One of artificial intelligence’s most enduring selling points is its promise to dramatically boost productivity. In theory, smarter tools should allow people and businesses to do more with less, lifting incomes and accelerating growth.
In practice, Anthropic is warning that who actually benefits from those gains may depend less on ingenuity and more on geography and wealth.
In a recent analysis of how its Claude chatbot is being used worldwide, the AI startup found that richer countries are adopting AI far faster than lower-income nations, with little sign that the gap is narrowing. The findings raise uncomfortable questions about whether AI, rather than leveling the global economic playing field, could end up reinforcing existing inequalities.
Anthropic’s study examined more than one million conversations from individual users on both free and paid versions of Claude, alongside another million interactions from enterprise customers. The pattern was consistent: usage was heavily concentrated in high-income countries. Lower-income nations lagged significantly behind, and Anthropic said there was “no evidence yet that lower-income countries are catching up.”
The reasons are not hard to identify. Advanced AI systems require reliable electricity, fast internet, modern hardware, and, in enterprise settings, deep integration into business processes. All of that costs money. For companies and governments in poorer countries, the upfront investment alone can be prohibitive, before questions of skills, training, and long-term maintenance even come into play.
The concern has also been expressed by others. Microsoft recently published research showing that AI adoption in the “global north” has nearly doubled over the past year compared to the “global south,” while overall usage remains far higher in wealthier economies. Peter McCrory, Anthropic’s head of economics, summed up the risk bluntly, telling the Financial Times that if AI-driven productivity gains materialize, “you could see a divergence in living standards” that favors places already ahead.
That warning cuts to the heart of the AI debate. Productivity gains are not automatic, and even when they occur, they do not guarantee shared prosperity. The experience so far suggests that the relationship between AI adoption and economic benefit is far messier than many technology evangelists suggest.
Evidence from early adopters is mixed at best. A study by MIT last year found that 95% of businesses that had invested in generative AI tools had yet to achieve a net-positive return on that investment. Rather than immediate efficiency gains, many firms are still grappling with integration challenges, unclear use cases, and organizational friction.
Workers’ experiences tell a similar story. According to a survey by Upwork, around half of employees said they do not know how to deliver the productivity improvements their employers expect from AI. More strikingly, more than three-quarters reported that AI tools have actually reduced their productivity and added to their workload, at least for now. Instead of replacing tasks, AI often introduces new layers of oversight, editing, and coordination.
This matters because even if AI eventually does raise productivity, history shows that higher output does not automatically translate into higher wages or broader economic well-being. In the United States, worker productivity has nearly doubled over the past 50 years, driven in part by technological change. Pay, however, has failed to keep pace, while corporate profits and executive compensation have surged. Technology boosted efficiency, but the rewards were unevenly distributed.
Against that backdrop, Anthropic’s warning lands as both an economic and moral question. It is notable that a leading AI company is openly acknowledging that income inequality is real and that its own technology could intensify it. That stance stands in contrast to more utopian claims from parts of the tech world, where some executives argue that AI will soon make everything so cheap and abundant that concerns about inequality will fade away.
The harder question is what follows from that acknowledgment. If the builders of AI systems believe their products risk amplifying global inequality, should market forces alone be allowed to determine who gets access? Or is there a role for policy, international cooperation, and deliberate investment to ensure that productivity gains do not remain locked within wealthy economies?
There is also an uncomfortable tension in the debate. Even as companies like Anthropic warn about inequality, they continue to scale technologies that require vast capital and infrastructure, conditions that inherently favor rich countries and large corporations. That contradiction is not lost on observers, especially in a world where AI founders themselves sit among the global elite.
For now, Anthropic’s analysis points to the fact that AI’s promise is not just a technical challenge but a distributional one. Productivity, on its own, is not a guarantee of shared progress. Without intentional choices about access, skills, and investment, the next wave of technological advancement may end up widening the very gaps it claims to help close.
