Anthropic Files for IPO Near $1 Trillion as AI Cost Concerns Mount

Anthropic filed paperwork this week to go public in the United States, a step that would place one of the leading developers of advanced AI systems on the stock market at a valuation approaching 1 trillion dollars. The move comes as companies that have poured resources into AI tools begin to question whether the returns justify the costs, and as broader public skepticism toward the technology surfaces in universities, workplaces and even the Vatican.

Anthropic, founded five years ago by former OpenAI executive Dario Amodei, recently raised private capital at a valuation above 965 billion dollars. OpenAI, its main rival, has been valued at 852 billion dollars and is also reported to be considering an initial public offering. These figures reflect investor enthusiasm for AI systems that can generate text, code and other outputs at scale. Yet they coincide with signs that some corporate customers are recalibrating their spending.

A Bain survey of nearly 1,000 companies found that many have not seen meaningful cost savings after investing in AI, with 40 percent reporting gains below 10 percent. Sam Altman, OpenAI’s chief executive, recently described corporate concerns about AI expenses as the most legitimate criticism the industry faces so far. Reports of unexpectedly large bills, including one case in which a client incurred hundreds of millions of dollars in a single month on Anthropic’s Claude model, have prompted some firms to explore cheaper alternatives, including open-source models.

This corporate hesitation arrives alongside more visible forms of pushback. Pope Leo XIV issued an encyclical last month calling for regulation of AI development, warning that the technology risks new forms of dehumanization. The document follows earlier statements from religious leaders expressing caution about systems that could displace workers or concentrate power in unaccountable companies.

At several universities this spring, graduates booed commencement speakers who offered optimistic forecasts about AI’s economic effects. Inside companies, some employees have begun quietly resisting internal mandates to integrate AI tools into their workflows. Proposals to build large data centers, which require substantial electricity and water, have also encountered local opposition, prompting some technology firms to explore less conventional locations, including proposals for orbital facilities.

The infrastructure build-out continues despite these signals. Nvidia chief executive Jensen Huang recently predicted that Marvell Technology, a semiconductor designer whose chips support data-center connectivity, could become the next trillion-dollar company. Alphabet, Google’s parent, separately announced plans to raise 80 billion dollars, much of it earmarked for AI-related spending.

Anthropic’s business customers have been a source of strength, with the company recently surpassing OpenAI in the number of enterprise accounts. Those clients typically pay higher rates than individual users. If spending slows, however, revenue growth could prove harder to sustain just as the company seeks public-market investors.

The pattern echoes earlier episodes in which technological change produced concentrated gains for investors and uneven outcomes for workers. Past waves of automation in manufacturing displaced large numbers of employees with limited retraining or income support. Current AI systems target a wider range of cognitive tasks, raising similar questions about who will bear the adjustment costs and whether new forms of regulation or public investment will accompany deployment.

Policy responses remain limited. While some lawmakers have proposed modest transparency requirements for AI models, comprehensive rules on labor impacts, energy use or market concentration have not advanced. The gap between the speed of private investment and the pace of democratic oversight continues to widen, leaving many institutions to react after systems are already embedded in workplaces and daily life.