Anthropic Assembles In-House Custom Silicon Team to Power Next-Gen AI Hardware
In an effort to cater to the rising enterprise needs for Claude AI models, the artificial intelligence startup company Anthropic will reportedly enter the domain of hardware creation. The company stated that its San Francisco-based AI research lab is recruiting a team of semiconductor engineers who will create custom silicon for its software architecture.
Instead of relying solely on off-the-shelf accelerators, Anthropic will “co-design” custom silicon and subsequent generations of its AI models. By creating hardware that can be used by Anthropic to optimize Claude’s data processing, the company hopes to speed up execution times and reduce inference costs substantially.
Job listings for Anthropic’s new “custom silicon team” include salary offerings from $320,000 to $485,000 and are looking for engineers with experience in creating real-life silicon chipsets.
However, despite the attempts at creating proprietary silicon, Anthropic is not going to cut ties with its existing hardware partners. It was stressed that custom chips would complement the firm’s multi-chip strategy and its cooperation with major players such as Nvidia, AMD, Amazon Web Services, and Google Cloud.
Developing an advanced accelerator for AI requires expensive investments, typically over $500 million per chip development project because of the huge cost of engineering and stringent manufacturing requirements. Despite no timeline and no official disclosure of the production partners of Anthropic, the lab was initially reported to have conducted manufacturing talks with Samsung Electronics.
This represents part of the growing trend that we see happening in the entire AI industry as companies such as OpenAI, Google DeepMind, and Meta pour substantial investments into developing their custom hardware in order to avoid any supply shortage problem and minimize dependence on any other hardware manufacturer.
By taking control of the hardware layer of the stack, Anthropic will be in a position to conduct computations more efficiently in the future as automated AI adoption will become common in all sectors of the economy.
