The Claude maker is hiring silicon engineers to co-design hardware and models, its first public step toward developing proprietary AI hardware for its frontier AI systems.
For years, the artificial intelligence boom has relied heavily on advanced accelerators supplied by a handful of semiconductor companies. On Wednesday, Anthropic signaled that it wants to play a more active role in the hardware stack powering its AI models.
The company confirmed it is assembling an in-house team to design custom chips for its Claude models — the first time it has publicly acknowledged a proprietary hardware effort. According to Anthropic, the objective is to improve the performance and cost efficiency of running Claude at the scale demanded by its growing customer base.
Hardware and models, designed together
Anthropic's approach goes beyond simply designing a chip. Instead, the company is pursuing a hardware and software co-design strategy, engineering its silicon alongside its AI models so the two can be optimized together. The concept is already familiar across the AI industry. Purpose-built accelerators can be tailored to the computational and memory requirements of specific model architectures, potentially delivering better performance and efficiency than more general-purpose hardware for targeted workloads. Companies including Google and Apple have long used similar design philosophies for parts of their hardware ecosystems.
Anthropic is recruiting accordingly. Public job listings for its custom silicon team seek engineers with experience spanning both hardware and software development, with listed compensation ranging from approximately $320,000 to $485,000. The postings emphasize experience shipping semiconductor products and making technical decisions in small teams. They also ask for engineers who work across multiple engineering disciplines.
Designing advanced chips remains a complex, long-term undertaking, requiring extensive verification and close coordination with manufacturing partners before production.
A costly undertaking
Developing leading-edge AI chips can require investments of hundreds of millions of dollars, according to industry analysts. Engineering costs and design verification account for much of that, along with the preparation required for manufacturing.
Anthropic has not announced a timeline for its first custom chips or identified a manufacturing partner. Earlier media reports indicated the company had explored Samsung Electronics as a potential fabrication partner, although Anthropic has not publicly confirmed any manufacturing agreement.
Maintaining a multi-chip strategy
Anthropic has emphasized that its custom silicon effort is intended to complement its existing hardware partnerships rather than replace them. The company says it will continue using infrastructure from Amazon Web Services, Google, Nvidia, and AMD as part of a broader multi-chip strategy.
That approach reflects Anthropic's significant existing compute commitments, including previously announced plans to access large-scale TPU capacity through Google beginning in 2027. Developing proprietary silicon is therefore expected to proceed alongside the company's current infrastructure partnerships.
An industry-wide trend
Anthropic joins a growing list of major AI companies investing in custom hardware. OpenAI has reportedly been developing custom AI chips with Broadcom, while Google has long relied on its Tensor Processing Units (TPUs) for AI workloads. Meta has also continued expanding its Meta Training and Inference Accelerator (MTIA) program for AI infrastructure.
The broader trend reflects the increasing importance of optimizing both hardware and software as AI models grow larger and more computationally demanding. For companies operating at frontier scale, custom silicon offers the potential to improve efficiency, reduce long-term operating costs, and diversify hardware supply.
Comments