Anthropic says the value of Claude AI used inside the company has grown so rapidly that, at customer list prices, it exceeded monthly personnel expenses in April 2026. The comparison excludes stock-based compensation and reflects the retail value of AI usage—not what Anthropic actually paid to run the systems.
The company disclosed the figures in an amended filing with the US Securities and Exchange Commission as it prepares for a potential public offering. The filing says the list-price value of internal Claude usage increased more than 250 times between January and September 2026.
Claude Writes More Than 90% of Code Contributions
Anthropic reported that Claude wrote more than 90 per cent of contributions to its codebase in September. The average daily volume of code contributed per engineer during the third quarter was also more than 45 times the average recorded in 2024.
The company says employees increasingly use Claude to run experiments, prepare training data, automate evaluations and safety audits, and write or review code. Engineers are spending more time directing and checking AI-generated work rather than writing every line themselves.
AI Is Being Used to Build More AI
Anthropic describes a development cycle in which Claude helps create and improve later versions of the model. The company believes this feedback loop, combined with AI use across research, engineering, sales and operations, can help it grow output without increasing its workforce at the same pace.
The filing also points to a widening gap between revenue and headcount growth. Anthropic’s workforce expanded more than fivefold between the end of 2024 and September 2026, while quarterly revenue in July–September 2026 was reported at 86 times the level in the final quarter of 2024.
The Figures Need Context
The internal-use figure is calculated using the prices Anthropic charges customers. It should not be read as a direct measure of cash savings or as proof that AI has replaced an equivalent amount of paid labour. The filing does not provide a corresponding measure of code quality, defects or the amount of human review required.
Anthropic also acknowledged that using AI to help develop more capable AI systems could create safety risks, particularly if progress accelerates. It said its safety policies are intended to monitor and mitigate those risks.
The disclosure offers a striking example of how heavily an AI company relies on its own models. Whether the same productivity gains can be replicated by other businesses will depend on their workflows, costs and ability to evaluate AI-generated work.
