Anthropic’s chip strategy tests Nvidia’s pricing power

Anthropic is spreading its AI computing across four chip families: Amazon Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs. Software routes workloads to the hardware where they are most productive, allowing the company to manage capacity from multiple partners as a shared pool. Analysts say that flexibility could strengthen its negotiating position and reduce exposure to fluctuations in Nvidia’s chip supply.

The approach required years of investment and remains difficult to replicate, particularly for model training. Smaller companies may lack the resources to justify switching platforms. Anthropic also continues to spend billions on Nvidia computing capacity. Analysts see limited immediate danger to Nvidia, citing its advantages in performance, networking and software.

OpenAI intends to buy computing chips from Google, according to a person familiar with private discussions cited by Bloomberg. Final deal terms remain unknown. The company has also announced plans to begin using its own chips later this year as it seeks additional processing capacity.

For Nvidia, the longer-term concern is pressure on pricing rather than an immediate loss of dominance. Broader access to competing chips could increase customers’ bargaining leverage, while excess computing capacity presents another risk. Citadel CEO Ken Griffin has warned that oversupply could cause GPU rental prices to fall sharply.