Sources have admitted that the relevant plans are still in the early stages. The company has not yet finalized any specific solutions, nor has it assembled a dedicated team. Anthropic may ultimately choose to only purchase chips rather than design its own. However, even the mere possibility is enough to signal something significant.
Currently, Anthropic relies on both Google parent company Alphabet’s TPU (Tensor Processing Units) and Amazon’s Trainium chips to develop and run Claude. Just this week, the company also signed a long-term agreement with Google and Broadcom—the latter being the key design partner for Google’s TPUs. While inking a multi-billion-dollar external procurement deal on one hand, quietly exploring self-developed chips on the other, this “two-handed” strategy strongly resembles what Meta and Microsoft were doing a few years ago. Today, both companies already possess their own proprietary chips.
Designing a top-tier AI chip is estimated by industry insiders to cost around $500 million. But beyond the price tag, what deserves more attention is the industry signal behind Anthropic’s move. When a pure model company begins seriously considering self-developed silicon, it indicates that the hardware battle over AI inference has entered a new level of intensity.
