Anthropic’s $750 billion compute costs threaten profitability
Anthropic’s $750 billion computing costs raise concerns about its ability to make a profit before its IPO
These high costs could make it difficult for Anthropic to start making a profit.
Most of this money would pay for the computing power needed to answer AI questions and train AI models, especially those used by its Claude platform.
Every time someone uses Claude, the system processes small pieces of text called “tokens.” Processing these tokens costs money, so frequent use can lead to very high running costs.
Recent estimates suggest that monthly subscription payments do not cover the full cost of processing these tokens. In some cases, serving a customer could cost Anthropic as much as $8,000.
The main planned costs include $160 billion to lease processors from Alphabet, $111 billion for Google Cloud services, and $110 billion for Amazon Web Services.
Other planned spending includes $86 billion on Nvidia computer chips and $31 billion on Microsoft’s Azure cloud services.
Concerns about the future
Most of these spending agreements cannot be canceled. This means Anthropic must pay the agreed amounts even if it does not use all the computing power it has arranged to get.
This could cause problems if fewer people start using Claude or if other AI companies become stronger competitors.
During the first six months of 2026, Anthropic spent more on computing power than it spent during all of 2025. This shows that its costs are rising.
Although the company has an $80 billion loss, the situation may not be as bad as it first seems. However, important risks remain.
As the AI industry changes, investors may pay close attention to how well Anthropic controls its spending.
Whether Anthropic can make a profit in the future will depend largely on how much people use its services and how well it competes with other AI companies.