Anthropic announced it will use 1 million of Amazon’s custom AI chips [1] following public criticism of Nvidia’s hardware strategy.
The move signals a potential shift in the AI infrastructure landscape, as one of the world's leading AI startups attempts to reduce its dependency on the dominant chip maker.
In a blog post published last week, Anthropic CEO Dario Amodei criticized Nvidia's approach to the market. Amodei said Nvidia’s pricing and closed-model strategy limit competition and make AI development less affordable for developers.
This public break comes as the industry struggles with the high costs of computing power required to train and run large language models. By pivoting to Amazon's custom silicon, Anthropic aims to bypass the constraints imposed by Nvidia's market position.
The agreement involves the deployment of 1 million custom chips [1] provided by Amazon. This scale of hardware adoption suggests a deep integration between the AI startup and the cloud giant's infrastructure.
Nvidia has long held a near-monopoly on the high-end GPUs used for AI training. The criticism from Anthropic highlights growing tension between the chip maker and the companies that rely on its hardware to build generative AI tools.
While many AI firms continue to rely on Nvidia's H100 and Blackwell architectures, the shift toward custom-built chips, such as those from Amazon, represents a strategic effort to lower operational costs and increase hardware efficiency.
“Anthropic will use 1 million of Amazon’s custom AI chips”
This shift indicates a growing desire among AI labs to vertically integrate or partner with cloud providers to escape Nvidia's pricing power. If other major labs follow Anthropic's lead, Nvidia may face increased pressure to open its ecosystem or lower costs to maintain its dominance in the AI hardware market.
