AI's Economics Don't Make Sense [Ad Free]
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
GitHub Copilot's shift to usage-based billing on June 1, 2026 is the latest sign that AI subscription economics are fundamentally broken. For years, companies like Microsoft, Anthropic, and OpenAI subsidized compute costs — often losing $20+ per user monthly — to build user habits while hiding true token costs. Now that subsidies are ending, real costs are emerging: Claude Code averages $13/developer/day ($273/month), with heavy users hitting $630/month. The piece argues this was deliberate deception, as flat-rate subscriptions never made economic sense for variable LLM workloads. The broader data center economics are equally dire: a 100MW facility needs 100% utilization just to achieve ~5% gross margin in year one, and Oracle's Stargate Abilene project requires OpenAI to generate $852 billion by 2030 to avoid catastrophic failure. The author concludes that generative AI is fundamentally unprofitable, that subscription pricing was tantamount to fraud, and that the transition to token-based billing will expose the true ROI gap that most enterprises have been ignoring.
Table of contents
The Core Economics of Generative AI Are BrokenMonthly AI Subscriptions Are All Part of AI’s Subsidy Scam, A Deliberate Attempt To Separate Generative AI From Its Actual CostsCan The Average Company Afford To Move To Token-Based Billing? Anthropic Estimates Users Spend $13-$30 a day ($7K+ a year) On Claude Code, As Large Organizations Spend Hundreds of Thousands or Millions A YearThe Economics of AI Data Centers And Compute Do Not Make SenseAI Data Centers Are Expensive To Build, Expensive To Run and Make Very Little Actual RevenueStargate Abilene Is A Disaster — $2.94-per-GPU-per-hour, $10 Billion In Annual Revenue, Years Behind Schedule, One Tenant That Loses Billions of Dollars A YearIf OpenAI Fails To Raise $852 Billion In Revenue, Funding, and Debt Throughout The Next 4 Years, The Stargate Data Center Project Will Kill OracleAnthropic Is Just As Bad As OpenAI, Committing To Up To 10GW ($100BN+ Annual Revenue) In Compute From Google and AmazonThere Needs To Be $156.8 Billion In AI Annual Compute Revenue To Support The 15.2GW of AI Data Centers Under Construction, and $1.18 Trillion To Support All 114GW AnnouncedGenerative AI Is Unprofitable and Unsustainable, And Only Getting More ExpensiveLLMs Are A Ripoff, And Customers Have Been Lied ToSort: