Anthropic's planned stock market debut is set to become one of the biggest tests yet of investor confidence in frontier artificial intelligence. But while the Claude maker's proposed valuation reflects expectations of enormous future demand for advanced AI, enterprise buyers face a more immediate question around how much it will cost to put these technologies to work. Will the productivity gains justify the investment?
Anthropic is targeting a valuation of up to $1.4TN, based on its initial public offering (IPO) filing, with Nasdaq trading expected to begin on November 19.
The scale of the proposed listing reflects the extent of the capital flowing into frontier AI development. However, the economics of building these models and the economics of using them inside enterprises are two different considerations.
Ramprakash Ramamoorthy, Director of AI Research at Zoho, told CX Today that enterprise customers need to consider the costs of accessing foundational models alongside the practical returns they can expect.
“Even though they might have changed the way we work personally, but the way businesses [work], it's a very different ball game.”
The Rising Cost of Putting Frontier AI To Work
The capabilities of large language models have developed considerably since the arrival of OpenAI’s ChatGPT-3, moving from generating text and summarising information to connecting with APIs to retrieve data and taking actions across business applications.
These developments are increasing the potential value of AI agents, but they also create greater reliance on the foundational models powering them.
Ramamoorthy points out that enterprises have relatively few choices when accessing the most advanced models. They can pay external providers or consider hosting open-source models themselves, which requires investment in data centers, GPUs, cooling and technical expertise.
Meanwhile, model providers face enormous infrastructure and computing costs. As companies such as Anthropic attract increasingly large investments, enterprise buyers should consider whether providers will seek to recover their spending through future pricing changes, premium services or more restrictive usage arrangements.
There is also the question of how consumption affects overall expenditure. Although advances in model architecture and inference software are helping improve efficiency, lower token prices do not necessarily translate into lower bills.
“Yes, the cost of a token per se has come down, but then you're using it so much that the bigger bill is so expensive,” Ramamoorthy noted.
Agentic workflows can involve multiple model calls, API requests and repeated processing before completing a task, and then integration, monitoring and security requirements add further costs.
Businesses should also consider whether every task requires a premium frontier model or whether smaller, cheaper models can handle routine activities.
Productivity Gains Need to Justify the Investment
The other part of the calculation is how much measurable productivity AI delivers.
The benefits have been particularly in software development, with work that previously required a 10-person team and a six-month development cycle now able to be completed by two or three people in around 15 days, including security, compliance and privacy checks.
However, Ramamoorthy cautioned against assuming similar improvements across other business functions.
In customer support, AI can help resolve straightforward first-line tickets, route enquiries and summarize customer histories but more complicated issues still require human intervention. Healthcare and insurance also involve complex processes and sensitive information that introduce additional compliance considerations.
Businesses in these areas are seeing productivity improvements of around 10–30 percent, rather than the multiples sometimes associated with software development, Ramamoorthy pointed out.
This means CX leaders need to measure improvements across the complete customer service process, including the proportion of interactions that still require human handling, rather than relying on automation figures alone.
Claude's Retention Policies Raise Data Sovereignty Questions
Cost is only one consideration. As businesses connect AI models to customer relationship management (CRM) platforms, help desks and other systems containing sensitive customer information, questions about data residency, ownership and permitted use become increasingly important.
Zoho's 2026 Digital Health Study found that more than 75 percent of U.K. businesses did not want their data stored outside U.K. jurisdiction.
Organizations need to understand where their information is held, how providers process it and whether it can be used to train external models.
“Your data is your secret,” Ramamoorthy said.