing infrastructure. To date, Anthropic has established various collaborations with Amazon Web Services (AWS), Google, Nvidia and AMD, providing access to the artificial intelligence hardware it needs. Despite this, it is considered that being solely dependent on external suppliers is not sufficient for long-term growth targets. Specially designed chips can offer more flexibility in capacity planning and can be optimized for specific AI workloads.
Anthropic does not act alone in this field. In June, OpenAI introduced a special artificial intelligence chip called Jalapeño, developed in collaboration with Broadcom and designed specifically for inference operations. While Google DeepMind has long leveraged Alphabet’s Tensor Processing Unit (TPU) processors, Meta continues to develop its own artificial intelligence accelerators called MTIA. Microsoft, on the other hand, has announced special accelerators called Maia AI Accelerator for use in Azure data centers.
Competition in the artificial intelligence industry is no longer just between models. Computing infrastructure, energy efficiency and special hardware designs are also becoming important factors that determine the competitiveness of companies. The new team established by Anthropic stands out as one of the latest examples of this trend. It has not yet been announced when the special chips the company will develop will be available or which manufacturer it will work with. On the other hand, the recruitment of engineers clearly reveals that Anthropic aims to control its own artificial intelligence infrastructure more in the long term.
Join Channel