Premium: The AI Data Center Financial Crisis

Source: Where’s Your Ed At Date: 2026-02-13 URL: https://www.wheresyoured.at/data-center-crisis/

Summary

Premium deep-dive arguing that the AI industry faces a structural financial crisis: every major generative AI company is unprofitable with no credible path to sustainability, funded by a chain of venture capital that cannot sustain itself indefinitely. Zitron exposes the training-cost accounting trick — AI firms exclude training costs from gross margin calculations, making financials look healthier than reality. Anthropic’s true 2025 gross margin was negative 53% when training is included. Both Anthropic and OpenAI project “magical” transitions to profitability (2028 and 2030 respectively) without credible mechanisms, while data center operators burn millions daily on idle infrastructure.

Implications

• The accounting deception is load-bearing. The entire AI investment thesis depends on investors not asking how training costs are classified. Once that question becomes standard due diligence, the reported unit economics collapse.
• Cascade risk. Every segment — GPU manufacturers, data center operators, model providers, downstream startups — operates at losses sustained by the same upstream funding. A stumble at one node (failure to hit revenue targets, VC pullback) propagates through the whole stack.
• Oracle’s exposure is the canary. $4.69M/day burned per 100MW of idle infrastructure at Oracle means Stargate’s economics depend entirely on OpenAI hitting revenue projections that have never been met. This is the most concrete near-term stress point in the system.
• Watch: Q1 2026 earnings from Oracle, Microsoft Azure AI revenue segment, and any revisions to Anthropic/OpenAI revenue guidance.