Anthropic’s IPO filing shows rapid growth in enterprise AI spending. It also raises a question for the companies buying its models: Are those investments paying off?
Anthropic reported $16.3 billion in revenue for the first half of 2026. As of Sept. 30, it had more than 11,900 customers that each generated at least $100,000 in revenue over the previous 12 months. That was up from more than 950 a year earlier, according to the SEC filing. Those figures show the scale of Anthropic’s commercial growth, but they do not show whether customers are earning a return.
Broader use brings more costs
AI agents can now write code, analyze information and automate multistep processes, expanding their use beyond experiments. Jeet Pattanaik, SAP API program manager and founder and CTO of enterprise AI vendor Glokal AI, said better models and easier purchasing are helping adoption. “The simple reason is that the models are now good enough to do real work,” he said. Existing cloud contracts can also help companies move beyond pilot projects.
More use does not necessarily mean better results. Employees may use AI regularly without completing work faster or more effectively. Kathryn Harrison, global vice president, global product organization at technology and services company Concentrix, said spending reflects both confidence in AI and pressure from boards and CEOs to move faster. “It shouldn't be read as proof that companies have figured out the economics,” she said.
Deployment costs extend beyond model prices and subscriptions. They include integration, human review, maintenance and employee training. Tim Doll, chief digital officer of Generational Group and CEO of tech consultancy Precocity, said companies often underestimate the cost of scaling, especially when they use expensive frontier models for routine work. Poorly prepared data and disconnected systems add difficulty. Harrison said changing technology can also force companies to revisit or rebuild workflows.
Lower prices may increase total spending
Anthropic says wider use of Claude across enterprise workflows and products drives its revenue growth. Its filing also describes how lower costs per task can encourage more use and broader adoption. For customers, that increased use still has to deliver enough value to justify its cost.
Lower-priced models may make some workloads cheaper to run, but total spending can rise with usage. “Cheaper models are good news, but they rarely lower the total bill,” Pattanaik said. “When something gets cheaper, people use it more. So, each task costs less, but overall spending goes up.”
Read nextSK hynix reports record third-quarter revenue and ends Solidigm IPO reviewSwitching providers can also become harder as companies build applications, coding workflows and agents around one model. Pattanaik said buyers should ask whether they could move workloads to another provider within three months. Keeping prompts, workflows and testing systems in-house can make that easier.
AI providers generally charge for usage rather than customer results, Doll said. That leaves buyers to define success and measure it. Harrison said companies should compare an AI task’s full cost with the previous process, including time spent reviewing and correcting output, then check whether the entire workflow improves service costs, completion time, customer experience or revenue.
