Anthropic wants to build its own AI chips to power Claude

Anthropic, the company behind the Claude AI assistant, is now hiring engineers to design custom chips for its models. The move follows reports that the company was in talks with Samsung to manufacture these chips. Anthropic confirmed it wants to co-design hardware and AI models to make its technology faster and more efficient.

This isn’t a first in the AI world. OpenAI recently revealed its own custom chip, Jalapeño, built by Broadcom. Google has long used its own TPU chips for AI, while Meta is developing its MTIA accelerators. Anthropic’s job listing makes it clear they’re serious about this, seeking experts in chip design for a new custom silicon team.

The push comes as demand for Claude grows, and AI companies race to secure their own computing power. Anthropic already partners with AWS, Google, Nvidia, and AMD for hardware, but building its own chips could help it scale without relying on others.

Anthropic started as an AI safety-focused research lab but shifted to building large language models like Claude. As more users and businesses adopt Claude, the company needs more computing power to keep up. Custom chips could give it an edge in performance and cost, especially if it can tailor them to its specific models.

This matters because it shows how AI companies are taking control of their tech stack. If Anthropic succeeds, it could mean faster, cheaper, and more reliable AI for everyone using Claude. But it also raises questions about whether this will lead to better AI or just more competition in an already crowded field.

What happens next depends on how quickly Anthropic can build this team and whether Samsung or another partner steps in to manufacture the chips. Watch for announcements about partnerships or early prototypes in the coming months.

Do you think AI companies building their own chips will lead to better products, or is this just a power play to control the market?

Would you trust a company like Anthropic to design chips that are truly efficient, or is this a risk for users?

---

Filed under: AI, Chips, Anthropic, Claude, TechNews