AI Inference Costs Hit $47M Daily as Industry Faces Pricing Crisis

The Math Doesn't Add Up

ChatGPT costs approximately $47 million per day to operate as of early 2026, according to analysis reported by Jacobin. With 800 to 900 million weekly active users but only 35 million paying subscribers, OpenAI faces annual operating costs around $17 billion for inference alone—the computational work of generating each response.

This represents a fundamental economic problem that distinguishes generative AI from previous technology platforms. Unlike social networks that scaled efficiently, each AI query consumes significant computational resources including electricity, water for cooling, and expensive hardware. The company expects to spend over $150 billion on inference costs through 2030.

Research firm SemiAnalysis estimated ChatGPT was already costing $700,000 daily in 2023. As models grow more sophisticated, they become more expensive to run—the opposite trajectory of most digital services.

Manufacturing Demand That Isn't There

The AI sector has attempted to solve this gap between costs and revenue through what Jacobin describes as "fictitious demand creation." Major tech companies have mandated AI usage among employees regardless of genuine need.

Meta and Shopify created internal leaderboards tracking token consumption. Nvidia CEO Jensen Huang stated he'd be "deeply alarmed" if engineers weren't using $250,000 worth of tokens annually—this after Nvidia invested $30 billion in OpenAI. Accenture told senior staff in January that regular AI tool usage would be required for promotion consideration.

Uber's experience illustrates the problem. After spending $3.4 billion on AI in 2025 and pressuring its 5,000 engineers to maximize Claude usage, the company exhausted its entire 2026 AI budget by April. Per-developer consumption increased five- to twentyfold, but no public benchmarks show matching increases in output value.

The Shift Toward Enterprise Stratification

Anthropicâ€TMs release of Claude Opus 4.5 in late 2025 marked a strategic pivot. The agentic AI model—capable of autonomous web browsing, code execution, and complex workflows—is dramatically more expensive to run than conversational chatbots. A single agentic task can consume millions of tokens compared to hundreds for basic chat.

Since early 2026, Anthropic has introduced token-based surcharges, premium inference tiers, and separate billing for autonomous agents. In May, the company announced that Claude subscribers would face separate monthly credit meters for agent tools billed at full API rates starting mid-June.

An anonymous large language model researcher told Jacobin that model providers are "increasingly prioritizing agentic capacities and B2B deals, while deprioritizing end consumers." The most powerful AI capabilities are being rationed to white-collar workers at large companies that can afford premium pricing.

Why it matters

The AI industry's cost structure is forcing a transition from universal access to tiered pricing that favors enterprise customers—a shift happening without public debate about who benefits from productivity gains versus who bears environmental and economic costs. Unlike previous platform monopolies that could scale efficiently, generative AI's per-query costs create inherent barriers to democratic access as models grow more capable.

These details were first reported by Jacobin in an analysis by Sophie Bandarkar, a writer and PhD student in economics based in Paris.