AI Demand Is Overstated — Only Anthropic Is Being Realistic

The AI demand signal is currently skewed, with only Anthropic actively addressing the true cost of AI consumption. Many companies are making significant investments without fully understanding the associated risks, driven by the perceived coolness of AI. Every AI interaction incurs token costs, which are the basic unit of AI consumption. While simple chats use a few hundred tokens, new AI agents, capable of browsing the web and completing tasks autonomously, can run for hours, consuming millions of tokens unnoticed. This leads to substantial, unbudgeted expenses, as seen with Uber maxing out its full-year AI budget by April. Companies are overrunning initial inference budgets significantly, with AI costs potentially rivaling engineering headcount. This is exacerbated by "tokenmaxxing," where employees are incentivized for AI usage rather than productivity, leading to inflated consumption. The CEO of Databricks and Eric Glyman of Ramp observe this inefficiency, noting that advanced models are often used for simple tasks where they are not necessary. The "unlimited usage problem" further complicates matters. Flat-rate plans, like ChatGPT Pro's unlimited messaging, are proving unsustainable, especially with AI agents burning through tokens much faster. Anthropic has recognized this by cutting off unlimited subscriptions for popular third-party tools and moving enterprise customers to per-token billing. This shift means that consumers and companies who budgeted for flat rates may reduce their AI usage. This potential pullback in token demand is critical because the entire AI investment cycle, from Nvidia's chip sales to data center construction, is built on the assumption of continuous growth in AI usage. If a significant portion of this usage is driven by gaming leaderboards, inefficient agent loops, or unsustainable budgets, the current infrastructure being built may be over-sized for an unreal demand. There is a clear overinvestment in AI, with companies making billion-dollar bets on demand that has not yet materialized, creating a "cone of uncertainty." Anthropic's move to per-token billing and verifiable demand modeling positions it differently from competitors, emphasizing prudence and long-term strategy in a market that is still sorting out real demand from inflated projections.