AI's Economics Don't Make Sense

2026-04-28 · Source: Ed Zitron's Where's Your Ed At · Field: Finance & Economics — Capital Markets & Investment Management, Economic Analysis & Policy · Depth: Intermediate, extended

Summary

A recent analysis argues that generative AI's economics are fundamentally broken and unsustainable. Current subscription models for services like GitHub Copilot, Anthropic's Claude, and OpenAI's ChatGPT are heavily subsidized, causing significant losses for providers. GitHub Copilot's shift to usage-based billing on June 1, 2026, exemplifies this, as Microsoft previously lost \$20-\$80 per user monthly on a \$10 subscription. AI data centers are prohibitively expensive to build (e.g., a 100MW facility costs \$4.4 billion) and operate, yielding low revenue and gross margins, often becoming unprofitable due to debt and rapid depreciation. Major projects, such as Oracle's Stargate Abilene for OpenAI, are years behind schedule and depend on OpenAI's "ridiculous" projections of \$852 billion in revenue/funding by 2030, a target even OpenAI's CFO doubts. Anthropic similarly commits to 10GW of compute from Google and Amazon, representing over \$100 billion in annual revenue commitments, despite only \$5 billion in lifetime revenue. The industry is deemed unprofitable, unsustainable, and built on a "subsidy scam" that intentionally hides true costs.

Key takeaway

For investors and executives evaluating generative AI investments, recognize that current business models are largely unsustainable "subsidy scams." You should scrutinize AI service providers' long-term financial viability and demand transparent, token-based billing models. Be prepared for significant price increases and reduced service quality as companies shift away from subsidized offerings. Your due diligence must account for the true, often hidden, compute costs and the lack of proven ROI.

Key insights

Generative AI's subscription models and compute infrastructure are economically unsustainable due to massive subsidies and hidden true costs.

Principles

• LLM services are not economically feasible without token-based billing.
• AI data centers require 100% utilization for minimal profitability.
• Yearly GPU upgrade cycles make data centers near-obsolete quickly.

Method

AI companies intentionally obfuscate token burn and actual costs through monthly subscriptions and vague usage limits to grow user bases, creating an unsustainable economic model.

In practice

• Evaluate AI ROI against actual token burn, not subsidized rates.
• Question AI service providers on their long-term cost models.
• Be wary of "jagged intelligence" when paying per token.

Topics

• Generative AI Economics
• AI Compute Infrastructure
• Subscription Billing Models
• OpenAI Financials
• Anthropic Compute Deals
• GitHub Copilot Pricing

Best for: CTO, VP of Engineering/Data, Director of AI/ML, Investor, Executive, Tech Journalist

Related on AIssential

• AI's Economics Don't Make Sense [Ad Free] — Generative AI's subscription model is economically unsustainable, masking massive compute costs and leading to inevitable price increases.
• Premium: What If...We're In An AI Bubble? (Part 3) — The AI industry's financial models, particularly for compute demand and revenue growth, are built on highly speculative and potentially unsustainable projections.
• Do AI models actually make enough money to cover their costs? Live with Epoch AI — The economic sustainability of high-valuation AI companies hinges on balancing immense training costs with revenue generation.
• 📈 Data to start your week — Global market data reveals shifts in solar energy, US economic growth, developer resource usage, and AI hardware/valuation dynamics.
• The bubble is slowly popping, investment isn't able to keep up — The AI market faces a critical debate regarding its investment sustainability and the long-term viability of current pricing models.
• The LLM Gamble — LLM usage costs are unpredictable, akin to a slot machine, challenging sustainable business models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Ed Zitron's Where's Your Ed At.