Anthropic's new Fable 5 model is driving infrastructure spending across major cloud and chip suppliers—Nvidia, Amazon, Google, and SpaceX—each of which provides computing power or hardware that the model relies on.

What happened

Anthropic unveiled Fable 5, its most powerful language model to date, designed for autonomous multi-step work rather than single responses. The company is using hardware and cloud infrastructure from Nvidia (Blackwell and Vera Rubin GPU architectures), Google Cloud (custom Tensor Processing Units for training), Amazon Web Services (tied to a $100 billion（約16兆円） commitment over the coming decade), and SpaceX (300 megawatts of capacity with over 220,000 Nvidia GPUs, at $1.25 billion（約2000億円） per month for three years).

Why it matters

Large language models require massive parallel processing power during training and deployment. As Fable handles more demanding workloads—software engineering, application building, chart and diagram interpretation—the infrastructure partners will see higher volumes of chip procurement and greater use of their cloud platforms. For these suppliers, Fable's release translates into sustained hardware and cloud capacity demand.

What to watch

SpaceX's new arrangement grants Anthropic 300 megawatts of capacity with a cluster of more than 220,000 Nvidia GPUs at $1.25 billion（約2000億円） per month over the next three years. This positions SpaceX as an emerging hyperscale compute provider, converting data center infrastructure originally built for its own needs into a steady revenue stream.