TechFlow News, June 14: According to ARK Invest’s disclosure on X, Anthropic currently pays SpaceXAI approximately $1.25 billion per month to lease its Colossus supercomputing cluster.
By this metric, Anthropic’s annual spending on compute leasing totals roughly $15 billion. ARK notes that the overall construction cost of the Colossus cluster is approximately $30 billion—meaning, in theory, rental revenue alone could recoup the project’s capital expenditure within just two years.
ARK contends that this case illustrates a new capital-expenditure paradigm emerging in AI infrastructure. Historically, neither cloud computing, mobile internet, nor traditional data center construction has seen such intense and sustained demand for compute resources.
Market analysts observe that as frontier model training costs continue rising, competition among leading AI firms is shifting from model capability to compute-resource access. The entity securing greater access to GPUs, network bandwidth, power supply, and supercomputing clusters stands a stronger chance of gaining an edge in the next round of model iteration.
Meanwhile, this unprecedented scale of capital investment is reshaping value distribution across the AI supply chain. GPU suppliers, optical module manufacturers, data center operators, power infrastructure providers, and cloud service providers have all become key beneficiaries of the current AI investment boom.
If these figures are accurate, Anthropic’s annual compute expenditure now approaches the annual capital expenditures of some large technology companies—further confirming that today’s AI race has entered the “trillion-dollar infrastructure era,” whose investment scale far exceeds that of any prior technological cycle.
