Anthropic confirms the creation of its own chip team for Claude with salaries of up to $485,000

Anthropic has publicly confirmed the creation of an internal team specializing in semiconductor design to power its Claude language models. As detailed by a company spokesperson to Business Insider, the objective of this initiative is co-design hardware and algorithms in a coordinated manner. This approach seeks to execute its clients’ workloads faster and more efficiently at scale.

The confirmation comes after the publication of several job offers in which the startup is looking for engineers for its custom silicon division. The positions offer salaries between $320,000 and $485,000 annual. The company is requiring candidates with direct experience packaging and shipping commercial chips to market, in a bid to speed up development timelines after having previously held talks with Samsung as a potential manufacturing partner.

Multi-chip strategy to reduce inference costs without abandoning your current suppliers

Despite starting the development of its own processor, the company has clarified that it does not plan to break with its traditional infrastructure suppliers. Anthropic will maintain a multichip strategy in which accelerators from NVIDIA, AMD, Amazon Web Services and Google will continue to play a central role in their expansion plans. The new proprietary silicon will be targeted at specific workloads where optimization between model and physical architecture reduces inference operating costs.

Anthropic’s move reflects a widespread trend among major development labs for Break free from exclusive dependence on third-party hardware. At the beginning of summer, OpenAI presented its Jalapeño accelerator developed together with Broadcom, while Google is working on specialized processors for Gemini aimed at optimizing energy consumption per query made in its data centers.

Technology companies such as Meta have also joined this race, which is preparing the production of the new generation of its own processors, and European startups such as Mistral, which are evaluating making the leap to dedicated silicon. Although designing custom chips requires million-dollar investments and years of technical developmentthe volume of queries managed by these systems justifies the effort to stabilize the supply chain and cut energy expenditure in data centers. In fact, the goal is to have, not only a more capable AI, but also a more efficient and cheaper one.