Dead Money

Key Points:

• Hyperscalers like Microsoft, Google, Amazon, Meta, and Oracle have heavily invested in AI data centers and GPUs, but much of this hardware remains uninstalled or unused due to power, labor, and infrastructure bottlenecks, creating an illusion of demand that does not reflect real AI compute usage.
• Current AI revenues, heavily concentrated in Anthropic and OpenAI, are far below the levels needed to justify the massive capital expenditures made by hyperscalers; estimates suggest that annual AI revenues would need to reach $2 to $3 trillion by 2030 to break even on these investments.
• AI startups beyond Anthropic and OpenAI are largely unprofitable and dependent on continuous venture capital funding, with many offering commoditized products and limited intellectual property, making their long-term viability and contribution to AI compute demand questionable.
• The AI data center industry faces a "doom loop" where rising debt costs, supply chain inflation, and slow construction increase expenses and financing costs, while uncertain customer revenues threaten the ability to service massive debt loads, raising concerns about widespread defaults.
• Anthropic’s leaked 2025 financials reveal severe losses and poor economics, worse than OpenAI’s, casting doubt on its ability to meet compute commitments and sustain operations; this highlights broader financial risks in the AI sector that investors and stakeholders must urgently address.