Anthropic is considering designing its own AI chips, according to a Reuters report citing three sources familiar with the situation. The exploration is in the early stages, and the company hasn’t committed to a specific design or assembled a team yet. They might still opt to buy chips from third parties. A company spokesperson declined to comment.
This exploration follows Anthropic’s sharp revenue rise. The annual revenue run rate has surpassed $30 billion, up from about $9 billion at the end of 2025. This growth has led to a scale of compute demand that makes custom silicon economically appealing. Anthropic currently uses a mix of chips, including Google’s tensor processing units, Amazon’s custom chips, and Nvidia hardware, matching workloads to the best-suited chips.
Just days before the Reuters report, Anthropic signed a long-term deal with Google and Broadcom, accessing about 3.5 gigawatts of TPU compute from 2027, according to Broadcom’s SEC filing. This is around three times its consumption earlier in 2026. The deal is contingent on Anthropic’s continued success and builds on its 2025 commitment to invest $50 billion in US infrastructure.
Broadcom, also a chip partner for OpenAI, is central to the custom AI silicon market, emerging as an alternative to Nvidia’s GPUs. Anthropic developing proprietary silicon mirrors trends in the industry, with Meta and OpenAI also working on custom chips.
According to Reuters, developing an advanced AI chip costs about $500 million due to hiring specialized engineers and validating manufacturing. Although Anthropic is currently unprofitable, this cost is more manageable given its tripled revenue run rate in four months.