Anthropic AI chips bold move: powerful response to global crunch

Anthropic AI chips are back in the spotlight as the San Francisco-based startup weighs building its own processors to power the next wave of its Claude models.

The company is exploring custom semiconductor design in response to soaring demand for artificial intelligence computing and a persistent shortage of advanced chips worldwide.

People familiar with the discussions say Anthropic is assessing whether it makes business sense to design Anthropic AI chips in-house, but no final architecture has been chosen and no dedicated engineering team has been formed yet. The firm could still decide to rely entirely on external suppliers such as Nvidia and Google, which currently provide much of the hardware behind its Claude chatbot and other systems. Industry estimates suggest that developing a cutting-edge AI chip can cost more than 500 million dollars once design work, specialist hiring, testing and large-scale manufacturing are factored in.

Today, Anthropic leans on a mix of Nvidia GPUs, Google tensor processing units (TPUs), Amazon's Trainium and Inferentia chips, and Broadcom-designed hardware to train and run Claude. Just days before news of the Anthropic AI chips exploration emerged, the company signed a fresh long-term agreement with Google and Broadcom for multiple gigawatts of next‑generation TPU capacity from 2027, largely located in the United States. That deal builds on Anthropic's earlier commitment, announced in 2025, to invest about 50 billion dollars in strengthening US computing infrastructure.

Designing Anthropic AI chips would mirror moves by rivals who see control over infrastructure as too strategic to leave entirely to outside vendors. Success could give Anthropic more predictable access to compute, better optimisation for its models and potentially lower long‑term costs, though the technical and financial risks are substantial and the company may ultimately stick with buying chips instead.